Camera testing bias

Ken Rockwell goes on a tear with his new camera, a medium format digital.

As his habit, Ken Rockwell exhibits a bad case of selection bias. For example, let’s take this quote from the first article:

All the 35mm rangefinders and DSLRs look pretty much the same, and the point-and-shoot is the worst.

I’ve also shown the fallacy of falling for claims of 12-bit, 16-bit or 24-bit image processing in-camera.

As those of us who have done this for a living since the 1980s know, the noise level of any of these sensors is much larger than even 12-bit processing. Throwing more real bits at the ADC only serves to quantize the noise more accurately; there isn’t any meaningful image data needing that precision.

Well anyone can see from his sample the 35mm cameras are not the same: the Nikon D3 exhibits tonality better than the Canon 5D Mk II and the Leica M9, as it should. And those aren’t even the right 35mm cameras to be testing against—I will bet you’ll get nearly the same result as the Mamiya DM33 in the Nikon D3X (with a Zeiss ZF optic on it). He does similar manipulations of outcome bias in order to get the result he is wants to get before hand in his high ISO test.

Just because I shoot a Nikon and Leica doesn’t make me biased against the tests—I’ve been hoping to switch to digital medium format for five years now. Comparing them to a $22,000 camera, however, does not make for a fair fight. (FYI, $22k for a camera like this is a good deal.)

If you are wondering how the Nikon D3 manages so well on this test and the ISO test, it’s because the latest Nikons and Canons have gapless microlenses. (The Leica has a CCD instead of CMOS with a very weak anti-alias filter so can’t achieve as high ISOs.) Furthermore, the Nikon D3 and D3s trade off the pixel pitch of the Canons (resolution and sharpness) for higher ISO and a moderately higher dynamic range. Now by cramming technology and tricks like this, the latest 35mm dSLRs seem to be crushing medium format film cameras in the same way that APS-C dSLRs beat film 35mm. By the way, medium format digital is not really “full frame” either— it’s closer to half frame. In terms of sensors, the D3x and the Sony seem to be as good, or better than, the $60k+ digital medium formats of five years ago.

But things get decidedly unfair (in medium format’s favor) when your engineering hits its heads against the laws of physics—when diffraction effects kick in, for instance. Or when the economics of spending 5x on a manual focus prime simply allows the lens to be machined to tighter tolerances.

Ken hovers around being correct. In this series, it is true the Mamiya is going to best any 35mm in sharpness since it has both more megapixels and a less diffracted lens, and you could do a stopped down test that will be even more enlightening. You could even go all Michael Johnston and start talking about bokeh. But when he’s wrong, he’s so wrong as to be misleading. And his influence is a verbal avalanche that spreads across the camera world until everyone is quoting gospel of bullshit—that’s why so many photographers dislike him. (I’m not too sure where I stand on him. He seems to be both helpful and harmful with the seasons, sort of like a glacier in the Arctic. It’ll take a lot of measurements over a number of years before there can be much consensus.) In the meantime, he does amuse. :-D

By the way, examining the quote above pretty much sums Rockwell to a real scientist: “Throwing more real bits at the ADC only serves to quantize the noise more accurately.” No, Ken, it’s “quantizing noise” more precisely, not accuratelybig difference! Not only that, but the test that would show what he describes would be if you could somehow have an 8-bit DSP, not RAW file, and that doesn’t exist.

Ken always talks precisely, but never accurately, giving laypeople the illusion he knows what he’s talking about, but ultimately leading many to uninformed decisions. He’s the photographic equivalent to a caliper with its screwed sheered off still dutifully reporting three decimal place accuracy, but you have no idea what it is measuring.

6 thoughts on “Camera testing bias”

  1. So, we shoot Canon, and that's probably not going to change. However I'd trade out more megapixels for increased dynamic range any day of the week. What are your thoughts on how high the megapixel count will go before camera companies put an end to counting megapixels as the standard for camera quality?
    My recent post Top 10 Wedding Videographers

    1. I don't know. Megapixel seems to be topping out and I feel the DxO tests have had a huge impact on that. The tests seems to bias less megapixels because of double counting (ISO performance and color fidelity). Currently Canon has opted for the megapixel and Nikon has opted for the low light performance (and dynamic range comes along for the ride).

      For web ready output 2.5 megapixels is "enough", for 8×10's output 6 megapixel should be enough. For double truck's 12 megapixels is "enough" That's mathematically dictated by printer output and our vision systems.

      However, if a scene you are shooting has relatively flat contrast. Or if you are going to be taking an HDR photo and the scene is still (so multiple exposures and time is not of importance). Then I'd take the megapixels over the dynamic range and high ISO also. It's also important to remember the following two caveats:

      1) To double the resolution, you need to quadruple the number of megapixel. The amount the megapixels variance between manufacturers is way to small to mean resolution has a noticeble difference outside pixel peeping.

      2) On the other hand, the increased dynamic range and high ISO performance is also usually not very noticeable until you reach the pixel peeping stage.

      The net result is that if you shoot Canon, be happy. You're not giving as much as you think in order to get the extra megapixels. ;-)

  2. anyone can criticize anything. i wonder, would you be so kind as to give 3-5 specific examples of where Ken is wrong, in your opinion, and if you can? you would, of course, need to be both precise and accurate. only then can we take you seriously.

    1. The article already points out inaccuracies in the statement "quantization of noise" from both a physics and mathematical perspective. Please mouseover the teal blue text for an explanation of what a precise and accurate comparison would entail.

      A previous article I wrote explains how the most common noise is affected by digitization. The general gist of it is it is correct that in the highlights region, noise levels will span multiple digitization bins and thus, for the most part, are "wasted." But in the shadows, this is not the case. In fact, many cameras actually save lossy "compressed" RAW files that do a mapping in order to contain this that throw out data (which would be under shot noise) in the highlights but preserve 100% of the data in the low bits. This may turn a 16-bit raw image down to a 12-bit (Nikon D70) or 8-bit (Leica M8) image! But comparing these, not seeing noise, and stating there is a "fallacy" in claims of high bit processing just because the final file is less than 8-bits is improper, since these systems work manipulate the image in the high bit space (16-bit or larger) before downsampling the image down to the final file. (The same mistake follows why for years Ken Rockwell incorrectly recommended slide scanning over digital cameras less than hundreds of megapixel—slide scanners scan in very large bit depth… and recommended that RAW was a "waste" over JPEG—incorrectly understanding that JPEG compression introduces both artifacts and bakes in stuff in a manner that is not as lossless as a lossy RAW compression with respect to noise).

      As for selection bias, I point out specifically in the mouseover how a previous test used totally different criteria to hype Leica lenses and razz Nikon camera bodies.

      I've already, in previous articles, pointed to many inaccuracies he has had (esp. concerning on how "film has higher resolution than digital" and "JPEG vs. RAW" in his old articles).

  3. You seem to misunderstand Ken Rockwell's purpose. It is not to inform his readers, but to sell camera gear. I was privy to the affiliate marketing stats of one of the big NY camera retailers, and Rockwell accounted for an ungodly percentage of these. I wouldn't be surprised if he makes well over $20,000 a month in affiliate commissions. For all practical purposes, his site is his main career. That's why his prose makes outlandish claims – to be memorable. Each person who debunks him only increases his notoriety, which is exactly what he wants. As for the recent change of heart with Leica, I guess he figured out the increased commissions from expensive Leica gear outweigh the risk of losing sales by alienating people who own brands he disses.
    My recent post Organizing with Delicious Library

    1. Yes, I gathered that. The fact that he hasn't taken a photo professionally (to my knowledge) in the last 5 years (at least) probably emphasizes this. His site has an insane amount of google juice, and that is unfortunate.

      I don't see the problem in debunking him. But I guess I should not link him.

      What I find amusing is how some people feel the need to defend him. ;-)

      As for Leica (and Nikon). His site makes me wish he was a Canon shooter. He does every Nikon owner a disservice.

Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>