50D review is very flawed (as are all DPReview reviews)...here's why

Started Nov 15, 2008 | Discussions thread
ForumParentFirstPreviousNextNext unread
Flat view
writingsama
New MemberPosts: 10
Like?
50D review is very flawed (as are all DPReview reviews)...here's why
Nov 15, 2008

The 50 D. review is ludicrous!

I'd like to apologize for small grammatical errors such as the above, as I'm writing this with voice recognition software and sometimes it's just not worth correcting it.

I have a number of problems with the 50d review, but before I continue I'd like to first give a little information about my background. I've been a professional software developer for 11 years, and I have a very good grounding in mathematics and imaging. Please don't argue with this post (I know, it's a lot to ask) unless you are sure that you fully understand what I'm saying.

I will also say that this critique mainly addresses the comparisons with the 40D and other cameras in the review, but should, however, apply to most comparisons between most cameras on the site.

Firstly, I'd like to clear something up. The "Nyquist frequency" is not properly understood by the reviewer. This is not exactly high mathematics; information on why (resolution / 2) is not the true limit of detail resolving capability is available on Wikipedia in plain English, but I'd like to put it my own way.

Computers currently use something called sub pixel rendering to render smother fonts on an LCD screen. Basically, a row of your screen looks like this:

...RGBRGBRGBRGB...

With R corresponding to a red pixel, G corresponding to a green pixel, and B corresponding to a blue pixel.

A white pixel is typically one with the red, green, and blue at their brightest settings; a black pixel is one with them at their darkest. So, let's say that their range is from 0 to 9, two typical white pixels would be:

RGB RGB
999 999

And two typical black pixels would be

RGB RGB
000 000

I only separated them for clarity's sake; in an LCD monitor they are packed together one after another.

So, most programs, when they want to draw a series of black and white pixels, will do something like this:

RGB RGB RGB RGB RGB RGB RGB
000 999 000 999 000 999 000

Te reason that this works is because our eyes see the red, green, and blue pixel as one, combining the colors together to create white or black or anything. And if the Nyquist frequency was the true theoretical limit for detail resolution, then this would be the end of this particular facet of the discussion. You would not be able to move the second white pixel over to the left any without joining it with the first, because in this 6-pixel example, the theoretical detail resolution limit based on the Nyquist frequency is three alternating black and white lines. Example:

RGB RGB RGB RGB RGB RGB RGB
000 999 999 000 000 999 000

But you see, we humans are clever.

RGB RGB RGB RGB RGB RGB RGB
000 999 009 990 000 999 000

Did you see what happened there? If you do that, the pixel on the screen appears to move by 1/3 of its actual size, because your eye is still seeing one red, one green, and one blue, surrounded by black. Let's take it a bit further.

RGB RGB RGB RGB RGB RGB RGB
099 909 990 999 099 909 990

I just managed to fit five white lines on a black background, thereby exceeding the Nyquist frequency of 3. Cool how that works, huh? Even cooler is that because digital camera sensors are a regular grid, this effect applies there too.

Of course, with different colors, and different contrasting colors, the amount of leeway above the Nyquist frequency changes; in some cases (especially on digital cameras as I will explain below) there is more leeway (complex 2d color patterns), and in some cases there is much less (let's say, perhaps, one pixel wide red and blue stripes), and in some cases there is none.

On a Bayer filter camera (which is basically all digital cameras for practical purposes except Sigma Foveon based), things get a little bit more complicated, since rows aren't RGBRGB etc... in fact, they're (depending on the type of Bayer filter) a two-dimensional grid with various patterns, which you can see examples of at Wikipedia easily enough.

The point of all this is that the Nyquist frequency is absolutely not the limit of detail resolution capability for a sensor, and that most digital camera reviewers, due to a lack of understanding of the underlying principles, misread resolution test charts.

The second flaw in the review is discussed in much more detail than I can cover here at the following link that I highly recommend: http://www.slrgear.com/articles/focus/focus.htm Basically, in summary, very small errors in focus point (and AF generally introduces very large errors) lead to misleading results, especially for resolution tests; there is a solution, and I don't know if dpreview uses it (and it's not careful manual focusing with 10x magnification in live view mode).

The third, perhaps most glaring flaw, is the choice of lens used to compare the 40D and 50D. The reviewers note that not much "per pixel detail" is gained even though there is a 22% increase in resolution on each axis. Clearly, it is simply because the 40D is already at or near the limit of what that copy of the 50mm f1.4 (a $400 lens) can resolve. Of course if you put a cheap optic on a 10 megapixel camera and a 15 megapixel camera with the same sensor size and don't focus properly, you won't see a difference! That's just plain common sense!

read next reply for pt. 2 (length limit...)

ForumParentFirstPreviousNextNext unread
Flat view
Post (hide subjects)Posted by
pt 2New
pt 3New
ForumParentFirstPreviousNextNext unread
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark post MMy threads
Color scheme? Blue / Yellow