DPReview announced this evening that they have put up test images of these cameras for comparison. They give the obvious reason … there is tremendous interest in these three cameras as they are (a) the three most interesting APS-C cameras out there right now, and (b) in what appears to be a dead heat for image quality.
The DPReview noise tests show that the Canon leads the pack … but that coin test is just one subject, and a fairly high contrast one too. To me, by far the most interesting test is how well a camera retains low-contrast details and 3-dimensional texture at ISOs like 3200. Anything above that is a gift with APS-C, but 3200 is mandatory for me. I already get excellent imagery with the D300, so I expect these cameras to blow me away.
I am going to show a few screen shots from parts of the jpeg images using their interactive tool. These shots will focus on specific areas of the images and I will focus on whether we are seeing good 3-dimensional texture or not.
I will use the current undisputed king of high ISO, the Nikon D3s as the benchmark by which all are judged, just to make it clear what we are looking for. The D3s is pretty close to noise free at 3200 ISO.
Crop 1 is probably the most important one for me. I focus on the yarn with a bit of the hair from the 1000 Italian Lire note behind its left edge. Here is the crop:
Click on that image to get the full sized crop.
- The D3s is perfect. Everything is crisp and clean and the yarn has texture and 3-dimensionality.
- The D60 is pretty poor. You see texture in the hair on the left, but the yarn has only coarse texture. The fine texture is literally all gone.
- The D7000 is pretty stunning on the yarn, coming very close to the D3s for both texture and that feeling of 3-dimensionality. The hair is not as good, being bested slightly by the Canon.
- The K5 is in between the Canon and the Pentax. More texture than the Canon with even a hint of the 3D effect the Nikon gives, but less of the realism of the D700o shot. This is mainly from the fact that the yarn appears flattened, which tends to happen when noise reduction plays havoc with edges and texture. You end up with planes that meet instead of curves that roll into each other.
This crop is an easy victory for the D7000. It’s not a D3s, but it might be able to play one in a movie …
This is the little tuft of fur that sits in the shelf just below and to the right of the mini globe.
- The D3s has stunning detail. Just wow …
- The 3 APS-C cams are in a dead heat here for detail retention. They all look fairly good, but they all create the clumping effect that I call helmet-hair.
- There is one thing I notice though … a very subtle impression of 3D from the D7000 that is not there on the others. Perhaps too subtle, but what I find is that excessive clumping of hair tends to flatten the whole image because it removes tone gradations that are needed to provide the sense of texture in the hair. There is a bit more tone to the Nikon, but only a bit. And on my TN panel, the inaccuracies pile up enough to make it all but invisible. It is visible, though, on my MVA panel.
This one is a tie. I like the D7000 best, but this is a classic toss up.
Crop 3 (remember to click on the image to see the real crops)
This is the globe itself … a favorite image of DPReview’s to distinguish cameras by their pixel level details.
- The D3s shows everything with crystal clarity. Edges are intact and letters are legible.
- The K5 is very clear. Seriously clear. The jpeg engine looks to have the perfect sharpening settings.
- The D7000 is close, in fact it looks a lot like the D3s with the characteristic softer sharpening. But it is less legible than the D3s.
- The 60D should win this one easily since it has the most pixels … 50% more than the D3s. But alas, all thise extra pixels means extra NR as well, and that shows. A general smoothing of the surfaces and a softening of the letters, reducing legibility, especially on the continent itself, where edge contrast with the letters is lower.
So, an easy victory for the K5. The D7000 is close enough that I suspect a click or two on the sharpening setting would equalize. So bully for both Nikon and Pentax so far … these APS-C cams are impressive, considering also that they boosted the pixel count by 33%. Not quite as excited by the Canon, but it is still pretty amazing.
This is one you don’t see shown very often. I like it because it has an extremely difficult section of a step wedge, and a very saturated logo with orange and yellow in it. A really tough test overall. Here, we look more for edge integrity and tonal variations than detail retention.
- The D3s shows the 5 tones on the wedge, as would be expected. Note: if your monitor is not calibrated, please don’t leave a comment telling me that you cannot see the five tones … I already *know* that you cannot. Get a hardware calibrator and a good panel and revel in the joy of clarity :-) … the D3s also have perfect edge integrity on the lettering and a nicely saturated rendering of the Kodak logo. All as expected.
- The D7000 looks great this time. The lettering has been very slightly impacted by noise reduction, but nothing serious. The steps are still visible, despite the presence of more noise than the D3s image. And the logo is pretty decent, if a tad blurred by noise reduction.
- The 60D looks nice as well, although the steps are not visible, having been overwhelmed by chroma noise, which shows up very clearly on my TN panel. The logo, on the other hand, looks great. The yellow is a bit washed out, but the overall effect looks good.
- The K5 fares less well in this test. The step wedge is just mush … overwhelmed by some form of luminance blotching. Coarse noise is really damaging and it shows here. The lettering also shows the effects … badly eaten edges here. Very visible degradation. The logo looks pretty good, although just as blurry as the Nikon from noise reduction.
So … another pretty clear victory for the D7000 overall.
This one is about contrast, sharpening an saturation.
- D3s cannot be used as much of a standard here since it is out of focus because of the narrower depth of field of the full frame sensor.
- The D7000 has the least details, as it appears to have the lowest contrast and sharpening straight from the camera. But all the detail appears intact and the saturation is fine.
- The 60D and K5 both present more detail, in the Canon’s case because of darker reds and what appears to be slightly more contrast. Saturation is characteristically lower than the D7000.
- The K5 has strong contrast and sharpening and some pretty dark reds. Good saturation on the yellow as well.
I think the K5 is the clear winner here with the 60D coming in second. But the D7000 only shows less of everything, which can be boosted in post processing. Quite a different issue from, say, the earlier yarn crop (1), which had non-repairable issues …
Crop 6 (back to texture)
This feather detail is pretty tough to call … my analysis:
- D3s is again near perfect.
- The 60d’s extra pixels finally assert themseves with the veins showing all over the red and the dots. Looks pretty good.
- The D7000 looks like the D3s shot for tone and color (duh) but does not show veins in the dots very well.
- The K5 has a pretty butty looking amount of saturation ion the dots, which makes then look unnatural and wipes out any chance of the veins showing at all in the dots.
I think the 60D takes this one by a nose. The others are pretty nice, although I find the K5 colors a bit funky here.
There is little to differentiate these cams. Each has strengths and weaknesses, but to channel Thom Hogan – if you cannot get superb results from any of these cameras, your technique is the problem. Seriously.
That said … *for my tastes* … the Nikon wins because I hate seeing 3-dimensionality sacrified to noise reduction and the D7000’s ability to retain that level of subtle tones at 3200 ISO is amazing to me … this has me tempted to exchange my D300 for a D7000. I hate buyiong a new L bracket every time, but I think I could live with that.
Disclaimer: I don’t spend all that much time pixel peeping when I look at my own work, but here I was looking for subtle things like edge integrity and texture retention versus noise reduction. I really do find that I prefer looking at images with more of a sense of 3D textures and tones, and that noise reduction is the most damaging culprit.
Your Mileage May Vary …