Canon EOS 350D Digital RebelCanon makes an impressive update to their wildly popular "Digital Rebel."!
<<Reference: Datasheet :(Previous) | (Next): EOS 350D Sample Images>>
EOS 350D IMATEST ResultsReview First Posted: 3/23/2003, updated: 6/4/2005
Canon Digital Rebel XT Preliminary Image Analysis
By Dave Etchells
Canon Digital Rebel XT Hands On Preview
Canon Digital Rebel XT Photo Gallery
Thumbnail index of Rebel XT Test Images
Canon Digital Rebel XT Preliminary Image Analysis (this page)
Rebel XT timing analysis
Canon Digital Rebel forum (Original and XT models)
We'll have our usual exhaustive analysis of the Canon Digital Rebel XT's test images in another week or two, and hopefully our FULL standard review of all the camera's features and functions in a similar time frame. Meanwhile though, here's what I've found to date, examining shots we've taken with the Rebel XT thus far:
Resolution and Sharpening
From the outset, the Canon Digital Rebel XT's resolution was impressive, but in reality it's not all that big a step up from that of the original 6-megapixel Rebel. If you look at the actual difference in the number of pixels in each linear dimension (3,456 x 2,304 vs 3,072 x 2,048), there are only 12.5% more pixels in each direction. Definitely not that much more resolving power, when you come down to it. In closely examining test images shot with both cameras under controlled conditions though, I found that the Rebel XT's default in-camera sharpening algorithm uses a smaller "kernel" in its processing, producing a tighter sharpening effect. The overall result is a somewhat greater difference in perceived resolution than the raw pixel count would indicate, but the difference between the two cameras is nonetheless slight. To my mind and eye, the XT's higher resolution sensor really isn't a justification for upgrading, if you already own one of the original Rebels.
Looking at prints made on the Canon i9900 printer we use as a sort of "reference standard" here, there was no question that the Rebel XT's prints produced great-looking 13x19 output. That said, and even with its slightly tightened in-camera sharpening, I found that added unsharp masking in Adobe Photoshop(tm) made a very noticeable difference, without introducing undesirable artifacts. The two crops below show an example of an image directly as it came from the camera, and with unsharp masking applied, at 130% and a radius of 0.3 pixels. This was the amount that seemed optimum for images shot with the camera's default sharpness setting. Slightly improved results could be obtained by setting the in-camera sharpening control to its lowest level, and using a much higher percentage in Photoshop's unsharp masking operator. (The example images were shot with a Canon 100mm f/2.8 EF macro lens at f/8, a very sharp and distortion-free prime lens.) Bottom line, the Canon Digital Rebel XT captures a tremendous amount of detail very cleanly, and its internal sharpening algorithm delivers a good compromise between sharpness and artifacts, slightly bettering that of the original Rebel.
Color is obviously a key factor in digital cameras, and for the most part, I prefer to let readers arrive at their own conclusions, based on test shots and "gallery" photos we take. It's safe to say though, that the Canon Digital Rebel XT has excellent color rendition. Like most "prosumer" digital cameras, it does over-saturate some colors a bit, particularly those that are highly saturated in the subjects in the first place. The amount of its over-saturation is slightly less than that of the original Rebel though, and the Rebel XT also offers slightly greater hue accuracy as well. Most Canon cameras that I test tend to shift cyans slightly toward richer blues, a tendency that I suspect is a deliberate choice made to enhance sky colors. (Rich blues are also typically shifted slightly towards purples, which is a bit more problematic in my experience.) In the case of the Rebel XT, this cyan-to-blue shift is noticeably reduced, as is the blue-to-purple shift other colors are likewise somewhat more accurate as well.
This can be seen in the illustration below, in which color-error plots (from Imatest) for the Canon Digital Rebel XT and the original Rebel can be compared by mousing-over the image. This does a pretty good job of graphically showing the differences between the two models' color handling.
(Mention very little loss of saturation at high ISO)
In this plot, zero color saturation is at the center of the figure, with progressively higher saturation as you move out radially. Hue varies as you travel in a circle around the center. The "correct" colors are represented by the small squares, while the actual colors captured by the camera are shown as circles. A camera with theoretically perfect color rendition would have the circles laying on top of the squares. Most digital cameras (consumer models more so than professional ones) boost color saturation in some areas of the spectrum, which would show up in the plot as the circles falling outside the corresponding squares, on an exact radial from the center of the graph. As you can see, the circles for the blues in the original Rebel are rotated slightly counter-clockwise from their ideal positions, while those of the Rebel XT are much closer to their proper locations. Likewise, many the circles for the XT's as-captured colors aren't stretched quite as far outward as are those of the original model. The resulting color will be technically more correct, but may not appeal to some people as much, if their personal preference is for the brighter, more saturated color of most consumer digital cameras. (Fortunately, the XT has a color saturation adjustment that permits tweaking of its color in fairly fine steps. Users preferring a more saturated look may want to just routinely shoot with the color saturation control bumped up a notch.)
Ultimately, plots like this only help us to understand what's going on in a camera's color management: I strongly encourage you to ignore numbers like "mean delta-E error" and "mean camera saturation" when comparing digital cameras, and instead look at images shot with them to see how they appeal to your own personal aesthetics. Color is a very subjective experience, one that translates very poorly into graphs, charts, and numbers. -- But charts like this do in fact make it easier to grasp the generalities of what each camera is doing.
While I didn't bother to show it here, one characteristic of the Canon Rebel XT's color handling that deserves some comment is how it behaves at high ISOs. Most cameras cut color saturation at high ISOs, as this has the natural consequence of also cutting image noise. Oddly, the original Rebel boosted its saturation at ISO 1600, but the Rebel XT leaves color saturation almost entirely alone. While its default saturation is a bit lower than that of some competing models, it's laudable that the saturation doesn't deteriorate any at its maximum ISO.
Since we're talking about high ISOs, this is probably a natural place to look at the image noise characteristics of the Canon Digital Rebel XT. Consistent with what we've come to expect from the entire line of d-SLRs using Canon's advanced CMOS sensor technology, images from the Rebel XT are very clean, with low noise levels and generally attractive "grain" pattern. Subject to my usual disclaimers that such graphs are ultimately fairly meaningless, here's a plot of how its noise-vs-ISO curve stacks up against some of the competition:
As you can see, the absolute magnitude of its noise levels are very close to those of the original Rebel at all but the very highest ISO levels. As usual though, noise-magnitude plots of this sort actually tell us precious little about what the noise actually looks like. - I show them because readers continue to ask for them, but I'm honestly considering dropping them entirely because they lead to such absurd camera-bashing in online forums between zealots for one brand of cameras or another. There's also the issue of how much detail the camera trades away in pursuit of low noise levels, something that such plots (or swatches cropped from the flat tints of a MacBeth chart) do nothing to reveal.
With that in mind, let's look at some crops from the same pine tree subject shown above, showing what happens as we go from ISO 100 to 1600:
What's interesting here is how well the image holds together at ISO 1600, and how well the camera continues to resolve detail in the low-contrast areas where the pine needles overlay each other. There's clearly some loss of sharpness at ISO 1600, but compared to what I'm accustomed to seeing from digital SLRs at that ISO level, it's really quite minimal.
It's also important to note again the difference between viewing images pixel-for-pixel onscreen, versus looking at prints. Looking at prints from the Rebel XT made on the aforementioned Canon i9900 studio inkjet, I found that the ISO 1600 shots held up surprisingly well, even when printed at 13x19. At 8x10 print sizes, the ISO 1600 results were almost indistinguishable from those at ISO 100, apart from a slight shift in the sky color toward a lighter, more purplish hue. (You can see this in the crops above, but it was much more apparent when the images were printed full-page.) -Bottom line, if you're concerned about high-ISO noise in any digital camera you're considering, I strongly encourage you to download our test images and print them out on your own printer, to see for yourself whether the image quality is sufficient. - Don't fall into the trap of judging everything by what you see on your CRT, viewed 1:1.
After our recent preview of the Nikon D2X, a number of readers wrote in asking me to show real-world high-ISO results under incandescent lighting. They raised an important point, in that image noise in digital cameras is significantly higher in the blue channel than in either red or green, and when compensating for the heavy yellow color cast of incandescent lighting, the blue channel is the one that has to be boosted the most, thereby amplifying its noise levels. The net result is that some cameras may do quite well at high ISO in daylight-balanced lighting, but much worse under household incandescent.
To shed some light (no pun intended) on the Rebel XT's high-ISO performance under incandescent lighting, here are some crops from my "Indoor Portrait" test, showing results from ISO 100 to 1600.
As we might expect from the preceding discussion, the XT's ISO 1600 image noise is quite a bit more apparent in this shot than the outdoor shot of the pines, but it's still impressively low. Printed at 13x19, the noise here is quite prominent, at 8x10 it's visible but probably acceptable to most users, and at 5x7 almost entirely ceases to be an issue. (The print at 5x7 showed a slight softness in Marti's hair, but if I didn't have the ISO 100 example to compare it to, I probably wouldn't have noticed.)
Since people will naturally be interested in comparing the performance of the new XT against that of the original Digital Rebel, here are crops from high-ISO images shot with both cameras under incandescent lighting. While they were shot at two different times, in two different settings, the color balance of the lighting in these two photos should be almost exactly identical, as the same type and grade of light bulbs were used to illuminate both scenes, and I've found from testing multiple brands of household light bulbs that bulbs of the same wattage and lifetime rating are almost always within 50K of each other.
Looking at the crops above, you can see why I emphasized the similarity in the lighting between the two scenes, given how differently the two cameras rendered it. I believe that this is a result of the two models' different color handling, as discussed above. The original Rebel's colors are a little brighter by default, and at ISO 1600, its color saturation increased somewhat, while the saturation of the XT's shots decreased a bit at maximum ISO. The net result is that the shot from the original Rebel was quite a bit more saturated and contrasty than that from the XT. Since higher saturation and contrast would both act to increase noise levels, I took the (very unscientific) step of adjusting the tonal balance and color saturation in the XT's shot to match that of the original Rebel's shot more closely, trying for the best match in Marti's skin tones. The differences in contrast, saturation, and (obviously) image size make a direct comparison a little difficult, but in scrutinizing the full frame of even the tweaked XT shot, the net result is slightly lower image noise levels. (Image noise in the untweaked file was very noticeably lower, but for the reasons noted, that probably isn't a fair comparison.)
Dynamic range (the range of light to dark values that can be adequately reproduced by a digital camera) is always of interest, and here again I'm finding that numeric measurements of it don't correlate all that well with actual experience in the field. I don't have what I'd consider a good standardized subject for showing dynamic range with, but like to play with shots from our "gallery" shooting to see how well a camera handles shadow detail in real-world situations.
In the case of the Canon Digital Rebel XT, my gut sense was that it did very well, albeit not quite as well as some recent high-end D-SLRs I've tested. (Notably, the new Nikon D2x and the Canon EOS-1D Mark II.) Still, there's a ton of detail in the XT's shots, even in very dark shadows. The crop below shows how it handled the deeply shadowed undersides of a large tree limb, in a shot Shawn captured with Canon's nifty new 10-22mm f/3.5-4.5 ultra-wide zoom EF-S lens. Despite the depth of the shadows here, playing with the Levels adjustment in Photoshop pulls out a lot of detail, with relatively modest noise levels.
Imatest Resolution Results
Having opened with a discussion of the XT's resolution, it's probably fitting to close with res-target results from Imatest. Visually, the Rebel XT's performance is very similar to that of the EOS-20D, with "strong detail" visible to about 1,500 line widths/picture height along the vertical axis (corresponding to the horizontally-oriented resolution wedges), and somewhere around 1,650 lw/ph horizontally. One could argue for higher numbers in each direction, particularly along the vertical axis, but there's quite a bit of aliasing at higher resolutions, so my conservative nature says to not rate the camera as being capable of distinguishing finer detail than that described.
Looking at the test target with Imatest suggests that Canon has indeed tightened up the in-camera sharpening on the XT a fair bit, in fact coming fairly close to the results obtained with the standardized 1-pixel sharpening in Imatest. As a result, the raw and corrected resolution numbers from Imatest come in very close to each other. Using Imatest's "MTF 50" criteria (which seems to correlate well with visual sharpness), the average uncorrected value was 1511 LW/PH, while the corrected average came in at 1506 LW/PH. These are both quite good numbers, albeit somewhat below the average 1,610 LW/PH resolved by the 20D when its output was corrected to a standard 1-pixel sharpening. (Images from the 20D are softer out of the camera, but take sharpening post-exposure slightly better. The XT beats the 20D's numbers when set to its defaults, but loses ground when the images are corrected to reflect a standard sharpening operator.)
Looking at the intensity profile across one of the edges in the target, we see that the unadjusted image has slightly more than the "standard" amount of overshoot, and almost no undershoot in the shadows. The resulting edges are very "clean," but do include some artifacts from the sharpening process. As noted earlier, the best results are obtained with the XT when shooting with the in-camera sharpening turned off, applying strong/tight unsharp masking post-exposure in Photoshop or other imaging software. For most users though, the XT's in-camera sharpening achieves a very good balance between good apparent sharpness and relatively little distortion of the underlying image data.
Conclusion and Signoff
This wraps up my image analysis for this go-round with the XT. Since the sample we have is a full production version of the camera, I'll be proceeding ahead with analysis of our full test suite pretty quickly, as well as preparation of a full review. At this point though, I'm confident in saying that its image quality is first-rate, slightly bettering that of the original Rebel in most respects.
Read Shawn's Canon Digital Rebel XT Review & User Report to see what we liked and what we didn't.
Browse our Canon Digital Rebel XT Photo Gallery to get an idea of what this camera can do!
Most of our standardized test shots are in fact already captured - See the Thumbnail index of Rebel XT Test Images
Top 3 photos this month win:
1 Canon PIXMA PRO-100
2 Canon PIXMA MG6320
3 Canon PIXMA MG5420