Canon digital and 35mm format issuesWhat about the full size (35mm) of a solid-state imaging sensor

In my original article about the Canon 5D I reflected on the technical and more philosophical aspects of the 'Barnack'-format for digital cameras. I noted that the 5D was a milestone camera because of feasibility of a large sensor at an affordable price. It was and it is my view that the 35mm format and the matched optics to create classical viewing angles and classical depth of field gradients, in addition to the artistic aspects of using the 2:3 format that has to be mastered before you can compose interesting pictures.

Olympus went for the easy way with the 4:3 format, that is much easier to use and conforms to the ubiquitous TV screen format that is a constant visual companion on today's culture.

In the current issue of AP, Geoffrey Crawley looks at the aspects of image quality tht is attainable with the full format (35mm) and the APS-C format that is roughly equivalent to half frame, and retains the 2:3 relation. He concludes that image wise there is a draw: both sensor areas deliver the same imagery. He compares two systems that have about the same pixel size and then it is not surprising that theoretically and with test charts the same image quality is possible. But there is much more to analyse here. In my comparison of the M8 with the 5D, the Canon gave somewhat better resolution despite having a larger pixel size. It is the software stupid! You can hear Bill Clinton shouting. And my Siemens star results indicated that the MTF values in the region from 30% to 10% of the Nyquist maximum are critical for effective image quality.

Presumably the debate will go on for a while and that is fine. We simply do not know that much about digital capture and digital processing as we know about chemical processing the silver halide grain. Here we have an history of 100 years of cumulative experience, but in the digital arena our knowledge spans hardly a decade. And myths are already all over the place!

You can like or hate Canon, but one theme is obvious: here we have a company that has a very steady course and a very clearly defined goal for the next ten years. Some cameras that were introduced over the last thirty years might draw negative comments and did not become world beaters. There main fault in retrospect has been to focus too long on the amateur market and leaving Nikon alone in the professional pond. But since the EOS body emerged around 1985, the company exhibits a singular drive.

The new D1 packs the sensor of the 5D in a really robust body, the film-loading 1V (end to that era?). The capture speed is very high and there is that mysterious comment that the 1D has no memory buffer, presumably wring directly to the flash card. The new 1Ds shares the same body and brings the pixel count to 22 million on a 24x36mm area. The most intriguing remark is Canons statement that from now on there will be no more 1.3 crop sensors. The strategy then is clear: the amateur market will be served by APS-C with 1.5/1.6 crop factor and a new range of lenses. The professional high-end market will be exclusively served by full-format sensors allowing all Canon lenses to operate at the true computed focal length and viewing angle. Canon seems to be quite confident that the problems with the 35mm format can be addressed and overcome. There is now also an ISO 6400 value available. The new cameras will be formidable instruments, the 1D attacking the professional market for mobile photography and the 1Ds (with 22 M pixels) attacking the medium format stationary (studio) photography. There is a risk here: many professional reportage photographers do not want nor need that huge amount of pixels. Is Nikon smarter in this respect and listening more closely to the market?

Nikon continues to state that they will not embark on that route and stay faithful to the APS format derivatives. For how long we may ask?

The 1D will retail for 4500 dollars and will be cheaper than the Leica M8. This is not a clash of civilizations, but a minor clash of belief. The M8 couples a mechanical film-loading body to a solid state sensor and retains as much of the classical values as can be done within the technological constraints.

The 1D couples a fully electronic film-loading body to a full 35mm format solid state sensor and skips as much of the classical values as can be accomplished within operational constraints. There is no doubt in what direction the mainstream buyer will move. Canon is shaping the market and the others are more or less responding to market trends as they are perceived by the gurus.

Leica has been troubled over the last decade by a most erratic decision making process by a series of CEO's and this decision making is reflected in the current stable of products. One really hopes that there will be now a steady developing strategy with a clearly defined future vision for the product range to be developed.

The Canon strategy is very consistent and very profitable. Their choice for the classical 35mm format for the high-end camera system is somewhat surprising, given the fact that they are alone in this choice, but then the market power of Canon is quite strong.

Now the competition must react.

What about the full size (35mm) of a solid-state imaging sensor
Intro
Around 1955 two icons of German engineering and drive for perfection were introduced. One was a BMW motorcycle, the full swing R50 and the other was the Leica camera M3. Both products exhibit the almost missionary passion to be an optimum of design excellence and to maximize the engineering appeal. The subsequent history of both marques is well known. The R50 stayed in production till 1969 and was followed by the R50/5. The /5/6/7 series was stubbornly defended by chief engineer Mr von der Marwitz, who refused to accept that the design was no longer suited for the new motorcycle requirements of the eighties. But with that typical German blend of overestimation (when things go well) and despair (when things go bad) major changes were not introduced until the management in Munchen radically swept away the motorcycle design group and allowed car engineers to produce new ideas. This was the start of the four-cylinder inline series and later the air/oil-cooled boxer twins.

The M3 stayed in production till 1969 (coincidence?) and was followed by the M4. The current M7 can be compared to the /7 series, an update within the limits of the basic design parameters, but a radical change it is not. Now the typical Leica clientele do not want changes, but demand at the same time that Leica delivers new products that incorporate the best of the current crop of (digital) products that are on the market. These requirements are obviously in conflict. BMW made a clear choice: how good the old designs, a radically new mindset is required to create future oriented products, fit to challenge the competition.

Full size sensor (35mm format)

The current mindset of many photographers assumes that a sensor with the size of the classical 35mm negative would be the best choice as it has a large area and can handle the full range of film based lenses without any reduction in field of view.
Many persons want big sensors, at least the size of the 35mm negative format because they assume to get really good image quality. There is however a fundamental conflict between these demands. The only good reason for a big sensor at this current state of the art is the reduction of noise. The pixel area is relatively large and that helps the reduction of noise. The Olympus E-1 is not usable at ISO 400 and higher and even excellent programs like Noise Ninja cannot cope with the massive noise the E-1 generates.
But a big sensor requires extremely bulky lenses to be able to exploit the inherent image potential of the sensors. The current generation of digital sensors allows outstandingly good image quality with a physical size of the sensor that starts with the area of the 13 x 17mm or twice the size of the classical Minox format. The optical designs however need to be upgraded to take full advantage of these characteristics. If we want lenses without vignetting, without distortion and with superior resolution and high contrast, there is only one route to take. This is not a new insight. The classical Zeiss Contarex lenses were quite large and Zeiss defended the size by noting that optical requirements simply demanded a large size. We can see the logic of this position when we look at the size of the micro-lithographic lenses, used for making chip patterns. These huge lenses are indeed aberration free and the size is no issue.
The Leica M lenses are a good example of this principle. The M lenses are relatively small, but do deliver superb performance. This is done by a quite elaborate design and construction, but some compromises have to be struck and some distortion and vignetting will be visible with most M lenses (and R lenses too).
To boost contrast and resolution over the whole image area for the 35mm negative format would imply a significant increase in size. I have commented on this topic on several occasions.
It may be counter intuitive to think according to these lines: a sensor area that has one quarter of the area of the 35mm negative requires lenses that are at least twice as bulky as lenses designed specifically for the 35mm film format.

The obvious counter argument is the Canon EOS-1Ds Mark II with a 24x36mm sensor. Here we have a camera with a sensor size that is identical to that of 35mm negatives. Professionals use this camera with the current lenses from Canon that were designed for film based imagery. If you have a large selection of these lenses and are happy with the results, this is a good interim solution. Extend this argument to the new Mamiya ZD that is equipped with lenses designed for film based performance. And we may add Hasselblad and Phase One backs to argue that classical lenses cope very well with the performance requirements of solid-state image sensors. This line of reasoning is not well founded.

A case in point is the introduction of the new Canon macro lens 2.8/60mm that offers vastly improved performance compared to the older 2.5/50mm macro lens for film based photography. Canon explicitly states that the lens cannot be used on large image areas (be they silver halide or solid state). Canon therefore implies that it is impossible to deliver the equivalent quality when having to cover larger capture areas.
Canon then tells you in essence:
Superior optical performance can be delivered when you restrict yourself to small sensor areas and accept lenses that are quite bulky relative to the sensor size. We are not able to deliver the same performance when we have to cover the full 35mm negative area. The first solution is the route to follow in the future.

Our current mindset is to assume that a larger image area coupled to a less potent lens is a better combo than a smaller area coupled to a superior lens. Many years ago Mr Crawley compared a 6 x 6 Chinese camera with a middle of the road lens to a 35mm camera with a state of the art lens and concluded that size mattered most. This result has been widely published and became a rule of thumb. We see this rule of thumb emerging again and again when persons declare that image sensors of 20 Mp or more are eagerly awaited, as they seem to promise to surpass the resolution watershed of silver halide emulsions. My personal testing (to be published soon) shows that a 5MP sensor with dedicated optics can match the performance of a 35mm negative of superb quality coupled to a lens of equally superb quality.
The dilemma is this: a 22 Mp sensor on a large format chip will require lenses that cover a wide area and that is only possible with huge sized lenses. A 22Mp sensor on a small area (and therefore with individual pixels that are extremely small) will require lenses that resolve impossibly fine detail and these lenses will be huge too! The idea that a larger image area will support intrinsically better image quality and that current lenses designed for this larger size are in a better position to deliver the goods is part of the wrong mindset.
The negative has many advantages and the theoretical superiority in relation to solid state sensors has been explained many times. But the main problem is the transparency of the negative. The image embedded in the emulsion is of stupendous quality, but when enlarging the image scattering and irradiation will reduce the resultant image to a much lower quality. With the digital print this reduction does not happen, It is as if we do a contact print from a large negative.

The technology of digital capture and digital printing has made this rule obsolete. But there is another non-technical argument that defines the current mindset.

Art versus technique: a spurious confrontation!
There is wave of opinions on internet discussion groups that the issue of optical performance and image quality is irrelevant for true photographers who are engaged in making good pictures. They will cite many examples of famous pictures that were made with simple cameras and plain glass. It is true that the artistic value of a picture does not depend on the technical quality of the photograph. On the other hand new techniques have always introduced new possibilities and improved performance has always been a challenge for the user to find ways to harness the power and invent new applications and goals to accomplish. We may return to the paragraph at the beginning of this essay: a BMW motorcycle of 1955 is perfectly adequate for today's traffic demands, but a modern BMW R1150 is light years ahead in driving pleasure and safety and comfort.

Better image quality will deliver finer tonality, more subject details, clear colour hues, more textural and surface definition, stronger delineation of outlines, better definition over a wider field of view, different out of focus representation, and so on. All these characteristics are part of the visual language of photography. No one will deny that a language, richer in grammar, concepts and words, can be more effective in expressiveness and description of fact and emotion.

Enhanced image quality simply enriches the vocabulary and grammar of the visual language: compare Mapplethorpe, Newton, Avedon and Stern with Eugene Smith, Cartier-Bresson and Brassai to understand that image quality is not a goal in itself, but when applied with a full mastery of the medium can deliver new images of great impact and force.
Upshot 1
The technology of image capture and the design of lenses is closely related. A lens designed for imagery with silver halide crystals is not at its best when employed for imagery with solid-state pixels.
Size of the sensor area and size of the individual pixels (grain clumps) are less important in the solid-state sensor than in the silver halide emulsion.
A lens designed for silver halide recording can be relatively small compared to sensor area.
A lens designed for solid-state recording must be bulky in relation to sensor area.
If we want the same image performance with solid-state imagery that we have now with the best of silver halide performance, we need very bulky lens designs.
A large sensor size in solid-state technology allows for a bigger print size, but not for better image quality.
The image embedded in silver halide crystals is bound to follow physical/optical laws.
The image defined in a computer file is governed by mathematical rules only.
In the solid-state world sensor size has less importance as a criterion for image quality than in the silver halide world.
In the solid-state world size of the lens has the utmost importance as a criterion for image quality.
Upshot 2
A large (solid-state) sensor size coupled to a lens system designed for silver halide capture will deliver less image quality than a small solid-state sensor size coupled to bulky lenses designed for optimum performance for that recording technique.
Or in plain words:
A Canon 2.5/50mm macro lens and Techpan 35mm film will bring comparable image quality as a Canon 2.8/60mm and a eight Mp solid-state sensor.
A Canon 2.5/50mm macro lens and a 16 Mp 24x36mm sensor will bring less image quality than a Canon 2.8/60mm and a eight Mp solid-state sensor with an area of 17x 24mm. .