40d better than expected

[Misirlou]

Since I have broadband in my present circumstances, I also downloaded Misirlou's comparison images. I sort of half expected there would be a difference in exposure due to a different ISO accuracy between the two cameras, but I see that Misirlou has compensated for that. The main difference is that the 40D image comes out slightly more contrasty with the same ACR settings. At least the shadows are slightly darker.

The surprise to me is not that the 40D shows greater detail. There is slightly greater accutance in the 40D image but there's also something odd going on considering these shots were taken with the same lens. There's a noticeably greater difference in resolution at the far left edge than there is in the centre.

In fact if the difference in resolution at the far left edge extended across the entire image, I would consider that a worthwhile increase in resolution for the 40D that could not be compensated by any amount of sharpening.

Could it be the focussing in the 2 shots is slightly different and if so, why should it affect edge resolution more than centre resolution? Could it be Misirlou's 20D sensor just happens to be slightly under par in its performance on the left side, or could it be that the 40D has improved performance at the edge of the sensor which no-one but a pixel-peeper would notice?

The two sets of crops below compare upressed 20D images which are unsharpened, although both images had initial default sharpening in ACR.

In the centre crop I used a bit of gamma correction on the 40D image to lighten it slightly and used a bit of local contrast enhancement on the 20D image to equalise the tonality of the two images as far as possible.

[attachment=3800:attachment]  [attachment=3801:attachment]
[a href=\"index.php?act=findpost&pid=152062\"][{POST_SNAPBACK}][/a]

My own guess is that the cameras didn't focus at exactly the same point. I set the central autofocus zone on both cameras to read at the very contrasty border between two patches on the McBeth chart. I bet neither camera picked exactly the same distance. The 40D central focus sensor is noticeably more sensitive than the one in the 20D.

Manual focus might have made for a more consistent test, but I find the 20D to be a real bugger to manually focus in dim light, though the 40D is a little bit better. The 40D has Live View, which should make manual focus a lot more precise when tripod mounted. Oh, wait a minute. Serious photgraphers are not supposed to stoop to such pedestrian frivolities as Live View. Forget I mentioned it.

Now that I think about it, the 50 f/1.4 is probably not at its best at such a close distance. It might be worth repeating on an outdoor scene from a greater distance.

On the other hand, the 40D sensor supposedly has a much improved set of microlenses. It might be possible that the image really is a lot better at the edges simply because of the sensor improvement.

And to get back to my original point, 40D images are superior to 20D images, even without stooping to extreme pixel peeping. At least with my copy of each camera.

[Ray]

And to get back to my original point, 40D images are superior to 20D images, even without stooping to extreme pixel peeping. At least with my copy of each camera.
[a href=\"index.php?act=findpost&pid=152306\"][{POST_SNAPBACK}][/a]

Pity you've not been able to demonstrate it with those sample RAW images you've taken the trouble to post   .

The clear resolution advantage of the 40D at the edges of the frame in the ISO 100 shot is a bit of a mystery. Since I'd downloaded all your RAW images initially but hadn't taken the time to examine the other ISOs, I decided I'd check out if this noticeable softness at the edges of the 20D shot at ISO 100 is apparent in the other shots.

Surprise! Surprise! It isn't.

In fact, at higher ISOs the resolution at the edges is even more closely matched than it is at the centre. This could be explained by the fact that 35mm lenses, even on cropped format cameras, are not as sharp at the edges as they are in the centre. Any subtle differences in resolution between the 2 sensors will be more apparent with a sharper lens and almost all lenses are sharper in the centre.

I believe your ISO 100 20D shot is therefore an anomaly and should not be used for resolution comparisons.

In the comparison crops below, at 100% and 200%, I interpolated the 20D shots to the same file size as the 40D shots but did not apply any additional sharpening. Both images have had only the ACR default sharpening of 25 applied during conversion.

I also compared shadow noise in the darkest part of the images by greatly lightening both images with 'levels'. As you can see, both images are as close as matters regarding shadow noise. Conversion settings were: zero shadows and zero contrast with everything else at default.

Now I know that you didn't apply the in-camera noise reduction for theses 40D shots. If you had, then shadow noise would have been marginally improved. However, I would suggest that such improvement is negated by the fact that the 20D has a slightly higher ISO range than the 40D.

You are probably aware that Canon DSLRs have in the past typically understated the ISO values, but this has changed with the 40D. The true ISO range of the 20D is ISO 125 to ISO 4000, compared with actual ISO 100 to 3200 for the 40D. This is reflected in the consistently longer exposure values in all the 40D shots which you've had to apply to avoid underexposure.

Now I'm going to make what I think is a very valid point. If one is going to pixel peep, then do it properly which means taking all pixel-peeping factors into account.

If the 40D has marginally less shadow noise, when noise reduction is turned on, it's a consequence of a slightly longer exposure (more photons impinging upon the sensor). A slightly longer exposure is likely to contribute to greater camera shake and consequently slightly less sharp images which in some circumstances will also negate the slight resolution advantage of the 40D.

For example, if getting clean, sharp images is the goal, then the purpose of increasing the ISO setting is to get either a hand-holdable shutter speed to reduce camera shake, or an adequate shutter speed to freeze subject movement.

At whatever ISO setting you choose, the 40D is at a slight disadvantage here, compared with the 20D. The only circumstances I can think of where it's not at a disadvantage is the taking of stationary subjects on a tripod, as in your RAW samples.

If I need ISO 800 on the 40D, for example, in order to get a hand-holdable shutter speed of 1/25th, I can use 1/30th on the 20D at ISO 800 and the same aperture.

Now the differences in sharpness that on average would result from these differences in shutter speed (and it would have to be an average due to the variable nature of camera shake) would probably be of pixel-peeping proportions, just as the increase in resolution due to the extra 2 megapixels of the 40D is of pixel-peeping proportions. One negates the other, so on balance it ceases to be an issue or an advantage for the 40D, in most circumstances.

In conclusion, I can only repeat that any 'objective' image quality improvements of the 40D are only discernible at the pixel-peeping level, assuming the focussing is not an issue as it appears to be in your ISO 100 shot with the 20D.

Nevertheless, accurate focussing can be very important when using a shallow DoF. A more revealing comparison between these two cameras would be a series of images taken with a high quality lens, say the 85/1.2 at f2.8, focussing on the eyelashes of a model or the pet cat. If the 40D produces consistently sharper images at the point of focus due to more accurate and/or faster focussing, then that's a very worthwhile improvement.

EDGE COMPARISONS

[attachment=3900:attachment]  [attachment=3901:attachment]

CENTRE COMPARISONS

[attachment=3902:attachment]  [attachment=3903:attachment]

SHADOW NOISE

[attachment=3904:attachment]  [attachment=3905:attachment]

[Misirlou]

Pity you've not been able to demonstrate it with those sample RAW images you've taken the trouble to post   .

In the comparison crops below, at 100% and 200%, I interpolated the 20D shots to the same file size as the 40D shots.

Now I'm going to make what I think is a very valid point. If one is going to pixel peep, then do it properly which means taking all pixel-peeping factors into account.

In conclusion, I can only repeat that any 'objective' image quality improvements of the 40D are only discernible at the pixel-peeping level, assuming the focussing is not an issue as it appears to be in your ISO 100 shot with the 20D

Ok, I'm not sure I understand these apparently contradictory statements, but then I've admitted previously that I'm none too experienced with pixel peeping in the first place. Sorry.

You interpolated the 20D files to the same size as the 40D shots. Why? One can do all sorts of things during interpolation that have an impact on sharpness. My contention is that both cameras produce pretty similar results on a pixel by pixel basis, but a 40D has more of them. If we're going to start interpolating, why aren't we looking at 40D images next to 12MP cameras or something?

If the autofocus in a 20D is more likely to result in innacurately focused pictures, isn't that's a factor in image quality as well? What would you do with a 20D that would mitigate that? Manually focus? At least you can use Live View to manually focus a 40D with some degree of precision.

Why are we interested in testing either a 20D or a 40D with extremely expensive lenses? No 20D is worth very much now, and new 40Ds cost a good bit less than almost all L lenses. Presumably, anyone with a big collection of L lenses wouldn't be very interested in either camera. That's why I tested them with a 50 1.4.

One thing I've noticed recently with the 40D is no patterned structure in the noise in deep shadows like I used to get with the 20D. I'm probably more concerned with noise in most cases than I need to be, given what my prints look like without any extreme measures taken to reduce it. But the 20D nevertheless annoyed me with the banding in deep shadows at high ISO. I've yet to see any discernable banding with the 40D, even without High ISO Noise Reduction turned on.

I've also been favorably impressed with the Highlight Tone Priority feature, although it doesn't work at 100 ISO, and I still don't really understand how it's mechanized in the camera.

I'm afraid I won't be able to demonstrate any of that to you because my 20D is going to live with my mother tomorrow, and will no longer be available for testing.

And I will repeat, if you don't think the 40D produces better images, don't buy one. Nobody's asking you to. The non-sensor related improvements are really where the 40D shines anyway.

[chickenhawk212]

Go easy on each other guys!  

I just got back (to the east coast) from two weeks shooting in a bunch of national parks in california and i nearly died of happiness shooting with the 40d! Deep buffer, better viewfinder, improved focusing, all so great.  Everything from wildlife to landscape, shore birds to surfing girlfriends.  The camera really just made shooting extra fun and easy.  I shot with my rebel xt too, usually with a 10-22 mounted and was less happy with it than usual when comparing it to the 40d for usability.  i dont know about the pixel peeping but i definitly have more keepers from the 40d.  Probably not a fair comparisson but might go to show that all the nice upgrades that canon keeps sending us are worth the money, regardless of IQ...which i still think is better, but maybe only for the sake of continuing the argument  

Happy shooting all!

[Ray]

You interpolated the 20D files to the same size as the 40D shots. Why? One can do all sorts of things during interpolation that have an impact on sharpness. My contention is that both cameras produce pretty similar results on a pixel by pixel basis, but a 40D has more of them. If we're going to start interpolating, why aren't we looking at 40D images next to 12MP cameras or something?
[a href=\"index.php?act=findpost&pid=154276\"][{POST_SNAPBACK}][/a]

The best way to compare 2 images is to have them the same size on the screen, otherwise the smaller image will tend to appear sharper even though it's not.

Standard bicubic interpolation actually produces a slight softening effect and usually requires some compensating sharpening, which I did not, however, apply in the above examples.

If one is comparing images at high magnification in true pixel-peeping fashion, it does not help to have a hint of jaggies in the smaller image at 200%. This is also not how one would print images. Interpolating the smaller image to the same file size as the larger image seems to me the most practical and realistic approach. One could go one step further and print out the crops, in which case one would not want to use less than 240ppi for either image and again the images would have to be the same size for proper comparison.

If the autofocus in a 20D is more likely to result in innacurately focused pictures, isn't that's a factor in image quality as well? What would you do with a 20D that would mitigate that? Manually focus? At least you can use Live View to manually focus a 40D with some degree of precision.

If, then yes. But autofocus capability was not part of the test. We don't know if it's an issue which is why I suggested testing with a very sharp lens at wide aperture. Accurate focussing is generally only an issue under certain circumstance such as shallow DoF, poor lighting or a lack of image contrast and detail at the point of focus. If the 40D sometimes outperforms the 20D in this respect, then one could claim improved image quality but only sometimes. This is not quite the same as a blanket statement that 40D image quality is superior to 20D image quality above the pixel-peeping level.

One thing I've noticed recently with the 40D is no patterned structure in the noise in deep shadows like I used to get with the 20D. I'm probably more concerned with noise in most cases than I need to be, given what my prints look like without any extreme measures taken to reduce it. But the 20D nevertheless annoyed me with the banding in deep shadows at high ISO. I've yet to see any discernable banding with the 40D, even without High ISO Noise Reduction turned on.

This sounds like a camera quality control issue. It was what I noticed with my first copy of the 5D; an irregular pattern like a coarse burlap weave in the deepest shadows, usually a sort of burgundy red. I'd never seen such ugly patterns and banding in either my D60 and 20D. I returned the camera to the store and was given another copy. After testing the new copy, I was convinced there was an improvement, but it's still there and I believe it is a characteristic (if not a problem) of both the 1Ds and 5D. There was much discussion of this issue on the old Rob Galbraith forum at the time.

I've also been favorably impressed with the Highlight Tone Priority feature, although it doesn't work at 100 ISO, and I still don't really understand how it's mechanized in the camera.

I've read that shadow noise/detail suffers slightly through the use of that feature. I think there's probably no substitute for proper exposure to the right.

And I will repeat, if you don't think the 40D produces better images, don't buy one. Nobody's asking you to. The non-sensor related improvements are really where the 40D shines anyway.

Well of course. That's a given. I've never claimed there are no worthwhile improvements in the 40D. There obviously are. It's just that it appears to be the case that 'across-the-board' image quality improvement, in respect of both resolution and noise, is very much at the pixel-peeing level. This is a point that I think needs to be stressed.

If there are people out there who are a little disatisfied with the resolution and noise of their 20D or 30D, who perhaps have just bought an A2 printer or Epson 24" wide 7800, they might be disappointed if they bought a 40D partly on the basis that they would get noticeably sharper and more detailed prints with less shadow noise.

This is the matter I'm just trying to clear up. I didn't intend to spoil your fun  .

[Misirlou]

Well, I guess we'll have to drop the discussion then, or at least I will. We could go around this ring endlessly.

For example, the only way to test autofocus accuracy is to factor in probability. Any system that is subject to random error of any kind has to have its accuracy stated with a defined confidence interval. That would require many, many shots, and I'm not interested.

I'd rather just go out and shoot real pictures. In fact, I still have a couple of nice Rolleiflex 2.8s. One is a Planar and one is a Xenotar. Maybe I'll go compare them.

[Ray]

Well, I guess we'll have to drop the discussion then, or at least I will. We could go around this ring endlessly.

For example, the only way to test autofocus accuracy is to factor in probability. Any system that is subject to random error of any kind has to have its accuracy stated with a defined confidence interval. That would require many, many shots, and I'm not interested.

I'd rather just go out and shoot real pictures. In fact, I still have a couple of nice Rolleiflex 2.8s. One is a Planar and one is a Xenotar. Maybe I'll go compare them.
[a href=\"index.php?act=findpost&pid=154320\"][{POST_SNAPBACK}][/a]

It's not really a matter of going around the ring endlessly but correctly identifying the causes of any perceived improvement in 40D image quality. Doing so is of practical benefit to your skills as a photographer. If 40D images are superior because of more accurate focussing, don't you want to know that?

All the test images I've seen from competent testers, including your own shots, indicate the 14 bit A/D advancement and the extra 2 megapixels of the 40D sensor have a relatively small impact on over-all image quality. In fact, my guess is it was necessary for Canon to move up to 14 bit in order to maintain noise at the 20D level and implement a noise reduction system which doesn't appear to reduce resolution.

[Panopeeper]

I've also been favorably impressed with the Highlight Tone Priority feature

That surprizes me, because it does nothing useful to the raw data.

The effect of HTP is, that the selected ISO will be lowered by one stop (that's the reason it does not work with ISO lower than 200). However, this is not noted in the Exif data, which shows the selected ISO.

The advantage of HTP with JPEG images is, that it prevents overexposure (or at least reduces the chance) and corrects the brightness by boosting the exposure with 1 EV, except in the very highlights.

The advantage for those recording raw data is, that the brighness of the in-camera display is as high as it would be with the selected ISO, and the two embedded images are boosted by 1 EV with "highlight protection".

As the in-camera histogram is based on the embedded JPEG image, it reflects the corrected exposure.

The disadvantage (a big one IMO) of HTP is, that one can't judge the true exposure because of the manipulated histograms. However, this is meaningful only if the settings make the in-camera histograms appear somewhat like the true raw histograms, which has its own disadvantages.