With a total of three "low priced" (well, everything is relative, I guess) SLR digicams on the market or announced now (late August, 2000), there's naturally been a huge amount of interest in how their image quality compares. As of this writing, we'd had only a very short time with the Canon EOS D30, so our set of standardized test images isn't yet complete. We do have enough collected though, to make a meaningful comparison between the Nikon D1, Fuji FinePix S1 Pro, and Canon EOS D30 digital SLRs. We called this comparison the "Three Titans" to reflect the impact these three products will have on the field of professional photography over the next year.
In the examples below (as in all our testing) we've tried to keep things as absolutely equal between the three cameras involved as possible. Normally, we present all images exactly as they come from the cameras, but in this case wanted to avoid any possible perceptual biases from slight differences in sizes, and so scaled some of the images to produce a consistent sizing across all three cameras. This resizing was never more than 10% up or down for any imabe, but we wanted our readers to be aware of it, as it would at least slightly affect noise structure in those images for which we've provided "full sized" samples. Really exacting reader comparisons of that sort will have to wait for us to get our test images from the S1 and D30 into the Comparometer(tm). In the meantime though, these samples should be very illuminating to those considering purchasing one of these cameras.
You'll note in the color samples here that there are two different images from the Nikon D1: One labeled "Original", and one labeled "NTSC". There's been a lot of discussion on the 'net about the D1's color handling lately. A variety of fixes have been proposed to deal with various slices of the color spectrum. One of the more intriguing fixes we've heard about is to use a color space conversion in Photoshop(tm) 5.5. This particular theory holds that the D1's color space is somehow tied to NTSC, normally a television standard. We're not sure of any underlying truth to this theory, but just a little playing around with some of our test images revealed that the impact was pretty dramatic, and mostly seemed to be in the right direction. (We're avoiding the whole issue here of third-party products designed to fix the D1's color, such as Mike Chaney's Qimage Pro, or Eric Hyman's Bibble. These products work well, but we weren't looking in this example to produce a review of color-correction software, just give a better idea of what the D1 could do with some minor tweaking.) The specific process we used here was to set Photoshop 5.5's color management to convert the D1's images as if they were coming from an NTSC color space into a default sRGB space with a gamma of 2.2. We then boosted the gamma by 1.1 in Photoshop's "Levels" control to brighten the image slightly.
Well, enough of the preliminaries: Here are samples from the three cameras, cropped from our standard "Davebox" test. The uncorrected D1 did have some trouble with the bright primaries, substantially corrected by the NTSC trick. By contrast, the S1 and D30 are quite accurate "out of the box." The S1 seems to like reds quite a bit, greens less so. The D30 looks very accurate across the board, but shows some weakness in the bright yellow, and a minor weakness in the bright red. Overall, the Fuji does the best job uncorrected, with the D30 running a close second.
(NOTE - VERY large image, 2.3MB, 5000 pixels wide: Perhaps best to download to disk and view with imaging software)
In this sample, we wanted to look at the behavior of the three camras shooting Caucasian flesh tones under difficult lighting. (Caucasian skin is tough, not only because it's a "memory" color that people know what it should look like, but also because the pastel colors make them very sensitive to minor changes in the balance of the secondary and tertiary "contaminant" colors. Cross-channel contamination in the color space thus has unusually severe consequences here.)
This is an area that's seen the D1 roundly criticized for, and with some justification, judging from the unadjusted image on the left. Two things seem to happen to flesh tones with the D1, particularly under contrasty lighting like this: First, they tend to have an odd mixture of magenta and cyan in them, giving a somewhat sickly appearance. The second effect we've noticed that seems to occur only in strong lighting as here is that the highlights in the skin tones appear to shift markedly toward the yellow. Both of these effects were significantly alleviated by the NTSC trick.
The Fuji S1 and Canon D30 both turned in beautiful performances here, but with some notable differences. The S1 shows very bright, attractive color that is our personal first choice. We have to say though, that this bright color really isn't a completely accurate rendition of the original scene. In this test at least, the D30 probably turns in the most accurate performance overall.
Detail and resolution are of course critical, and all three cameras here have close to the same pixel resolution at the sensor level. We have some samples further down the page showing the performance on a laboratory test target, but always feel that "natural" objects like this picture of a house with foliage in the background often give a better "gut" sense of how well a camera handles detail. These crops are taken from our standard "House2" poster test. The original subject here is an ultra high-resolution poster, scanned as a 500 MB file from a tack-sharp 4x5 transparency, printed at 40x50 inches on a 600dpi laser-photographic printer. This allows us to test cameras with exactly the same subject, regardless of time of year, weather, etc. (This is a good test for detail, not so good for dynamic range (light to dark) because the poster doesn't have nearly the range of the original scene.)
In this test, the Fuji looks the sharpest, but we'd criticize it a bit for being just a little over sharpened, with some of the finest detail in the branches and pine needles lost. (Not much, but enough to comment on.) The D30 on the other hand looks rather soft, and the D1 is somewhere in between.
In this sample, we tried to level the playing field a little, by applying unsharp masking in Photoshop, striving for the best sharpness we could achieve separately for each image, without introducing unwanted artifacts. With this treatment, we feel the D30 actually now comes out noticeably ahead, with exceptional sharpness and no artifacts or artifical coarsening of the finest details.
ISO-12233 Resolution Test Chart
We suffered a lighting problem just as we were getting to the resolution target with the D30, but wanted to post what we could. Our full-target shots were poorly lit to the point that we don't want to post them in their entirety. Across the area of the individual resolution elements, the images were lit evenly enough to use here though. These images are all just as they came from the cameras, with the exception that we adjusted the tonal balance using Photoshop's "Levels" control, to render all at exactly the same brightness and contrast.
The results are interesting: All three cameras seem to have close to the same resolving power, with minor differences in the degree of aliasing they present. (An important note though: The S1 shot here was taken at the camera's "medium" non-interpolated resolution: The 6.1 megapixel interpolated file size does in fact produce a slightly higher resolution with slightly reduced artifacts. The differences are minor though.) Of the three cameras, we'd have to say that the D30 delivers the most usable detail in this test, thanks to the exceptional smoothness of the target lines, and the absence of "jaggies" from the aliasing. (The D30 does show a bit more color aliasing at very high frequencies though.)
This is the "other half" of the resolution test, measuring resolution in the vertical direction. Here, the results favor the D30 even more strongly, although in fairness the tone adjustment we did to make the images look the same visually emphasized the "jaggie" aliasing in the S1 and D1 samples. Bottom line, the three cameras are pretty close in resolution. Interestingly, the D30 seems to win the race for the most resolvable detail, even though it looks "softer" to the eye.
High-ISO Noise Levels
Everyone seems to want to know about these cameras' performance at high ISO ratings. The swatches above are cropped 1:1 from the medium gray swatch on the MacBeth chart in the Davebox target. The image from each camera was taken from the brightest sample of our Low Light test series, a light level of 8 foot-candles (about 88 lux). This is a moderately dim lighting condition, about a full f-stop below a typical residential interior at night. In this test, the D1 and S1 seem about equal in their noise levels, and the D30 shows about the same noise level in the green channel, but noticeably more in the red and blue. The one cauation we have here is that, while we were told that the D30 unit we were testing was "really equivalent" to what the production models would do, it nonetheless was an "initial production" model, so there's always the possibility that we'll see better results on the final production models. (The D1 and S1 were both full-production units.)