Saturday, October 31, 2009

F70EXR and F200EXR -- test Large versus Medium size -- which is better?

The short answer is that medium is better ... in *all* circumstances.

Well ... if you shoot bright light all the time and very fine, but very high contrast details ... and you like to pixel peep at 100% ... then shooting HR mode might satisfy you now and again.

But for general purpose shooting without requiring of constant thought or the use of the battery-eating EXR AUTO mode, just shoot P mode with M size.

My test today is a pair of shots of a stack of books I have to file at some point ... I focused on the label and title of the Nelson DeMille book Wild Fire (good thriller by the way.)

I cropped the focus area first, since it is bright and should be quite close between the two ... I found in fact that the labels and text are more readable on the 5mp image that was upsized to 10mp than it was on the 10mp native image.

Those who have read my thoughts on this in the past know that I consider the HR mode to be evil ... it shows weird edge artifacts, no doubt caused be a combination of the rotated SCCD matrix, and the strange pixel-paired EXR arrangement of the colors. Between these two, they are performing some ugly math to get a simple image matrix. With the binned mode, we no longer have the pixel pairs, so we are back to the original SCCD demosaic algorithm, which we know works very well.

Added to the strange bayer mask, we have Fuji's selective noise reduction. At 10mp, the details are very fine, yet there are a lot of weird artifacts that no doubt look like noise to the jpeg engine. So Fuji tries to smooth the flat areas, which inevitably damages edges of letters etc. This is *very* obvious in these images.

Starting with the crops ...


Remember to click on these images to get the larger 800px versions. The crops show just how much cleaner the edges appear. Start with the words "Pix Reg." on the circular label. Perfectly clear on the upsized image, but the x is destroyed by edge artifacts in the 10mp native image.

Next, have a look at the spine of the book "Follow the Stars Home" by Luanne Rice, just below the book by Anne McCaffery and Jodie Lynn Nye (The Death of Sleep.) The title is discernible on the 5mp image while it is simply mush on the 10mp image.

The next crop down is the spine of "Light: Science and Magic", a superb book on artificial lighting, especially of difficult subjects like glass and mirrors. The word Light on the upsized 5mp image is crisp and clear while the 10mp images shows it weakly with a lot of grain. The fine grain has fooled some people to declare that edge definition is much better in 10mp mode ... this, of course, is obviously not true. In some very few cases, the 10mp mode can resolve a bit more high contrast detail, but there is *always* the price of chroma noise and edge artifacts along with smoothing of any low contrast details.

The final crop is the Harry Potter spine ... it shows poor definition on the author's name and it shows a lot of nasty chroma noise on the main part of the spine. The noise on the 5mp mode is mainly luminance grain ... a much more pleasant form of noise.

What I find really interesting is how these issues are subtly visible even at tiny web sizes. Showing the upsized image first:


It looks pretty good. Everything is crisp and clear. The 10mp image looks superficially similar ...


But now take a look at some of the crop areas. Also look at the title of the Stephen King book "The Stand" ... you will see that the title stands out clearly against the red background with the 5mp image, which blending back in (the result I believe of NR and edge artifacts combined) to make it harder to read on the 10mp image ...

I see no reason to *ever* shoot 10mp on the F70EXR or 12mp on the F200EXR ... every time I see a pair of images from these cams in the two sizes, the issues with edges and low contrast detail stick out like a sore thumb.

I realize that many people don't see this ... but train yourself to look for this kind oft thing ... it won't be long before you see the kind of improvements that make a big difference in your image's ability to convey 3 dimensions.

7 comments:

Lili said...

What kills me on FTF is the same folks that lamented the death of the 'magic' 6mp Fuji sensor are the same ones, now that we've cameras that match in hi ISO detail and far exceed the F30/31 while in 5/6mp mode, are the ones that bitch about the EXRs perfomance in HR mode.
"Alas, I cannot use this new sensor at full resolution!"
*insert wails and teeth gnashing*
They're delusional that refuse to accept or understand the whole effing point of the EXR is pixel-binning!
I imagine Fuji engineers reading these posts and screaming.

Lili said...

"Far exceed in DR the F30/31"
posted too quick

Kim Letkeman said...

I quite agree Lily ... the Fuji engineers must find comments like "I never shoot DR mode because HR mode is better" utterly moronic ...

Robert said...

Kim,
Many thanks for your insights here as well as previously elsewhere. I returned my F200 a number of weeks ago in favour of the F72 (Jessops UK badged F70) - the F200 image quality and additional functionality did not convince me as upgrade to my F100. The F72 is a welcome pocket-able addition to the F100/G10/TZ5 and my other cameras.

As I still like the output I get from my F30 (I know some people call it plastic-like smeary but I never print poster-size) so was wondering if the F70 P-Med or P-Large would get closest (I stayed away from EXR mode for this).

After shooting some unscientific test shots (both indoors and outdoors) to initially check the 'print & monitor size' output (rather than 100/200% as I will never ever be able to see this when printed), I felt the output of the larger file size was most pleasing / closer to the F30. But I noticed this was mainly due to a shift in colour balance between the 2 shots. I initially discounted this, but it showed up consistently when doing the test shots, with the images on the Large setting being a hint more warmer. (I kept the same auto ISO and Auto DR - which both resolved back to DR100 for the shots I took).
I can even pick out the 'large' file size from the identical images I shot with Med & Large due this shift.

Have you seen this colour shift / warming as well during you testing?

Could this be due to invoking SW vs HW treatment on the lrg & med sized output in camera??

Would be interested in your views.

Regards
Robert

Robert said...

Well, I should have taken the time to study your examples in more detail. Unless my browser is deceiving me, the whites in the 9.99 price sticker example seems to show the warming in colour, although the book's authors underneath it 'Anne McCaffrey & Jody Lynn Nye' shows this even more clearly...

Kim, is there an explanation for this shift in colour you can think of and would this play in all circumstances?

Thanks
Robert

Kim Letkeman said...

Robert, the color shift is chroma noise, mostly a problem in the blue channel. This is caused by really tiny photosites trying to capture shadow detail. This also contributes to the breakdown of details in shadow at high res as ISO climbs.

Kim Letkeman said...

Robert, there are issues in L size that make it not worth pursuing. If you want warmer color balance, the standard method is to set "shade" as your white balance. I always recommend M size, as that is its best mode for noise. It will get you closest to the quality of the F30's output (exceeds it in my opinion when all things are considered.)