Medium Format Digital Camera Optical Precision
MEDIUM FORMAT DIGITAL CAMERA OPTICAL PRECISION
or
PROBLEMS IN THE LAND OF PRECISION
Mainly written in early March, 2008
Completed July 8, 2008
Updated and Published April 5, 2009
Incorrect Image Link Fixed April 15 HERE
Thanks are in order for Elmo Sapwater of ImagingInsider.com and Michael Reichmann of Luminous-Landscape.com
for providing the links to ensure that this article and the one which followed it a few days later would come to your attention.
For that newer article go HERE.
Supplementary article with comments on this one from other photographers posted here, April 20th, 2009.
PREFACE
July 30, 2010
Despite the tale of woe for quality control which you will encounter in this long article, I want to make it perfectly clear that I find medium format digital capture systems to offer tremendous image quality and I rely primarily on my medium format system for my own work and am quite fond of it. I also use a full-frame 35 system, which fills a variety of different roles, and which also can be used for serious high res work, especially when using stitching. I find it most useful to have both systems, as between them, they cover a very broad range of capabilities. I only bemoan the outsized "quality tax" on my main system's purchase. Nor does this mean that the 35mm systems are without QC problems. My new rule of thumb: never assume a camera's focussing screen matches its sensor with respect to the focus adjustment (do a test right away), and especially never assume that a lens performs as it should, rather assume lots of variation from sample to sample of a given type, even with the products of the very best lens makers. Test every new lens right away. The corners will usually tell the story.
INTRODUCTION
April 5, 2009
[Before I get started, I need to explain that, as you can see, I wrote this a year ago and am only now deciding to go ahead and publish it. I have just gone through it and made a few notes in places where things I have learned subsequently required it.]
I want to share the things I have been learning about high-end digital capture over the last year or two. I've been following this general topic closely since the mid-nineties and recently elected to make the switch from 4x5 color films to digital capture.
I first did what I could to evaluate all the most relevant properties of the various approaches, including the use of: scanning backs in view cameras, the new Seitz scanning back on an ALPA, the same on an ARCA R m3d, the new Seitz D3 pano camera, all four of the major brands of medium format backs at 33 or 39 million pixels, every medium format view camera, every 4x5 view camera, the Canon systems, all of the medium format SLRs, the other metal-plate type cameras (aside from the ALPA and the ARCA R), and the use of both single-frame shooting and multi-frame stitching methods with each of the above. The issue of tilt focus capability is a large one for me, even considering the very useful focus blending techniques that are available. The issues of optical defects in top brand (German) wide angle lenses and focus precision in digital back sensor calibration have unfortunately turned out to be huge. Those two things are what this article is largely about.
It's also true that the design properties of the cameras which can use the various medium-format and scanning back sensor options have, generally speaking, not fully caught up with the reality of the sensors. Most of the existing cameras were designed for either 645 film or 56 x 84 mm (6x9) film. The optical demands of 37 x 49 (or 36 x 48) mm sensors are vastly higher (four to six times higher in my estimation). So the precision required in the camera system is a huge issue and this includes not merely manufacturing precision, but also the things which allow the photographer to get a picture into focus in the first place, including the view through a viewfinder, the fineness of the focussing mechanism, the fineness of a tilt mechanism, where present, etc. The sheer size of the cameras also tends to be overly large, given the smaller size of the sensors. I'm not sure that anyone has yet to actually design a camera around these small sensor sizes, though some seem to have been designed for full-frame 645, i.e. 42 x 56 mm. To complicate matters, the camera and lens makers have to foresee that we may get larger sensors, including possibly square ones, before too long. So they need to be wary of investing too heavily in this format.
The first thing view camera users usually want to know is: Will it match the sheer detail that I can get from my 4x5 sheet film? The answer is an inordinately complicated one. Sometimes, yes, and then some. Sometimes no. [If you ask, can it, then the answer is most certainly yes, as we shall see at the end.]
Of course there are many important qualities of photographs besides the detail, but one must work to one's personal standards in every way, so for many of us -- landscape photographers in particular -- the sheer detail is a critical factor in deciding to switch, and it's perhaps the last area where sensors have reached the level that people like myself prefer to work at.
One aspect of digital capture with area array sensors (as opposed to linear array scanning sensors) is that it lends itself to relatively easy and high quality stitching of multiple frames into a single finished result. This can provide greatly increased detail, but also provides flexible aspect ratios without relying on cropping and its attendant loss of detail. Indeed it's possible for low-end digital cameras to produce extremely high resolution images under some circumstance by stitching.
Another important aspect of the high-end systems is that they tend to have total dynamic ranges approaching roughly 12 stops (though my latest test showed more like a useful range of 7 to 10 stops, with the bottom few zones not useful for heavy lifting -- i.e. keep them small and in the deep shadows). This is nevertheless a big advantage over transparency films in particular, and allows the CCD-based digital backs to rival most or all kinds of common B&W and color negative films for lattitude.
And the ability of digital capture to totally eliminate color crossovers is another profound advantage. A color digital capture can render gray tonalities with such incredible precision as to make pictures that look indistinguishable from black and white -- essentially an impossibility with chemical capture (color film).
Digital systems are faster to use than view cameras (except for the scanning back systems, which have, until the recent advent of the Seitz scanning system, all been slower). Digital systems can capture many times more frames without stopping to reload film or copy files. Digital systems are expensive, but they eliminate film, processing, and scanning costs. And they increase the quantity of data to be archived. But that is getting steadily easier as things like hard drives and optical disks continue to get bigger in capacity, faster, and cheaper. Fairly soon we may see CD-sized holographic optical disks that can store up to between 1 and 3 TeraBytes each.
There are many differences between systems. Non-retrofocus lenses impart color casts to captures due to varying angle of incidence on the sensor, more or less, depending on the sensor type, and this needs to be dealt with in post processing. Moire can be fairly fierce with some subject matter. Purple fringing in fine highlight detail can be a problem. The overall complexity can be rather overwhelming, especially considering all the software options.
So first I want to get on to some important optical basics.
OPTICAL ISSUES
1) As a lens's focal length is doubled, its depth of field (DOF) becomes 1/4 as great. A 50 mm lens has four times the DOF that a 100 mm lens has. The 50 mm lens has sixteen times more DOF than a 200 mm lens has. And so on. In this context I am describing the amount of depth of field (for a given circle of confusion size) as, for example, a relative distance around the outside of a focusing helical mount. So if the 100 mm lens were focussed at 20 feet, and were considered sharp from 10 feet to infinity, then the 50 mm lens could be set to five feet, and be in focus from 2.5 feet to infinity, at the same aperture (e.g. f/16).
2) As a lens's focal length is doubled, the distance which it must be moved to shift the focus from infinity to a given distance, say 10 feet, quadruples. So if a 100 mm lens is shifted slightly from its ideal infinity focus location to focus at 1600 feet away (very close to infinity), that same amount of physical shift with a 25 mm lens will cause it to be in focus at just 100 feet.
This latter point is particularly important as one tries to understand the toughest optical tests for a camera system, especially one for making pictures of distant scenes.
3) Diffraction causes all images captured with lenses to degrade on the plane of focus as the aperture becomes smaller. Apparently the physicists would have us believe that some of the light is bent by passing near the edge of the iris. If you ask me, one must keep a healthy skepticism about the craziness of the world of the incredibly tiny things that make up everything we know :-). In other words, if you have a plane in perfect focus, wide open, as you stop down, many things happen to the image, including depth of field overcoming various aberrations which may be present in varying degree, but including also a general blurring of the image which becomes quite extreme at very small apertures such as f/45, f/64 and so on. So when somebody says that a lens is a diffraction-limited optic, they are making the strongest brag possible about the lens, saying that its only (or at least its dominant) blurriness on the plane of focus is caused by the unavoidable effects of diffraction. Indeed the MTF (modulation transfer function) curves for some of the German digital lenses, particularly the Rodenstock HRs, show performance which is remarkably close to being diffraction limited. As we shall see, it's wise not to assume that this is the performance you will get when you buy one!
When using film, the blurriness added by the film itself obscures diffraction to the point where it's often not particularly obvious even at f/32, but when using one of these digital backs, which unlike the Canon sensors (for example) have no anti-aliasing filter over the sensor, diffraction can be seen almost immediately. Certainly it's obvious at f/11, though it doesn't tend to become ruinous until you go above f/22 -- if you use a good de-convolution sharpening tool to sharpen things up later, a shot made at f/22 can almost perfectly match a shot made at f/8, except that the DOF is much better. Nevertheless, M.F. digital systems naturally demand working at larger apertures than a 4x5 camera does, and this increases the demands for precision, even beyond the huge increase in precision caused by the 3X to 4X increase in sharpness of the sensor itself. Imagine an 8 x 10 or a 5 x 7 camera being shrunk to the size of a grapefruit, but retaining all of its precision. You will learn to start thinking about microns.
4) When shrinking an optical system from the 150 mm diagonal of 4 x 5 film to the 60 and 61 mm diagonals of the 33 and 39 MP sensors (or the new 50), one can expect the DOF from a given equivalent focal length (e.g. 60 mm vs. 150 mm) to be 6.25 X greater (2.5 x 2.5 = 6.25). However, if the sensor can see detail that's 3X to 4X finer than the film, it stands to reason that an acceptable circle of confusion size for the sensor will be about that amount smaller too. So the 6.25X DOF advantage erodes to perhaps only a 2X advantage over the 4x5 system. Then, because we want to avoid the small apertures (certainly f/32 and smaller, if not f/22 and smaller) entirely if at all possible, we may wind up with not a whole lot more DOF than we had with the 4 x 5, while also having more speed from shooting at a larger aperture and higher quantum efficiency in the sensor. I think in all fairness that the MF systems have a good deal more effective DOF, even considering how much detail we're trying to get out of them, but considering all the factors is not so simple.
5) Focussing a 4x5, though complicated, is something that I have almost never had trouble doing with great precision. Nearly every exposure I have made over the 35 years that I used a 4x5 was made with some use of the back tilt. At first without a loupe for the groundglass, but more recently with a 6X loupe, I could readily see where the plane of focus was and get it into the right place, even when shooting at a wide aperture, e.g. at f/8, which was, however, rare. M. F. digital backs require far more precision though. A groundglass used on one of the M.F. cameras is generally thought to demand a 10X loupe for marginally adequate focus. Even the large, beautiful viewfinder image of the Hasselblad H system simply does not provide the magnification necessary to focus manually, except under relatively easy circumstances (fast lens of medium to long focal length, with high-DOF subject matter and a decent light level). Fortunately Mamiya sells a 2X accessory eyepiece magnifier which at least makes it just barely possible to see well enough in the central region of the frame only, for manual focussing. I figure this view is about equal to using an 8X loupe on a groundglass with the same sensor attached to one of the M.F. view cameras. Whichever approach to focussing is taken with these sensors, the precision required is far beyond that of the better view cameras when using sheet film. The required solution may include: more viewfinder magnification, a stronger loupe, a finer helical focussing mount, or a more accurate auto focus mechanism, as the case may be. Or one might use a through-the-sensor mechanism, such as focussing with a live preview with instant zooming to 100% and then back to full screen (imagine focussing on an iPod Touch with a wireless link to the camera, or even better, on a 3" LCD right on the back). Any tilt mechanism should be much finer than for the view camera as well, so that minute adjustments can be made with confidence, and the zero position must be incredibly close to a perfect no-tilt condition.
I have high hopes for the gear-driven tilt in the eagerly awaited ARCA-SWISS R m3d, though it may not be precise enough for making very subtle tilt adjustments. An ALPA tilt adapter prototype is showing promise for extremely fine adjustment, but it will only work with lenses of 80 mm and longer -- and apparently only Schneider lenses at that. The 6x9 view cameras with the greatest precision all seem to be too bulky and heavy for landscape photography. The ARCA-SWISS F Classic 6x9 with the Orbix option may be the best among those monorail cameras reasonably suited for outdoor work where one want to avoid high weight. The Linhof M679CS is perhaps a little more precise, but is also a lot heavier and also bulkier. Besides, I think that for a sensor of these small sizes (e.g. 37 x 49 mm), it's best to avoid swing capability entirely, but to have tilt, plus possibly a rising front and perhaps other shifting movements as well. At least for the kind of images that I'm after.
TESTING AND RESULTS
To test a system -- camera body + digital back + digital back adapter (if any) + lens + focussing system -- for its optical capabilities, I think it's best to go straight to the acid test -- making pictures of all-infinity subjects with one's widest lens at its biggest or second biggest aperture. This test is specifically best for detecting errors in the sensor plane location or focus calibration errors in other components, such as the location of an infinity stop on a lens's helical mount. Using a scene which approaches being infinity for the entire picture eliminates alignment as a source of focussing error better than do close-up tests, unless one has some kind of precision-calibrated copy camera arrangement to guarantee parallelism between the sensor and the plane of a flat subject. Even so, optimal performance at infinity is more important to landscape work than is optimal performance at distances of only a few feet. This is because, in my experience, the most likely single optical failure for a view camera type lens is curvature of field, and this tends to be optimized for one distance at the factory (if you're lucky). Careful spacing with shims behind the shutter, adjusting the gap between the front and rear elements of the lens, appears to be the main method of adjusting curvature of field. What we want is perfectly flat focus at infinity. If we were doing copy work or enlarging, we would want perfectly flat focus at a low reproduction ratio (closer distance). In landscape photography, closer-up subject matter is not often as optically flat as any distant scene is. Please understand that a distant scene which may vary greatly in its relative distance is usually nevertheless all, in effect, at the same distance optically.
Dalsa 33 MP and Kodak 39 MP chips are so sharp (sensors do use pixels the size of bacteria) that they reveal the most subtle flaws in any imaging system, or at any rate, far more subtle flaws than color film could reveal. As I mentioned above, the sensors appear to be between three and four times sharper than modern color film, per inch of sensor or film. The Kodak 39 MP sensor, the KAF-39000 (http://www.kodak.com/global/en/business/ISS/Products/Fullframe/KAF-39000/specs.jhtml?pq-path=12249/12161), which is used in the Phase One P45 and P45+, as well as the Hasselblad CF-39, CF-39MS and HF-39 backs, has a 6.8 micron pixel size. The 33 MP Dalsa sensor has 7.2 micron pixels. By comparison, a Canon 5D [Mark I] has an 8 micron pixel size, and a BetterLight Super 6K back has a 12 micron pixel size. Some of the point and shoot cameras are down to 2 microns or even smaller, but of course their chips are tiny, so their lenses are also tiny. The self-contained cameras, like the point-and-shoot models and the Canon SLRs, tend to avoid the focus issues that I've been seeing with the M.F. systems, though they have other kinds of problems, such as lenses that fall off rather dramatically in sharpness in the corners, and sensors that are softer due to anti-aliasing filters, inadequate dynamic range, etc.
Some of the superwide lenses that are available, particularly the 35 HR lens from Linos/Rodenstock (http://www.linos.com/pages/no_cache/home/shop-optik/rodenstock-foto-objektive/digitale-fachfotografie/?sid=13381&cHash=6a89402634#sid13381 and http://www.linos.com/pages/mediabase/original/rodenstock_apo-sironar_digital_hr_e_2473.pdf) have stunningly good MTF specifications compared with all such lenses built for use with large format film. When lens makers create MTF curves, which show the ability of a lens to reveal subject detail, they use computer simulations -- not measurements of actual production samples. Naturally one must wonder how closely the shipped lenses match the published MTF data.
During the many years I was using fine view camera lenses, I never encountered a lens which it seemed to me was failing to more or less match its published MTF specifications. This was a somewhat superficial as well as an un-quantitative group of observations, but the results with the modern, computer-designed lenses that I've used since the early 1980's were nevertheless always superb. Not once can I recall having had trouble getting an entire horizon into essentially perfect focus, for example. And never was I able to observe any curvature of field at working apertures either, where the plane of focus isn't a plane, rather a paraboloid, like you might see if you were standing inside the nose of a modern submarine, looking forward to see the shape of the hull, or a similar shape. With curvature of field, you may be in focus at infinity on axis (in the middle of the picture), but the corners of the image may be in focus on a plane which crosses the optical axis much closer to the camera. I was able to see some modest curvature of field with some large format lenses wide open on the groundglass, and I avoided purchasing those lenses as best I could.
Many experiments which I have carried out recently with the help of various friends on a substantial number of Schneider and Linos/Rodenstock lenses using a variety of high end backs have shown what seems to be a markedly different picture from that of the large format view camera optical situation. A high percentage of the lenses shipped by these two most-respected vendors appear to be somewhere between mildly and grossly defective. I will provide many JPEG samples at 100% for your examination.
Of course, I can only make observations based on the actual samples I have seen, but I know that even a sample as small as eleven lenses, when they have been selected at random, not on account of a known defect, can give a quite statistically valid general view of the kind of lenses being examined (German, view-camera type, super-wides, of four models only).
First Success:
But before we look at failures, let's take a good, careful look at some successes. These pictures prove how good medium format digital can be with respect to detail in the situations which are the most difficult and demanding with respect to focus calibration issues and lens performance issues -- super wide lenses shot at very large apertures. The testing has been done mainly at between f/4 and f/11. Again, keep in mind that the difficulty here has to do with the depth of focus being tiny and the fact that we are not focussing on a ground glass with a bellows camera -- we are in effect predicting the plane of focus location through what can be somewhat indirect means (but then, even a groundglass is indirect, as its surface may not match the location of the sensor surface). Only focussing through the chip electronically can avoid the potential pitfalls of focus calibration entirely. Depth of focus is like depth of field, only inside the camera, and for very short lenses, it is very tiny -- the opposite of the depth of field with very short lenses, which is huge.
This picture, courtesy of a serious amateur photographer in Taiwan by the name of Jia-Yan,
http://www.callme.com/lens_tests/24_A-D_f5-6_Inf_Exc_Leaf33.jpg (12.1 MB) ** NOTE: Until Wed. April 15th at 19:45 GMT, this link was to the wrong image! I made a clerical error in adding the link. The old file was one of the bad 35 Apo-Digitar samples which also appears lower down in the article. Now the link is to the good 24 Apo-Digitar, wide open at 5.6, as intended. My apologies for the confusion no doubt engendered in anyone looking closely, as I hope you all will! Thanks to Henning Wulff for noticing and letting me know.
was made with a 24 mm Schneider Apo-Digitar lens on an ALPA, focussed at infinity, and shot wide open -- at f/5.6 -- on a Leaf Aptus 75, thirty-three million pixel digital back. Jia-Yan is quite certain that it was shot wide open as a test of the lens (captures on cameras like this do not contain aperture data). The sensor size is about 36 x 48 mm, with a diagonal of about 60 mm (image circle radius of 30 mm). The pixel size is 7.2 microns. Please drag the completed JPEG from your browser window into Photoshop for examination at 100% magnification.
Unfortunately, it was sharpened in the Leaf raw conversion software, so the true detail is somewhat obscured (had it been unsharpened, better sharpening techniques could have made it look much, much better at 100% magnification). Nevertheless we can still see, looking at this picture, that the optical performance of the lens is really superb -- particularly for a super-wide, and a very wide one at that. Corner to corner, it's very sharp. Now, it's true that this sample's subject isn't really far enough away to give us the ideal test, i.e. infinity focus everywhere. But the focal length is so very short, at 24.9 mm (plus or minus the manufacturing tolerance for actual focal length of one percent), that this subject is a reasonable approximation of an all-infinity subject. A lens that can do this, wide open (picture an 18 mm lens on a 35mm camera or a 62 mm lens on a 4 x 5), is really quite exceptional, even when using film, let alone a sensor. But at the same time, since we have to shoot at big apertures sometimes to get the most out of these systems with their proportionally much more diffraction and more depth of field, we really need optics which are capable of the small miracle of performing well, wide open or nearly so, on a sensor with these smallish pixels (7.2 to 6.8 microns).
Just for further reference, a 22 MP back has pixels of 9 microns, and indeed they do seem to have a somewhat easier time with the optics. The BetterLight 12 micron pixel size is, in itself, a big advantage with respect to avoiding optical failings, but there are many other differences too, including disadvantages of camera precision compared with some of the M.F. cameras, plus the inability to use the shorter lenses which are sharper and have less coverage and which are designed for digital use.
After taking a good look at this picture, at 100% in Photoshop, please then also take a good look at this MTF graph published by Schneider for this lens design, at f/5.7. The original PDF is available here: http://www.alpa.ch/files/products/151/AD24XL.pdf
The curves presented are for 15, 30 and 60 line pairs per millimeter, as you can see. Examine the graph at the left, the one for performance at f/5.7. All three of these graphs are for a reproduction ratio (1/Beta') of infinity, meaning infinity focus. The right edge of the graph shows the image circle radius of 30 mm (100% of u'max equals 30 mm), i.e. the corners of the image on the Dalsa sensor (with no rising front or any other view camera movements used). Lower lines mean less detail, so as you can see, the corners of this lens, when it's working to spec, are nowhere near as sharp as the center. Nevertheless, the image from this lens is pretty much sharp from corner to corner. This lens's recommended working apertures for optimum performance would be f/8 and f/11. At either of those apertures it would perform quite well w. respect to sheer focus, if the above JPEG example can be considered proof of what the MTF curves actually mean. The falloff is another matter though, making the use of a center filter with this lens quite important and indeed vital with a Kodak sensor. Note that the issue of "lens cast", as Phase One likes to call it -- the issue of color shifts due to increased angle of incidence of light on the chip -- is supposedly about three times less severe with the 33MP Dalsa sensors than with the Kodak 39MP sensors, and that as a practical matter, this means that this 24 AD lens should probably never be used on the Kodak chip, as the required correction for color shift can easily be more than what is possible. The 28 HR lens works fine with the Kodak chip, because it exhibits vastly less falloff and light hits the corners at a much better angle than is the case with the 24 A-D. Also, unlike with film, light falloff with wide lenses in digital use comes not just from the two sources traceable to the lens (inverse square falloff and cosine falloff) but also from the effects of light hitting the sensor at an angle -- the sensor is not equally sensitive in all directions.
So at 15 line pairs per mm, this lens, wide open exhibits 56% contrast at the corners. At 30 line pairs it's 37% contrast at the corners. And at 60 line pairs it's 18% contrast, on average at the corners. (In each case, averaging the radial and the tangential.) Theoretically!
And just for reference, if this lens is moved from its ideal infinity position for any reason by a mere 8 microns further from the lens, its plane of focus will shift to 200 feet (if my math is right). The maximum shift of this lens on its special ultra-high precision ALPA helical mount is just 1.6 mm (1600 microns) for the minimum focussing distance of about 2 feet. So a mere 8 micron error could be visible. So again, it really matters that each component be within 10 microns or so of the ideal for total system focus calibration to be right -- and even so, it could still be wrong, if the errors are additive. So some user-adjustability is indicated when the system is being integrated/assembled by the user, especially for cameras like the ALPAs where the three major sub-assemblies are coming from three vendors and not being tested as a unit (lens, shutter and lensboard calibrated by the lens maker, camera and adapter plate from ALPA and their manufacturer, Seitz, and digital back from one of the back makers). If the system is a Mamiya AFD II (or AFD III) body and lenses, or a Hasselblad H and lenses, the integration issue seems to be reduced, but only if the combined body and lens are calibrated to a maximum error of less than 20 microns, e.g. 10 microns. And still, the sensor calibration has to be quite good -- especially if the back is to be usable on other non-bellows cameras.
I'll have to check whether I already wrote this somewhere in this oh-so-long article, but an ordinary sheet of cheap laser paper is 100 microns thick. Ten microns is then, as you can see, a very small distance. A micron is 1/1000th of a millimeter, in case you missed that. So 25.4 microns equals 1/1000th of an inch, as 25.4 mm equals one inch (almost exactly). Yes, I did, but further down -- see the section called Microns, below.
These two things together (a successful sample image with a distant subject and the MTF data for the lens), can allow you to calibrate your own sense of how sharp a lens should be, when forming an all-infinity image onto one of these sensors (Dalsa 33, Kodak 39) with every part of the camera system working very nearly perfectly. If you can get your hands on the MTF data for any given lens, it can then show you how good that lens should be at the specified magnification ratio and aperture, right across the image from center to corner. Unfortunately, when attempting to make all-infinity pictures at very large apertures with super-wide lenses, results this good are far less common than they could be and should be, based on the experience with testing that I've had so far. The German digital view-camera style lenses from Linos and Schneider are supposed to be the best lenses available for medium-format work, and some of them do live up to their MTF specs, but in my experience so far, that would be well under half of them! When they do live up to their specs, the Rodenstock/Linos and in some cases the Schneider digital, view-camera style, fixed focal length lenses do indeed deserve to be regarded as the best lenses in the ultra-wide range, and perhaps also in the medium and longer ranges, again based on what I've seen [refer to a more recent article about Mamiya lenses on this site for more info favorable to the better samples from Mamiya]. These differences would tend to be apparent at large or medium apertures, in comparisons to glass such as the zooms made for the Mamiya AFD cameras and the Hasselblad 50 to 110 zoom, all of which are surprisingly good in practice, overall.
A Second Success:
Here is a another example. This exposure:
http://www.callme.com/lens_tests/35_A-D_f11_Inf_Exc_Hass39.jpg (10.4 MB JPEG)
was made by my friend Alain Briot (http://www.beautiful-landscape.com), with a briefly borrowed Hasselblad 39MP back at ISO 50 on an ALPA 12 SWA with a 35 mm Schneider Apo-Digitar at f/11 (the optimum aperture for this lens when little DOF is required) and the lens set to the infinity stop. Alain was certain at the time that he had used f/11. More recently he was less certain. I will trust that it was f/11 and the results certainly resemble f/11 results. No rising front. It was converted with Hasselblad's FlexColor 4.6.5 raw converter with no sharpening and no "lens cast" removal and saved as a quality 11 JPEG in sRGB. This is the sample that convinced me that it was time to finally make the big switch to digital capture. Careful study of the image of the moon proves that this capture shows more detail than any of the three most important moons I have captured with my view camera, taking the focal length and the number of pixels into account. The noise level of this moon is also far less and the color accuracy is far more. I believe that this sample proves that 39 million pixels of interpolated (Bayer chip) data can yield detail equal to the very sharpest color 4x5 detail (latest Velvia). Since I can't stand Velvia, however -- I greatly prefer Astia 100F for making 4x5 transparencies, though it isn't quite as sharp, I figure this capture at Zabriskie Point is roughly as sharp as a typical, low depth of field requirement, 5x7 capture would be on modern color film of my preferred types.
So if I can manage to get a camera with a proper tilt capability and which is free of the various optical defects discussed here, I should be able to routinely get images about this good, and with even higher center-to-corner detail uniformity (given the use of generally longer lenses), depth of field requirements permitting. And just imagine making three or more overlapping vertical frames to stitch into a horizontal image of 80 million pixels and more. That readily brings one into the quality range of roughly 7 x 11" or wider of modern color film, with respect to detail, and far beyond it in other respects. Or if the capture conditions are not as favorable (heavy stopping down needed due to focussing uncertainties), then perhaps the result would be more like 5 inches by 7 or more inches the long way, depending on the aspect ratio.
Just to help clarify things, no pun intended, here is a sharpened version of the file (be sure to check out the moon):
http://www.callme.com/lens_tests/35_A-D_f11_Inf_Exc_Hass39_SS.jpg (12.4 MB JPEG)
A BetterLight Reference Image:
I find this result to be pretty spectacular, though the fine detail contrast (MTF) near the corners is noticeably lower, and even the center is still not on a par with the best BetterLight scan back 100% image detail. Just for reference, here is my favorite example of a BetterLight capture which one can view on the Web, using the Zoomify technology:
BetterLight 6K sample at 100%:
http://www.betterlight.com/fullRes_zoomifyLIST.html
After reading the camera details, take the link to the Half Dome picture. Prepare to have your mind blown, then zoom to full magnification and take a look. Please be sure that your LCD monitor is set to display at its max available res in the relevant control panel or system preference.
Using Smart Sharpen in Photoshop CS 3:
And here are three screen shots of the settings used in Photoshop's Smart Sharpen to do the sharpening of that second version of Alain's picture:
One other thing about this picture is that it shows what Phase One calls "Lens Cast" more vividly than any capture I had ever seen. Notice the large color shift across the sky, caused by the wide angle lens illuminating the sensor at widely varying angles. This defect must be processed out after exposure by using a record shot made through a white diffuser over the same lens, hopefully at the same aperture, and certainly with the very same camera movements, if any (the raw processing software subtracts the pattern from the image and renders it more or less devoid of this defect). This problem is negligible when using SLR M.F. cameras because their retrofocus lens designs illuminate the chip with light at a much smaller range of angles from the perpendicular. The problem is also about one third as great with Dalsa chips as with Kodak chips, if Sinar is to be believed [the 60 MP Dalsa sensor in the new Phase One P65+ indeed exhibits not only far lower "lens cast" than the Kodak chip in the P45 and P45+, etc., but the pattern of color shift is far more radially symmetrical, which is another good thing, thus the Sinar information shown here is essentially confirmed to my satisfaction]. Here are the graphs from Sinar which show something of this comparison. The data are incomplete, but they do show both more falloff and more color cast with the Kodak chip. Of course there are various other differences which favor the Kodak chip -- or else I'd not have bought one.
And in order for you to see the pattern of the color shift from varying angles of light with a 28 HR lens on a Kodak sensor, I am going to throw in a 20% scaled down JPEG of a "lens cast"/record shot, made with the Phase One white diffuser held over the lens -- matte side resting against the front element housing. Examining this file in Photoshop can show you both the magnitude of the light falloff (vignetting) of the lens at this working aperture) and the magnitude of the "lens cast" in this particular RGB space, ProPhoto RGB. Note that if you see banding in the image gradient, it's due to your monitor's calibration LUTs (i.e. it's not really in the image). Remember that the next time you see banding in a smooth gradient. Figure out a way to turn off your monitor's calibration temporarily, and you can see the improvement in the appearance of gradients on screen for yourself.
28HR Lens Cast Pattern on Kodak Sensor:
http://www.callme.com/lens_tests/28_HR_LensCast.jpg (140 KB JPEG)
The rub comes in trying to actually get the kind of detail that we can see in this 35 A-D sample, in other super-wide captures, with other M.F. digital equipment. So far, the majority of combinations of superwide lenses, cameras, and backs that I've seen proper test results from have failed this basic test.
And here is the MTF curve, again from Schneider, for this 35 A-D lens at f/11. Once again the curves are for 15, 30 and 60 LP/mm, but this time the horizontal axis extends to 35 mm of image radius, 5 mm out beyond the corners of the sensor (so ignore the last 5 mm for analyzing this capture made without any rising front).
Comparing the MTF data for the 35 A-D with that for the 24 A-D, you can see that wide open, the 24 is actually superior to the 35, but that at f/11, the 35 is superior to the 24. For that reason and because of its much lower total falloff, I regard this lens are a much better performer.
Downloadable MTF Data:
You can download the complete MTF data for the Schneider lenses, only, from the ALPA website (the Adapted ALPA Lens area):
http://www.alpa.ch/index.php?path=products/lenses/adapted_alpa_lenses
Some Apo-Digitar information is available on Schneider's site too, e.g. here:
http://www.schneiderkreuznach.com/foto_e/dig/dig.htm
The Linos/Rodenstock digital lens MTF data are harder to come by, though I managed to acquire essentially all of it by pestering Linos endlessly. Some of their lens data is available directly on their site, as all of it should be. Here are the links for all that Linos has posted (as of April, 2008), all PDFs:
http://www.linos.com/pages/mediabase/original/rodenstock_apo-sironar_digital_e_2472.pdf
http://www.linos.com/pages/mediabase/original/rodenstock_apo-sironar_digital_hr_e_2473.pdf
http://www.linos.com/pages/mediabase/original/A-Sir_dig_HR_28_Neuh_2006_e_5758.pdf
http://www.linos.com/pages/mediabase/original/A-Sir_dig_70_Neuh_2006_e_5756.pdf
[And while I'm at it, I'll mention the new HR Digaron-S 23, the HR Digaron-W 40 and 50, the latter not shipping as of April, 2009. These apparently quite spectacular lenses can be explored on ALPA's site, where PDFs for the 23 and 40 are available.]
Conclusions from the Successful Captures:
Notice that the horizon line in the Death Valley picture is pretty darn crisp from the left edge to the right. Even the corners look quite good still, despite the subject matter being perhaps only 30 to 40 feet away there -- I think they would have looked a bit better otherwise. One can see, however, a definite loss in fine detail contrast toward the corners, which is what the MTF data say we should expect. Nevertheless, the corners and edges are very much in focus, suggesting no curvature of field issue with this lens and no other defects for that matter. An excellent result for such a wide lens (equivalent to a 90 mm on a 4x5 or to about a 24 mm lens on a 35 mm camera).
Again, this sample and the corresponding MTF data, help to establish that a lens with MTF performance at the level shown in the graphs can give quite satisfying detail from corner to corner -- if everything is more or less perfect with the camera system, and the subject allows it. I should probably mention that the Rodenstock 35 HR performance data show far better performance in the corners than this 35 Apo-Digitar from Schneider, and the 35 HR's flange focus is 54 mm (back of the shutter to the sensor at infinity), versus about 38 mm for the 35 A-D, so the "lens cast" issue is much lessened with the 35 HR, as is the falloff -- making the HR highly desirable, assuming it lives up to its specs, albeit much less physically compact and more expensive. The HR's longer flange focus is especially beneficial for the Kodak 39 MP chip, due to its greater sensitivity to light at higher angles of incidence. Perhaps owing to the very kinds of problems I have encountered in production samples of the 35 HR, a rumor has it that this lens is slated for a re-design. More to follow. [More recent rumors and lens announcements and introductions by Linos , through March, 2009, continue to suggest that both the 28 HR and the 35 HR are to be taken off the market, and my own experience suggests that it may be because these designs make achieving adequate QC too difficult, but I am only guessing about that.]
Microns:
Now... lets consider the precision requirements to actually achieve the kind of performance seen in these two good examples. There are 25.4 mm in one inch. A micron is 1/1000th of a millimeter. So 25.4 microns equals a mere 1/1000th of an inch (normally pronounced "one thousandth of an inch").
A ream of laser paper is about 2 inches thick. The ream includes 500 sheets. Each sheet is then 4 thousandths thick (0.004"), or about 100 microns.
The distance that a 35 mm lens needs to be moved to shift its plane of focus from infinity to about 200 feet is roughly 16 microns. For a 28 mm lens, that would only require about 10 microns! And yes, when shooting at a very large aperture, focussing accidentally at 200 feet instead of infinity is not an inconsequential matter with a 35 mm lens. The lack of sharp focus at infinity will be plainly visible. So if you had a subject which was all beyond 1,000 feet away, and you tried to shoot it at f/5.6 with a 35 HR, but the focus calibration were off by 16 microns, it would hurt the result quite noticeably. Please keep in mind that the optimal performance on the plane of focus for the 35 HR lens is indeed f/5.6 -- according to the MTF data. And Linos understandably brags about this unbelievable performance. With modern view camera lenses (made after the late 1970's) a lens that is optimal at an aperture larger than f/16 is phenomenal, let alone at f/5.6, let alone a super wide!
When a system like an ALPA 12:
http://www.alpa.ch/index.php?path=products/cameras
or the new ARCA R m3d:
http://www.galerie-photo.net/index1.html?lang=fr&lang=fr&target=d45.html
(use the search box on that page for R m3d) and
http://www.viewcameras.blogspot.com/ [showing the "Mark I" version -- there is now a "Mark II" version and yes, ARCA-SWISS still has no web site as of April 4, 2009]
...is built, there are several components which contribute to the overall focussing precision: 1) The infinity focus calibration of the lens on the lensboard and/or in its helical mount; 2) The thickness and/or focus calibration of the camera body itself; 3) The thickness of the digital back-to-camera adapter plate; and 4) The location of the effective surface of the sensor inside the digital back. And 5) If one uses a ground glass, its plane must match that of the sensor. Of course everything has to be flat and parallel also -- no swing, no tilt, no concavity, no convexity. And let's not leave out rotation (since I learned of a case of an obviously rotated sensor in a Canon 1Ds Mk III the other day). And I've also seen a sensor that wasn't centered on the optical axis of an ALPA, for whatever reason (inverting the back changed the composition). Nothing is perfect. But it's so nice when cameras come close.
If each of these systems is successfully made to be within a mere 10 microns of the ideal (1/10th the thickness of a cheap sheet of laser paper), the sum of the errors between them may still be 20 or even 30 or conceivably 40 microns. As we have seen, an inadvertent shift of that magnitude in the focus is enough to noticeably degrade the sharpness of subject matter upon which one has focussed with a short lens at a big aperture. Of course, much larger errors will show up with longer lenses too, even at fairly small apertures. My first Phase One loaner back was off by about 175 microns. More on that later.
First Bad Example:
Beyond the focus calibration issues (sensors being in the wrong plane, infinity stops being set wrong, etc.), the issue of lens performance is huge. Here is perhaps the most vivid example of a grossly defective lens, which has turned up in my sample of shipped, super-wide, German view-camera-type lenses:
http://www.pbase.com/image/79813605/original.jpg
Note that the lenses for which I have obtained revealing sample images, and about which I am writing here, do not, unless otherwise noted, include lenses which came to my attention because they were defective! In other words, the lenses are a small but nevertheless representative sample -- not lenses that were singled out from a large number by virtue of their poor performance.
This full-res JPEG is from a 24 Apo-Digitar again, shot by JIa-Yan, again at f/5.6, on the very same ALPA and Leaf Aptus 75 back as that used in the first 24 A-D example above. It's hard for me to believe that Schneider would ship even one lens this bad in 100 years. Just take a good look. The image is very sharp on axis and becomes horribly unsharp as one gets further from the axis, in all directions. This may be an example of extreme curvature of field, but because the subject matter doesn't allow us to see where the plane of focus was (if anywhere), off axis, we can't see if it comes closer near the corners, or if this is some other kind of optical defect. It may be the result of elements being spaced incorrectly, and to be fair, it's not inconceivable that something happened to move an element after the lens's final quality inspection in Germany. But I don't believe the lens was ever subjected to any great physical shock, for example. I can only guess at what may have gone wrong. Thankfully, ALPA and Schneider made it right, returning in its place, the lens which made the prior, very sharp sample. We don't know whether it was the same lens, repaired, or a different 24 A-D. Hopefully it was repaired. There is a tale of a second bad 24 A-D sample below, which sample I found very interesting. It clearly does exhibit curvature of field, as there is nearby subject matter in one corner which shows the plane of focus being very close there.
Second Bad Example:
Here is a sample unsharpened image from a third ALPA set to infinity focus with a 35 HR lens, with a Phase One P45+ back (apparently in reasonably good focus calibration), set to f/5.6 (one stop down):
http://www.callme.com/lens_tests/35_HR_f5-6_Inf_P45+.jpg (8.3 MB JPEG) This test image and some of the others to follow were made by Mr. Don Wood with my assistance.
Despite the overexposure, in the results we can see that on axis, the garage across the street and the tree just in front of it and to the right are in focus and show excellent detail. That puts the plane of focus on axis at about 120 feet. We can also see, unfortunately, that the horizon at the left and right edges is horribly blurry compared with that in the middle, and shows obvious color fringing, and the close subject matter in the lower corners reveals that the plane of focus near the corners is at a distance of about 25 feet where the plane crosses the axis! This is extreme curvature of field, bringing the effective, wide-aperture performance of this lens (f/4 or f/5.6) down to perhaps one tenth of what it should be, in MTF terms. Or less, possibly a lot less. And this is roughly a $4,300 lens when mounted on the ALPA board, calibrated for infinity, and also custom-optimized by Linos for ALPA, for infinity performance. And yes, we did invert the Phase One back and re-shot, only to find that the back is not to blame for this focus shift toward the corners. The chip was only very slightly swung, causing a tiny shift between the two frames in the plane of focus, off axis. We also know from other results with other lenses that the chip in this back is not significantly spherical or parabolic, nor have I seen any evidence of this kind of defect in any back yet.
Note that the lens was set to infinity focus. This means that if we are to judge overall system focus calibration by the performance on axis, that this system is off by about 27 microns. A good portion of this is apt to have been contributed by the Phase One chip calibration, but some may have also been contributed by the ALPA lens calibration, or possibly an unauthorized modification of the shimming in the ALPA back adapter plate (though we believed this to be unlikely with this particular adapter). In any case, trying to focus on infinity and getting 120 feet instead is also, like the dreadful curvature of field, totally unacceptable.
This lens should make images that are razor sharp on one plane at f/5.6 when properly focussed on that plane. Here are the MTF data for this lens at f/5.6. Note that the line pair frequency range used to illustrate the lens's performance reach an incredible 100 lp/mm and that the data are for a 50:1 reproduction ratio, or about 5.8 feet. The diffraction limits are also shown on the left and right vertical axes. The stellar falloff performance of this lens is also shown in the curves to the right. Note that the lens even beats the cosine falloff limit wide open. One more thing: Tangential = Meridional, and Radial = Sagittal. Line pairs tangential to the outer edge of the image circle are Tangential, aka Meridional. Line pairs parallel to radial lines passing through the optical axis are Radial, aka Sagittal. Hard to remember which are which.
You may notice that in the corners, if you interpolate to 60 lp/mm, this lens has more than twice the contrast of the 24 mm Apo-Digitar and over half-again more than the 35 mm Apo-Digitar. The MTF data for the 35 HR are the best I've ever seen for a superwide lens by a substantial margin [as of April, 2008, prior to the introduction of the spectacular 23 HR -- indeed it's getting more and more common to see amazing superwide performance as people introduce more and more modern designs with 14, 15 and more element formulae].
This is ostensibly a world-class camera system (the ALPA, 35 HR and Phase One P45+), and yet the results being obtained are really quite terrible, although where the lens is in focus, the performance is spectacular (check out the tree to the right of the garage across the street). The main culprit in this case must be the Quality Control at Linos (maker of the Rodenstock lenses). This lens was returned as defective by the dealer. And I was assured by Thomas Weber, one of the three ALPAs, that this had never happened before with such an ALPA lens. I should note that it was only two days later, that I obtained the disastrously bad 24 A-D lens sample shown above, and that this was also a lens supplied and which had been subsequently replaced by ALPA, six months earlier. I am not certain how these facts would be reconciled, however I have continued to see, as we shall see, seriously defective digital lenses from both Schneider and Linos as testing has proceeded, including lenses from other channels. At the very least, most of these lenses are not capable of matching the capability of the sensors to reveal fine data on a plane of focus. But a few of them are, at least part of the time. I suspect that the spacing shims between the rear elements of this 35 HR and its shutter may be of the wrong thickness, but I can only guess.
Again, if a lens cannot match, at least fairly closely, the published MTF specs, when testing at infinity and the largest aperture for which MTF data are provided, it must be regarded as defective, even though for many kinds of pictures, this defect might not be noticed. And this means the lens should exhibit essentially perfect focus (no curvature of field causing the corners to shift focus relative to the center) across the entire field, on the plane of focus. It's not OK to merely be capable of focus at some distance in each part of the image. The focus must occur on a plane or something very close to it. Of course, at infinity, the "plane" of focus becomes very thick in an optical sense. (Depending on the lens, from a few hundred feet away, give or take, to the ends of the universe.)
Perfection is Claimed:
Indeed (as of April, 2008) Linos/Rodenstock makes [made - the language has been changed] this very bold claim on the Linos web site here:
"The Rodenstock Apo-Sironar digital HR lens series was developed for special applications with extremely high resolution CCD chip backs with pixel sizes even smaller than 10 µm such as can only be realized with smaller digital camera formats. These lenses utilize every technological possibility to get as close as possible to the absolute limit of diffraction-determined resolution. Among other things, even the optical properties and the thickness of the CCD protective glass were taken into the equation of the optical correction.
The resolving power and lateral chromatic aberration have been optimized to ensure that the resulting lack of sharpness or the color fringes do not amount to any more than a tiny fraction of the pixel size (which can no longer be resolved). As a result, even when the digital photos taken with the lens are enlarged to a maximum on the screen, absolutely no color fringes are visible, unless color fringes are added by the pixel structure of a one-shot back used or due to interpolation."
Note that they are actually claiming that on the plane of focus, at some unspecified reproduction ratio (distance), that the image formed by all of their HR lenses at the recommended working apertures is so sharp that the blurriness contributed by the lens (and the chromatic aberration, i.e. color fringing) is a tiny fraction of one pixel, even for chips with pixels down to 5 microns! This is basically an absolute claim of perfection in every shipping lens -- and then some! If only it were so! What I have found has been light years from this kind of result, on average. In the above sample one can easily see abundant color fringing on the horizon at the left, and I've seen fringing in other HR results as well (see below). And the resolution off axis is many, many times lower than what is claimed -- in this case due to curvature of field. Again, I presume this is due to just one failing among the many, many Linos manufacturing processes for this lens -- curvature of field adjustment -- but I can only guess. Other examples follow.
Third Bad Example:
If having just one tiny thing go wrong can trash the performance of these optical systems, and the tolerances are so miniscule, it's perhaps not all that surprising that my experience to date has been that a strong majority of these systems, at least when the short focal lengths are involved, are clearly defective. It doesn't help that multiple vendors are involved, but if you think the Hasselblad H systems are immune -- guess again. One of the worst results I've ever seen was a sample image from none other than the "Hasselblad Star Quality" image set (since removed from the set). The middle of the image, made with the 35 mm lens on an H2 at f/16, was absolutely beautiful. The right edge was horribly out of focus for no apparent reason. Curving chip? Swung chip? Curvature of field combined with a swung chip? My money is on the latter, because the left edge was better. But it was severe. But don't take it from me: here are two small sections, both at 100%, first the center, then the right edge:
It's probable that the contribution to this particular system's poor performance by the lens was quite typical of this particular lens design. So far, I have seen zero samples from any 35 mm Fuji/Hasselblad lens that were not junk. If one looks at the MTF data for the lens, however, it's not surprising. The design just appears to be bad -- even at f/8 (stopped down 2 1/3 stops), and just 3/4 of the way to the corners, the 60 lp contrast is apparently zero (but that doesn't explain what happened above -- that seems to be mainly gross misalignment of the chip to the lens):
http://www.hasselbladusa.com/media/1f9ad1ce-5a29-49cc-97c5-244a70f5d446-HC35.pdf
The remainder of the Hasselblad H lenses are much better -- at least according to their MTF data, as well as in the few other tests I've conducted with them. All in all, it's a very nice camera system -- but this lens design needs to be replaced, and the chip quality control also needs to be improved -- apparently! I don't mind mentioning that some of the best samples of M.F. digital results which I have seen have come from Hasselblad H systems. Some of the samples Hasselblad is showing off now on their site show really superb handling of noise removal and sharpening. This zoom-able example is particularly impressive (if it's from their 35 lens it is a clear exception to what I just said):
http://www.hasselbladusa.com/products/virtual-demo-overview/hasselblad-image-quality/clarity.aspx (the one on the left -- zoom in all the way and look around -- it's gorgeous)
Always Too Close?
I should mention that in all of the many tests I've done and samples I've acquired [through April, 2008], whenever there has been an error in the focus plane, it has always been too close (the sensor has been too far from the lens). Many cameras, many backs, many lenses. This makes it likely that someone doing system component focus calibration is intentionally erring on the side of focussing closer than infinity when the cameras are set to infinity. Or maybe some piece of calibration equipment is itself out of calibration! [I encountered one exception to this rule in early 2009 when I received a Phase One camera body with the focussing screen out of calibration, causing the focus plane to fall further away than I focussed when using manual focus -- I focussed at 10 feet and got 11 feet, 3 inches, typically. That body was accurate with Auto Focus, however, at least when the subject contained a strongly contrasty line running vertically through the AF sensor area, though not so accurate when the same line ran horizontally through it.]
Phase One Chip Placement:
Phase One tells me that their sensor calibration accuracy standard is 12 microns -- an excellent standard. Presumably this refers to the height of the center of the chip. And presumably the chips are within, at most, a few microns of being flat. And presumably the swing or tilt of the chip is never more than a few microns either -- at least in their standard. One of the P45 backs I have used showed easily detectable swing and tilt, both. The other two that I have used, much less so.
Unfortunately, I have also seen an inordinately high percentage of P45 and P45+ backs shipped which do not meet this overall focus calibration standard. Keep in mind, one of the features that sold me on Phase One over their direct competitors (Hasselblad, Jenoptik/Sinar, and Leaf) and their indirect competitors (Seitz, BetterLight), was my dealer's extensive experience with digital backs, and the fact that he had never seen a Phase One back that seemed to be out of focus, but that he had seen other brands that were. Indeed I subsequently received much the same assurance from a senior Phase One USA officer.
My very first back, a P45 loaner, was out of focus calibration by approximately 175 microns. After spending many days unravelling the bad performance I was getting and then returning this fiasco (it damaged a lot of pictures that I was making on a trip to Colorado), I got another P45 loaner, and it was out by only about 50 microns. Still not good enough, but close enough that many people might not notice (this depends a LOT on which kinds of pictures you are making. Remember, longer lenses are dramatically less sensitive to errors in the placement of the chip plane, and smaller apertures cover up the errors too). [4-13-09: I am posting one additional unsharpened image as of Wednesday, April 14th, showing a classic fence test, which was made with my first loaner back (3.4 MB JPEG) here: http://www.callme.com/lens_tests/Fence_Focus_Test.jpg. The results are not significantly affected by curvature of field in the lens, as there was none to speak of. This exposure can give you a graphic sense of how much shift I'm talking about. The particulars are written on the JPEG.]
As I build my system, I want to establish the best performance that I can, so that in those few cases where I could get a subject entirely into focus at f.5.6 or perhaps even f/4, that I can use it if I need the speed and have confidence in the result. Some of these issues can be solved by focussing through the chip -- the way BetterLight does with their scanning backs, but we're not yet able to do that in the field with the M.F. backs (without using a laptop). Perhaps we will soon see it become possible to use an iPhone or an iPod Touch as a viewing screen for a Phase One back with live preview through the chip. This offers a promising avenue for perfect focus, including with a tilt, and a fine image preview, even when using view camera movements and without using a sliding back adapter, a ground glass, or removing the back from the camera. Tap to zoom to 100% and tap again to return to full-screen view. Set a focussing point or points, and get audible or visible feedback telling when focus is optimal for all such points. And we would even be able to check focus when stopped way down!
Finally, my third back from Phase One, the P45+, which I had purchased, arrived and it's chip calibration was close to perfect. Unfortunately, along the way I was also hit with a grossly defective replacement focussing screen from Mamiya for the AFD II camera I was using. I purchased it because I wanted to do stop-down metering with a manual lens. The screen was fully 0.025" thicker (over 600 microns thicker) than the screen that came with the camera, and the focussing plane shifted about 280 microns when using the replacement screen. Imagine focussing on 130 feet and getting 60 feet with a 100 mm lens. Or something like that.
So, in my own, limited, random sample of P-series backs, two out of three were bad. Fortunately my friends Charlie Cramer and Bill Atkinson have had three P backs between them, and all three were good (I presume within 20 or 30 microns or less of error in the back itself).
So that makes two out of six backs out of calibration. Was I just fantastically unlucky? If a back outside of focus calibration were a 1-in-100 event, my bad luck with the two loaner backs would have been a 1-in-10,000 event. Doubtful to say the least.
Another friend, Don Wood, with whom I've been doing many of these tests as I said, and who is generously hosting most of the sample images linked in this article, also purchased a P45+ in 2007 and was given a P45 loaner back to tide him over until the Plus back could arrive. The P45 was good -- his pictures tended to be in focus. But the Plus back was bad! He had to send it back to Phase One for calibration of the chip, and after three weeks, the same back was returned. After calibration it was at least close to being right, although in many tests, the focus was still well this side of infinity when infinity was set (including the 35 HR sample above with the curvature of field). The problem is -- how does one determine with reasonable certainty which part(s) are at fault when a system of four components is out of focus calibration? The best way is to have a system that works perfectly, and to then assume that it's components are dead on (rather than the camera or lens calibration or the adapter being off one way in one amount, and the back being off much the same amount in the opposite way.
If a back is off -- focussing too close by being, say, 100 microns too far from the lens, what are the odds that the rest of the camera will be off in the opposite way, canceling the effect of the back, by pure chance? One in ten? If it were so, and if there were two cameras, and each of them focussed correctly with their respective backs, and if the backs could then also be swapped and they still focussed correctly, then the chances would be one in 100. So we can be sure enough that both backs are good in such a case.
Similarly, if there are two identical cameras, with two identical backs, and one system focusses right and the other doesn't, and we then swap the backs, and the one that used to work right now works wrong and vice-versa, we can be sufficiently certain that it was the back on the malfunctioning system that is the culprit, with a certainty on the order of 100 to one. That's how I diagnosed my first loaner back.
It certainly helps a lot to have a friend whose system is just like yours, with whom you can swap components and see which combinations give accurate focus -- in case you get into this mess. Of course it doesn't help that the lenses used might not be able to be in focus anywhere except on axis when the back calibration is correct, but it's probably wise to judge the back calibration based on optical axis performance, even though off-axis performance could give a better overall compromise in the case of mild curvature of field and the lens makers might therefore have calibrated for off axis focus at infinity. I doubt it, but it's a possibility. Then again, that doesn't explain the bad curvature of the 35 HR in bad example #2 above, because it was worse off axis, not better.
So let's see... that's three back backs out of eight that weren't good. With random sample results like that, there is simply no way that a Phase One back with focus calibration which fails to meet their own calibration standard is an unusual event. It is, however, quite possible that most photographers getting backs that are only off by not more than, say, 50 microns, might not be able to determine the error with sufficient confidence to complain about it. Even I assumed at first that my erroneous focus results with my new P45 loaner were my own fault (at 175 microns). After all, the system was new. But no.
More Bad Backs:
Mark Dubovoy informs me that he has had four Phase One backs -- two P45s and two P45+s. The first P45 was not a loaner, and was for the Hassy V mount. It had a bad electrical connection for use in the vertical orientation. The horizontal connector worked. When shooting horizontally, the right side of the picture was out of focus, though it was in focus on the ground glass. That back was returned and he got a Hassy H mount instead. That P45 worked perfectly. It matched the groundglass focus all over. Mark mainly uses the 100 HR and 180 HR lenses on his Linhof M679CS, and that camera is what he was using the backs with. He tends to shoot at big apertures. It's possible that the culprit was the Hassy V adapter plate by Linhof, but I doubt it.
When Mark later upgraded to the P45+, he first tested a loaner from Bear Images in Palo Alto, my preferred dealer. It was out of focus. He went ahead anyway and got a new P45+, and it was in focus. So counting these four backs, my sample set is now at twelve.
The pool of loaners may be rich with backs that were returned as bad. We don't know if Phase One is doing that or not (loaning bad backs without having repaired them first), but in my sample, the number of bad new backs has been two out of eight, whereas in the loaner pool, it has been three out of four! Or a total of five out of twelve that were out of focus.
If you appreciate the way probability works, and if the observational technique was sufficiently good, you'll see that it's basically impossible that this sample does not prove that a very high percentage of the Phase One backs are shipped well out of focus calibration (or their calibration changed after being shipped). If each bad back were a 1 in 100 event, the odds of having five bad ones out of five would be one in 10 Billion. So the odds of having five bad ones out of twelve would be... well, just a little bit higher than that. If indeed the loaner pool is a dumping ground for defective backs, we still have to face that two out of eight brand new ones were bad, plus that the various bad loaners had been shipped as new at some point too. Surely, the chips don't move in any significant way as a result of use. Could they? Perhaps there is an unrecognized problem with movement after the fact?
More Bad Lenses:
Here's where I start to wonder how many examples of failure to meet specs I should show you. I have sample images from at least three 35 A-D lenses. Only one of them is worth a darn (it's a beauty) -- the other two are complete garbage. The two bad ones were lenses supplied by Calumet to Don for testing. And these lenses were selected at random, not because they were bad, so the sample is small but still representative. The Cambo/Calumet lenses were probably assembled by Cambo from separately purchased front and rear element assemblies and a Copal shutter, and the critical element spacing step may have been omitted or otherwise done incorrectly, unfairly reflecting poorly on Schneider's QC.
Here are three samples from one of the two lenses supplied by Calumet, shot on a Cambo wide camera, at f/5.6, f/8 and f/11, all focussed at infinity, all with Don's P45+:
http://www.callme.com/lens_tests/35_A-D_f5-6_Inf_Cambo_P45+.jpg (4.6 MB JPEG Quality 10)
http://www.callme.com/lens_tests/35_A-D_f8_Inf_Cambo_P45+.jpg (5.4 MB JPEG)
http://www.callme.com/lens_tests/35_A-D_f11_Inf_Cambo_P45+.jpg (5.7 MB JPEG)
In this series, one can see the dramatically worse performance off-axis in all cases, but also the rapid improvement off axis as the lens is stopped down. Unfortunately, at f/11, this lens's theoretically best aperture, the center-to-corner difference is still rather extreme, especially in the lower right. The other of these two 35 A-D lenses performs about like this one. Neither seems to come close to the 35 A-D sample from Death Valley that is the second successful image shown above in this article (and which launched my quest for a M.F. system!).
The 28 HRs:
I have seen and carefully studied images from three 28 mm HR lenses. One of them shows mild falling apart in the corners (one corner actually, as the other lower corner has been cropped off), I believe considerably greater than what the MTF data show should occur, making that lens what I would consider defective, although it is nevertheless the best of the three and in other respects quite impressive. I think this 28 HR was one of the Sinar-supplied lenses -- potentially cherry-picked [I have since heard that these lenses were all built on contract for Sinar but that they did not take all of them]. I had a link to the cropped image, but it has apparently been taken down. It is by Rainer Viertlböck, a high-end architectural photographer from Germany, http://www.tangential.de. Rainer recently made a big splash with the announcement of the rather spectacular Sinar arTec camera, which he largely shaped with his feature requests, steering the camera toward ideal performance for architectural photography with MF digital backs. See: http://www.image2output.com/user_resources/TextFiles/pdfs/Sinar_Artec.pdf
and: http://www.sinarcameras.com/site/index__0-e-2019-50-2158.html
This sample 28 HR image was made from a very tall crane, over Munich. Only the lower left corner belies the imperfection of this lens. Perhaps I can get him to share this image again as an example of a nearly-to-spec 28 HR result. Or perhaps you've seen it and remember it.
The second 28 HR is the one used to make this picture:
http://www.alpa.ch/files/knowledgebase/23/pxm_000090.jpg
Examine this image carefully, and you will see that the infinity performance in the middle region is rather good, despite the focus being compromised for the foreground at the right, but the infinity performance all along the left edge is horrendous -- perhaps six times less detail than in the middle. And what little is visible of infinity along the right edge looks to match the left side. This 28 HR is the worst of the three.
The third 28 HR is the one used to make this picture, with an ALPA TC, focussed at infinity, on Don's P45+ (He believes it was shot at f/22, and the recorded shutter speed and ISO suggest this is so):
http://www.callme.com/lens_tests/28_HR_f22_Inf_P45+.jpg (9 MB JPEG Quality 11)
Overall, this is a very good result, but zoom to 100 percent, then follow the horizon line from the center to left edge, on screen in Photoshop. The detail should be equally crisp all the way across, but it's falling apart near the left edge. The right edge seems better. The lower left corner is questionable, but the subject matter makes is hard to be sure. But also take a look at the balcony at the left in the distance, behind the Rite Aid store, which shows a three-pixel color fringe. My sense is that this may be an example of a lens which is unsharp in one region only (not exhibiting good radial symmetry in its performance). I've seen a 35 HR which is quite obviously guilty of this particular failing (sample to follow).
Here is the small section from the left enlarged to 200%, making the fringing glaringly obvious. Recall that Linos promised no fringing! (Unless caused by the RAW conversion -- which this was certainly not.)
http://www.callme.com/lens_tests/28_HR_Fringing@200.jpg (0.2 MB JPEG Quality 10)
This is a USD $6,300 lens. Unless you buy it from Sinar, in which case they will ding you $11,000 or so (and you might get one cherry-picked from a lot provided by Linos). At this price -- or any -- obvious defects in a lens for finely detailed work are simply intolerable (even if they hadn't promised perfection).
Another 24 A-D Disaster:
This example is the exception to the rule, inasmuch as I did learn about the lens solely because it was defective. It is therefore not a representative sample. Nevertheless, because it is yet another example of a Schneider 24 mm Apo-Digitar XL lens being extremely poor (assuming it isn't the same lens returned by Jia-Yan above), it does prove that the first example isn't the only disaster of this lens type shipped by somebody. It was shot with a Horseman SWD-II at f/11, processed in FlexColor, sharpened a bit with USM at 100%, radius 0.5. Posted in Luminous Landscape's forum on February 9th, 2008 by Jordan Reeder http://www.jordanreeder.com (thank you Jordan).
http://www.jordanreeder.com/testing/rod_24mm_22mp.jpg
Perhaps the most interesting thing about this sample (which also shows both the extreme falloff and the severe "lens cast", both of which would be dealt with in practice, if possible!) is the overhanging eave of the house in the upper right clearly shows that the plane of focus has come very close to the camera in that corner of the image. This lens too is suffering from extreme curvature of field! Maladjustment at the factory? In this case, it may be entirely due to Horseman having failed to use the proper spacers behind the shutter when assembling the lens from components purchased separately -- in which case the fault would not lie with Schneider, though in the prior case, it apparently did.
Note that the falloff with wide angle lenses on these sensors goes well beyond that of the lens alone, as I mentioned above. When light hits the sensor from a wide lens at some angle of incidence greater than zero (i.e. perpendicularity), the back itself contributes to the falloff (much more so with the Kodak chip than with the Dalsa chip, apparently). So you can't predict how much darker the corners will be by looking only at the falloff data provided by the lens maker. This goes a long way toward explaining the striking relative appearance of this 24 A-D sample and the above 28 HR sample image of L.A. (which was also shot without a center filter and had no vignetting removal done in processing and no "lens cast" removal either).
Again, you must examine the images in Photoshop or the like at 100%, not at some lesser magnification in the browser, to see what I'm talking about.
The 35 HRs:
I have also seen and carefully studied the results from three different 35 HR lenses, again chosen at random. These lenses did not come to my attention on account of their problems.
The one mentioned above is the first of the three (the second bad example above). It's terrible and was returned to ALPA/Schneider by Samy's as it should have been (but not until Don and I had spent many, many hours in testing -- and in shock at the results).
The second 35 HR is the one that made this picture by Jia-Yan in Taiwan (ALPA, Infinity focus, f/8, Leaf Aptus 75, 33 MP):
http://www.callme.com/lens_tests/35_HR_f8_Inf_Leaf33.jpg (6.2 MB)
If you study this picture, you may notice that it's soft on the horizon at the right edge. When I had the photographer invert the lens on his ALPA, the blurry area shifted to the left side of the picture, proving that the problem is with the lens -- not the Leaf Aptus 75 chip plane.
Here is a lovely picture made with this same lens, which shows the soft region better. Compare the detail at the right edge, where the hill rises above the road, with the same image height at the left edge:
http://www.pbase.com/jyc3kimo/lens_test&page=all (this is Jia-Yan's page full of test images at full res)
http://www.pbase.com/jyc3kimo/image/93245925 (this is the JPEG I wanted you to look at, 12.2 MB -- choose the link to the original, i.e. full-size version)
So, this lens is not bad overall, but still has an apparent defect in the result which I would find difficult to accept in my pictures. This would force me to abandon nearly all large aperture (f/4 and f/5.6, if not also f/8) use of this lens and work only at f/11 and higher or even f/16 and higher. This lens is most definitely supposed to perform best at f/5.6 and also very, very well wide open, at f/4. It is not performing up to spec.
The third of the three 35 HRs is one that was again tested by Don with my assistance. It belonged to another photographer and our tests were done with that photographer's Phase One P25+ back, a 22 million pixel back, with 9 micron pixels. Again it was an ALPA, and again it was set to infinity and this time a variety of apertures: f/5.6, f/8 and f/11.
This system gave an excellent result at infinity focus, which you can see here. This is the f/8 capture, because the f/5.6 capture was too badly overexposed to be very useful. Nevertheless, this lens, camera and back are showing very uniform focus along the horizon, and excellent performance in the lower corners too (though the lower right corner is quite obscured by the foreground tree).
http://www.callme.com/lens_tests/35_HR_f8_Inf_P25+.jpg (4.5 MB JPEG)
This one and the f/11 capture are among the best results from a super-wide with M.F. digital capture that I've seen. I am leaving it to you to play with these unsharpened files (except as otherwise stated) as by sharpening them to taste. Do keep in mind that it's a little easier to look really sharp at 100% with a 22 MP back than with a 33 or a 39.
Unfortunately, when this third 35 HR was used to make this photograph just a minute later:
http://www.callme.com/lens_tests/35_HR_f5-6_20-Foot_P25+.jpg (4.1 MB JPEG)
where the plane of focus was set manually to 20 feet on the lens's helical mount, the picture changed terribly. The center portion of the horizon is only slightly blurred, but the left and right ends of the horizon are blured three or four times as much! Examining the corners of this capture, we can plainly see that the plane of focus is again curving strongly inward in the corners.
This kind of shift -- from excellent performance at infinity focus to terrible performance at 20 feet focus is quite inexplicable to me. I realize that lenses achieve optimal flatness of field at a given magnification ratio, but to shift by this large amount from such a tiny focus adjustment is well beyond bizarre. The weather was extremely warm. The rooftop pictures were made in 110 F ambient temperature, followed by working on the balcony a minute or two later, in lower temperatures. We don't know what effects the heat may have had on the system, but the Phase One Plus family of backs is supposed to work fine at up to 180 degrees F (and down to 100 below zero F). If we can't trust this lens design to continue to exhibit flat field focus throughout its medium-to-long distance focussing range, it would mean the design is useless. What's up with this lens? Could this be the main reason that the aforementioned rumor has it that this design is slated for replacement? Or is it rather that the design is inherently more susceptible to curvature of field problems due to slight spacing errors? I can only guess.
Concluding:
So let's count up the lenses:
1) Two 24 A-D's (not counting the third one), 2) Three 35 A-D's, 3) Three 28 HR's, and 4) Three 35 HR's. And out of these eleven, how many appeared to meet the MTF spec? Two for sure and three fairly close but not really. Or maybe three with two not quites. Two to five good lenses out of eleven. And I'm still nervous about some of the five. Really what I'm looking for is just to be able to have the same thing (or better) from MF digital that I've had with my 4x5 for decades, and even to achieve it using the larger aperture optimums that Linos claims they are achieving.
I take that as approximately a one-in-four chance of getting a good superwide, view-camera-type, German lens when you order one. And for every lens that gets rejected by one photographer, if the thing isn't made right by Schneider or Linos/Rodenstock or one of the other vendors who assemble elements to shutters (possibly including Linhof, Sinar, Horseman, and Cambo), it then just serves to reduce the chances of getting a good one for the other photographers to almost zero. And these things aren't exactly in great supply to begin with.
The sorry thing is that other kinds of super wide lenses which might be useful with this format are generally worse. The 35 mm Hasselblad HC f/3.5 lens (built by Fuji) is a real lemon, as mentioned above. One of those was Don Wood's Hasselblad HC 35 lens, which is so blurry around the edges that we both regard it as useless -- it's simply impossible to get a sharp picture, even if you stop down to f/22.
Here is a sample shot at f/5.6, with the lens manually set to the farthest it would go (beyond the infinity mark) and the plane of focus is still closer than infinity (presumably the sensor is off). Examine the horizon. It's a fright.
http://www.callme.com/lens_tests/35_HassyHC35_f5-6_focussed_beyond_Inf.jpg (7.9 MB JPEG)
So far, the Mamiya lenses that my friend Charlie and I have used -- even their three zooms -- have turned out to be a happy surprise, though contrasty edges near the corners do show a lot of fringing in the wider focal lengths, especially of the shorter zoom, the 55 to 110. Still, I've seen no evidence of curvature of field ruining the edges of any of my Mamiya exposures [later testing with other Mamiya lenses showed plenty of curvature of field problems with some lenses of a given type, and to sometimes greatly varying degree from corner to corner -- more on this in a subsequent article].
Again, remember that with longer lenses and smaller apertures, the focus calibration errors and problems become substantially less. And hopefully I will find that the longer German lenses all tend to meet their MTF specs. But I'm still worried about even that, and I know now that I will have to very carefully test every lens I purchase, with the assumption that it will not pass muster.
Linos and Schneider -- get your act together!!! Make sure every lens shipped comes at least quite close to the MTF specs you yourselves have computed and in many cases published, if not guaranteed. If anyone on Earth is capable of superior quality control, surely your two companies are.
And Phase One -- you too! Make sure every back shipped is within your own 12 micron focus calibration standard. And that should include the loaner backs you give people.
Photographers -- expect trouble and look for it. Make wide aperture test shots at infinity and examine the results. Check to see that all four corners and the middle are sharp, and more or less equally so. Expect to have trouble focussing your M.F. digital cameras -- simply because the precision required is so extreme. [I intend to write an article on a great method for facilitating accurate manual focus of the Mamiya and Hassy H bodies soon, in early April, 2009.]
Camera designers: user-adjustable calibration systems, like the ALPA adapter plate shim system, are, I think, vital. The ARCA R system, with its individually calibrated lens mounts using shims of variable thickness, could potentially provide even better flexibility for system calibration that the ALPA adapter plates -- if ARCA-SWISS will allow us to re-calibrate our own lenses with shims to get our systems into perfect calibration. User access to the infinity calibration of the Rodenstock and Schneider helical mounts would also enable a robust system to insure that the user can overcome slight errors of chip plane in the back as well as slight errors in every other system component.
I think we have to accept that these small errors are inevitable and provide the means to deal with them fully to the users. But at the same time, according to all of the above observations, the lens Quality Control and the chip plane Quality Control both need to be greatly improved.
On a happier note, I am very pleased with my P45+, given that it's focus calibration appears to be very close to perfect (perhaps just within the 12 micron standard, but if anything, focussing a trifle closer than it should). The interface of the back is great. The physical quality of its build is great. The dynamic range and detail are great, though the usable dynamic range is not as good as is claimed. The color accuracy is fantastic. The shot to shot speed is sufficient. The long-exposure capability is tremendous. The battery life is pretty good. The LCD is pretty good. The blinking highlight warning for overexposure works way better than Canon's [prior to the 5D Mk II, which came later and got better] because it's full-screen and it is totally effective at avoiding even the finest highlights being blown out if used properly. The histograms for underexposure assessment work very well (although they show nearly one stop being lost at the bottom which is not lost when doing the actual raw conversion). And the Plus backs allegedly have superb environmental toughness -- water resistance, extreme cold and extreme heat resistance. It's a terribly expensive piece of equipment, but it's pretty awesome too.
I have been waiting many months to receive one of the first batch of ARCA-SWISS R m3d cameras and a first lens, for close evaluation. I am optimistic about at least the possibility that this camera can do what it needs to do, with respect to precision. The zero detent for the tilt mechanism will need to be ultra-precise. The camera body's focus calibration will need to be extremely accurate. The infinity focus calibration of each lens will likewise need to be extremely accurate. And the ground glass, which I hope to be able to avoid using, even with tilt and shift movements, will also need to be super accurate so that focussing done on it matches the result from the Phase One back. These things are true even if I am ultimately forced to abandon hope of sometimes shooting as wide open as 5.6 or even 8, when I need the speed.
When all is said and done, if it works as I hope, the R camera should provide an optical capability which is, all things considered, a substantial step up from my trusty Linhof Technika and its seven beautiful lenses. if the R doesn't work out, I will have the option to use one of the ALPAs instead, but even with the new ALPA tilt adapter, I will be limited to using a tilt with lenses of 80 mm and longer. I really need a tilt down to the 47 mm lens.
And yes, I have used Helicon Focus to perform focus blending as a means of extending DOF. And you should too! Combined with a fine tilt, we'll be able to get more stuff in focus than ever... [As described in the article I posted on March 30, 2009, I was able to devote a great deal of time to collaborating with the primary creator of Helicon Focus to extend its capabilities and make it a much more reliable tool for increasing depth of field without a quality penalty. I should also mention that after my two-year quest to find the most effective medium format system with which I could continue my work, I surprised myself by choosing the Phase One body and using Mamiya lenses, and I am happier with them than I thought I could be. I may also write soon an article about <that> process and what balance of factors led me to choose the Mamiya/Phase system over the many other interesting alternatives.]
—Joseph Holmes
Nit-pickers of the world, UNITE! -- just kidding. :-) Upon further reflection, perhaps I'm not!
ADDENDUM
April 5, 2009
This article was essentially about one aspect of my quest for a digital capture system which I could feel good about using, and which would get me as many of the qualities I am after as is feasible. In stark contrast to the selection of the camera type which I used with film, this selection process was incredibly difficult, as there were many camera systems of interest, none of them seemed to have a clear edge, and some of the best ones were only prototypes, most notably the ARCA-SWIS R m3d, which required a lot of waiting to be able to assess properly. The mechanisms involved were mostly exotic machines that one might have to travel to New York or even Europe to see. And each system can't really be tested unless one has all of the components, but they come from different companies and sources. So actually getting to know one of these camera systems can be a nearly impossible task. That's one of the reasons I wanted to write and publish this article.
The camera systems are now so thoroughly intertwined with the software that works with the images, that one can't think of cameras anymore without understanding the processing issues and potential. Perhaps the two biggest examples of this have to do with the stitching process and the focus blending process. Each extends imaging in ways that have been impossible until recently and directly alter the envelope of the camera system. The hardest single thing to give up was the fine tilt capabilities of my 4x5. A number of the candidate cameras had tilt, but to get it they sacrificed several powerful advantages of a sytem like the Phase/Mamiya body and lenses. In the quest for excellent optical performance, in the context of the incredibly revealing sensor, I was very unhappy to discover the major discrepancies between the MTF data, published or not, for many lenses from lens makers I've always been able to count on (the Rodenstock and Schneider brands had never let me down before with a modern, computer-generated design). This makes a tough shopping problem dramatically tougher, because I loathe systems which can't give me corner to corner sharpness. Imagine having your own camera or lenses ruin your pictures. I'm the only one who should be able to do that.
As for our findings regarding the focus calibration of Phase One's backs, I still can't explain how it is that I have seen such a very high percentage of bad ones, yet my dealer's experience has been much better. Regardless, it has become crystal clear to me that the days of assuming that a lens or camera body are either in calibration with respect to focus, or perform up to the MTF specs of the lens are as well as other samples of that same lens, are over. I have repeatedly seen lenses be bad, digital backs be bad, and camera bodies (or perhaps just their focussing screens) be bad. This puts the photographer in an incredibly awkward position. Even if you have an angel for a dealer, the best you can hope for is to get the problem sorted out after having wasted a few weeks. Just to give you an idea, although many of the Mamiya lens designs yield quite stunning results, I have seen shocking amounts of lens-to-lens variation in quality. I had to test seven Mamiya 80 mm lenses (six of them were the spiffy new 80D and one was one of the older 80 AFs) to find two good ones. They were all pretty much the same if you didn't look at the corners, on the plane of focus.
At this point, I am quite happy with my camera system, despite the ordeals and apart from Phase One having apparently decided to put the 45-90 zoom lens on hold for a year or longer (Phase One now has controlling interest in Mamiya and is calling the shots). This was apparently a marketing decision, at least in substantial part, which is a big problem for me since I predicated my outfit purchasing decision in significant part on the arrival of that lens, to replace and to augment the coverage of the older 55-110 zoom, which has performance that isn't up to what we need with a digital back.
I also need to throw in a paragraph here to say that I have more than a suspicion that Linos/Rodenstock's new line of wide angle lenses for digital capture, the 23, 40 and 50 mm, are all going to be consistently spectacular and not suffer from the very high defect rates seen here in the 28 and 35 HR lenses. I have only seen pictures from a single 23, and it was stunning. Not only perfectly sharp allover, but with unbelievably low light falloff with a very strong rising front movement, and negligible color shift due to sensor angle issues, assuming no post-processing was applied. Probably the best extreme wide-angle performance in history. This doesn't prove either consistency or anything about the 40 or the 50, but somehow I can just tell — between the whispers of the 28 and the 35 being discontinued and my many years of finding Rodenstock lenses to be the best in the world in many categories for large format work, I just have a very strong feeling that these lenses will be reliably awesome, especially when coupled with a new P65+ back from Phase One, with it's larger area requiring more coverage than many of the digital lenses can provide, very low sensor angle problem (Dalsa chip), unbelievable detail, higher dynamic range, higher usable ISOs, etc. The Linos 70 is another great favorite of mine, as is the 100 HR. Avoiding the problem of removing the back to compose still looms large, however. Not being into architecture though, I can largely cover the wide end with stitching. I may get by with 45 mm as my shortest, though at times I'd want to go down to 35.
And on another more positive note, software and processing (8-core Mac Pro Nehalem with 24 GB of RAM and super-fast Intel E SSD-based RAID scratch coming soon) advances have finally led to my outfit being capable of coming very close to matching the holy grail of quality — BetterLight scanning back files viewed at 100% and bigger. Or to put it another way, photographs can now look like what I see, in terms of color, tonality, and detail, in a wide view. Very often, digital files have to be viewed at 50% on a display, to look both sharp and free of artifacts. But now, thanks to the superior interpolation and processing options in Brian Griffith's RAW Developer application (Mac only), from IridientDigital.com, I see 100% and 200% results that blow me away. This in turn means that 7000 pixels can look very sharp and clean at 40 inches of print size. My stitched files will typically be about 7,000 pixels on the short edge from this camera system. Any more really does start to get too sharp. I never imagined that such a thought would cross my mind. After all, black & white contact prints are the quality standard in printing that I've always enjoyed the most, with respect to image detail. If you should ever have the chance to see any of Brett Weston's 11 x 14 contact prints, take it.
It's been three years since I've exposed a sheet of film. My site doesn't otherwise give any hint of this, since I've not updated the pictures for four years, owing, in part, to the perfect being the enemy of the good. But it shouldn't be too much longer now. The last big technical transition in my 40+ year quest for fine color images of this dear Earth is nearly completed, and hopefully now I can get back to spending a big fraction of my time on adding to what I have to show you.
— JH
PS: The Holy Grail. OK, here's what I'm talking about. The JPEG visible below is less than 1/27th of one 39 million pixel frame, at 100% magnification, after processing in RAW Developer, including sharpening (yes, the interpolation and sharpening in RAW Developer are so good that I'm willing to sharpen near the beginning of the process, which has heretofore been taboo). The lens was a Mamiya zoom, the relatively lowly 55-110, at roughly 80mm and f/11, 1/60 second, and ISO 50 with the Phase One P45+ back.
Below this small section I have offered two downloadable files, both quality 12 JPEGs in my DCam 2 working space: the first is a 2 million pixel section of the image from the center, and the second is the entire image, so you can really check it out. I can see a very modest falloff in lens performance, especially at the right edge of the full image, which would not be present in one of the better lenses. If the corners had subject texture, we would see more falloff there in this image. Only when the image detail is uniform throughout the image does sharpening work just right, otherwise it's too strong in some areas and ineffective in others. The sharpening in the visible sample is about ideal, as this section is a little off center. Still, the result is not only fully sharp overall at 100%, but nearly devoid of any sharpening artifacts that I can detect (my other two Mamiya zooms, the new 75-150 and an old, manual focus 105-210 are both sharp, corner to corner, wide open, at all apertures, and usable from 5.6 on up).
Be sure to zoom to 200% also, and consider how your files look at 200% after making them look sharp at 100%, and what that means for big enlargements. Now consider that this one frame is 5428 x 7230 pixels, and that it's part of a pano with ten overlapping frames. Single frames from this camera can now be consistently sharper than the sharpest 4x5 I have ever done, and indeed better than all of the old-time B&W 8x10 that was my holy grail when I was younger. I still can't get over what I'm seeing here. For nearly two years after the 39 MP sensor backs hit the market, I had never seen a result that even approached this, and indeed I had never seen anything this good from any area-array capture device (except one aerial shot made with a 22 MP back) until just a few weeks ago. It was difficult to achieve, and to do this consistently, optimally, from corner to corner does require great lenses, very careful focussing and optimal processing, but what a relief to finally see this level of quality in a practical, albeit expensive, camera system. And the new P65+ can do even better.
Download a 2 MP cropped section from the middle of the file here. (2 MB JPEG)
Download the entire 39 MP frame here. (21MB quality 12 JPEG — thank Don for the bandwidth, and please save the file to your hard drive rather than downloading it repeatedly)
PPS: You can actually see the Old Man of the Lake floating in the water in the lower right in the full frame. It's an inverted tree trunk that's been floating around in Crater Lake for at least 113 years. See http://en.wikipedia.org/wiki/Old_Man_of_the_Lake.