Canon 1Ds Mark III Test Shots

Still Life

This is our new "Still Life" test target. We're combining some of the elements from previous shots (DaveBox and Res Chart) into this and the "Multi Target" shot below, plus added a number of elements that are very revealing of various camera characteristics and foibles.

Here's what to look for in this target:

• Tone-on-tone detail & noise suppression: The cloth swatches in the pinwheel were chosen because they show a lot of tone-on-tone detail, across a broad range of colors. This is just the sort of detail that noise suppression processing tends to flatten out. If you look at the detail in these swatches as the ISO increases, you'll see just where different cameras start to lose subtle detail. -- The white and tan swatches and the dark swatches tend to be particularly revealing of this. The label of the vinegar bottle (second from the right) is another great place to look for lost detail from noise suppression, as the image of the person at the top of the label is actually a depiction of a mosaic. The dark colors in the background and in the figure's clothes contain detail that's very quickly lost when a camera's noise suppression system kicks in. Cameras with really high-quality, low-noise sensors that require little noise suppression will be able to hold onto the detail in these areas, many others will show only a uniform swath of smudged color.
  
  Another place where you'll quickly see the effects of over-aggressive noise suppression is in the white salt grains of the salt grinder in lower left. Cameras are often more conservative about suppressing noise in highlight areas (because our eyes tend to see less of it there), but many cameras seem to have a hard time holding onto the subtle shadings that distinguish the salt grains from each other, particularly at higher ISO levels.
  
  
• Fine Detail: You'll find a lot of fine detail in the label of the beer bottle on the right, in its fine cursive text, but the other bottle labels hold a lot of fine detail as well. Fine text is often a good visual indicator of resolution, because our brains have an excellent idea of what the text should look like, so are very quick to notice even minor loss of detail.
  
  For really fine detail, look to the circular scale/calculator on the right side of the scene. Some of the fine lines there are extremely fine indeed. Looking at results from many different cameras with this target, we found that camera noise-suppression systems often confuse the fine lines with image noise, and so flatten them out. There's also a nice range of fine text sizes in this chart as well, once again great visual cues for resolution and detail.
  
  
• Highlight Detail: Three elements in this scene show off (or show up) a camera's ability to hold onto highlight detail. As mentioned above, the salt grains (and reflections of the studio lights) in the salt mill are examples of fairly subtle highlight detail that cameras' anti-noise processing sometimes obliterate. The folded white cloth under the mug on the right side of the frame likewise shows a lot of white-on-white detail that is easy to lose, particularly if a camera's tone curve is too contrasty. As it turns out though, the most sensitive test of a camera's highlight abilities seems to be the hank of white embroidery thread in the upper right corner. These fibers are unusually bright and reflective, so its easy for a camera to blow out detail in them.
  
  
• Shadow Detail: Several elements of this subject are useful for evaluating shadow detail, particularly the black mug and the pieces of folded black velvet, both under and inside the mug. The bottoms of the beer bottles also provide some gradations of deep shadow, and the clump of peppers in the bottom of the pepper oil bottle had a fair bit of detail that's far down at the shadow end of the tone curve.
  
  We were actually surprised when we constructed this scene just how dark the velvet and sides of the beer bottles ended up being. Even with the bright studio lights shining directly on it, the velvet in particular stays way, way down at the shadow end of the tone curve. With most cameras and on most monitors, the velvet will simply appear as an unrelieved swatch of black. To see whether it contains deep detail or not, in most cases you'll have to open the file in an image editor and boost the brightness dramatically, to bring the detail up into a visible range.
  
  
• Preservation of "Shape" in Strong Colors: As you approach the extremes of a camera's color gamut (its range of recordable colors), it becomes more and more difficult for the camera to show fine gradations of tone, because one or more of the RGB color channels are close to saturation. It's not uncommon to see a brightly colored piece of clothing or a vibrant flower appear in digicam photos as just a blob of color, because the camera ran up against the limits of its color gamut. The brightly colored embroidery threads in the upper right portion of the Still Life target are good examples of situations where this might happen. Pay particular attention to the bright red and dark blue colors here, as these are both colors near the edge of the typical sRGB color gamut.
  
  
• Color accuracy and white balance: It's pretty small in there, but we've included a mini-MacBeth chart, which displays very carefully controlled color swatches. Our Multi Target (see below) sports a full-sized MacBeth chart, but the one here serves as a good check of color balance and rendition, and is also useful for checking white balance on this particular shot.
  
  
• Image noise and detail vs ISO: As mentioned above, this target contains many elements useful for evaluating detail loss to anti-noise processing. We'll therefore always shoot a full set of test images of this target across each camera's ISO range, for every camera we test. (See below.)

Multi Target

Our new "Multi Target" is actually an interim design that we plan to replace with a modified version within the next few months. (By the end of summer, 2006.) In its current form, its just an ISO-12233 res chart with several other elements attached to it, covering portions of the target we don't generally use in our reviews. (We'll continue shooting the full res chart in SLR reviews for the sake of interested readers, but all lesser cameras will now show just the Multi-Target instead.) The added elements largely mirror ones that were present in our "DaveBox" test target, which we've now semi-retired. Here's some of what you'll find in this target:

Res chart elements: In placing the new elements on the ISO target, we were careful to leave the highest-frequency hyperbolic resolution wedges (the sets of fine lines that fan out vertically and horizontally), since these are what most people look at on the ISO chart to judge camera resolution from. There are also enough of the slanted black parallelograms available to use them to measure a camera's Spatial Frequency Response characteristics with Imatest.

Gray Scale: While reflective gray scales don't cover the full dynamic range of higher-end cameras, advanced readers may be interested in using the gray scale here to evaluate noise performance vs brightness level, and/or examine the shape of a camera's tone curve.

MacBeth Chart: This is about as common a color standard as you can get these days, very widely available for only mildly exorbitant cost, and quite well controlled in its production. It thus serves as a good basis of comparison between cameras and between test setups. Imatest also understands the MacBeth colors very well, and uses them to produce its color accuracy map that we feature in all our reviews.

Kodak Color Separation Target: We include this for reference because it's used by other reviewers out there, but caution our readers that it really isn't well-suited for use as an absolute color standard. As its name suggests, its actual intended purpose is as an aid in setting exposure levels for old-style film-based color separation, using panchromatic graphic arts film, RGB filters, and halftone screens. It's ideal for that usage, but the colors aren't well-controlled enough and are too subject to fading for it to be usable as a true reference standard for color accuracy.

Kodak Q60 target: This is another target that's perhaps not well enough controlled for quantitative measurements between cameras, but one that does have several useful characteristics nonetheless. For instance, it provides a good reference for the handling of various colors representative of common skin tones (the tan and brown patches along its right side), as well as of lighter, less-saturated tints of both additive and subtractive primaries (red, green, blue, cyan, magenta, and yellow). The MacBeth chart's colors lean heavily toward highly saturated values, and there aren't any pastel tones present at all. In the past, we've found that cameras with contrasty tone curves sometimes have trouble with the pastel tones in the Q60 target, making it a valuable reference that we'll continue to include.

"Sunlit" Portrait:
(This is our "Outdoor" Portrait test - read more about it here.)

The lighting in this shot is deliberately awful, about what you'd expect from noontime sunshine here in the Atlanta, GA area. (In fact, the color balance has been chosen to pretty well match the hazy sunshine here in mid-August.)

The reason for the harsh lighting is to provide a real "torture test" of how cameras handle conditions of extreme contrast; and in particular, how well they do holding onto highlight detail.

Look for:

• Overall color: Matching summer sunlight here in the South, the lighting in this scene is a bit more yellow-tinged than that in many parts of the country, or in the fall or winter. - So there may be an overall warm cast to the color. That said though, there's a fair range of color represented in the bouquet, presenting a tough challenge for the cameras. For some reason, the blue flowers seem particularly hard to handle, with many cameras rendering them as purple. (In real life, they're a light shade of navy blue, with just a bit of purple in them.)
• Skin tones: The overall slight warm cast will tend to leave Marti's skin tones a bit on the warm side as well. Nonetheless, look to see if her skin seems overly pink or if they have a too-bright tinge of yellow: Some cameras oversaturate skin tones (make their color too intense), leading to an almost sunburned look. A little oversaturation can make for a more "healthy-looking" complexion, but it doesn't take much variation for skin tones to look unnatural.
• Highlight detail: When Marti's skin tones are at a more or less normal level of brightness, how much detail can you see in her shirt? Does it blow out entirely to white, or can you still see the creases and folds in the fabric?
• Overall contrast: Most consumer digital cameras produce bright, contrasty images, because that's what most consumers like. Unfortunately, under bright sunlit conditions, many such cameras produce images with little or no highlight detail, and dark, plugged-up looking shadows.
• Shadow detail: The area under the flower bouquet is in quite deep shadow. Does the camera in question retain good detail here, with low image noise? To see, you may need to download the image and play with it in Photoshop(tm) or another imaging program. Brighten the image, and see how far detail extends into the shadows. Photo printers are generally much better at showing shadow detail than are CRTs, so you'll want a camera that preserves good detail here. The ability to boost brightness without encountering too much image noise is important if you ever have to "rescue" an underexposed image on the computer.
• Detail in areas of subtle contrast: Most digital cameras employ some sort of noise-suppression to remove electronic noise from their images. Noise suppression is a good thing, but only if it's not overdone. Too much noise suppression will "flatten out" subtle detail in areas of reduced contrast. You can often see this in hair, where the individual strands become blurred, and the image takes on an almost watercolor effect. Look at the detail in Marti's hair, and compare how it looks with different cameras in the Comparometer.
  

To view the entire exposure series from zero to +1.3 EV, see files E1DSMK3OUTAP0.HTM through E1DSMK3OUTAP4.HTM on the thumbnail index page.

The incandescent lighting used in most US homes actually has a very strong yellow color to it. Our eyes have an amazing ability to ignore color casts like this, something digital cameras struggle to emulate.

The incandescent lighting used for this shot is thus not only very common here in the US, but also very difficult for most digital cameras to deal with. While we probably want a little yellow color to remain in the image (to convey some of the mood of the original scene), too much will look unnatural and distort colors.

Most cameras' auto white balance systems have a great deal of difficulty with this shot, but many incandescent white balance settings struggle as well. (It seems that many cameras' incandescent settings are actually calibrated to the tungsten lighting used in professional studio systems, which isn't nearly as warm-toned as typical household lighting.)

If you intend to do much shooting indoors after dark, pay careful attention to this test, as cameras vary widely in this regard.

To view the entire exposure series from zero to +1.7 EV, see files E1DSMK3INTP0.HTM through E1DSMK3INTP5.HTM on the thumbnail index page.

ISO Series:
"ISO equivalent" refers to a camera's light sensitivity. ISO 200 represents twice the sensitivity of ISO 100, meaning that you can use a shutter speed that's twice as fast. Higher ISO settings are often required to get any picture at all when shooting after dark, but even in full daylight, using a higher ISO can help you freeze fast action. The problem is, increasing a digital camera's ISO also increases image noise. In practical terms, this means that higher-ISO images often can't be used to produce prints as large as lower-ISO ones. The tricky thing here is that high-ISO images often look much different when printed at various sizes than they do when viewed on-screen. In particular, for any level of image noise, you'll often find that while noise is quite evident at larger print sizes, as you reduce the size of the prints, there will come a point where it suddenly ceases to be an issue. We routinely print high-ISO photos from the cameras we test on our studio printer (currently a Canon i9900) at a range of sizes, and report our findings. If you're interested in investigating the effect of image noise for yourself, don't judge cameras' performance by how their images look on your CRT, viewed pixel-for-pixel. Rather, download the test shots linked in the table below and output them on your own printer, so you can see how prints of various sizes will actually look.

One additional note about this particular test series though: Because these images are shot under household incandescent lighting, the camera has to boost its blue-channel signal quite a bit to get back to a neutral color balance. Since the blue channel is generally the one with the most noise, this makes this shot a real acid test of noise performance. Noise levels in high-ISO shots taken under daylight conditions usually won't show as much noise. (See the "Far Field" test for examples of high ISO shots captured in daylight.)

Like several of our tests, these images are actually photos of a high-resolution poster, shot under studio lighting. As of this writing, the poster in question was produced from a high-resolution scan of a 4x5-inch transparency, scanned to a resolution of about 150 megapixels. Even at that, it's starting to show its age, as the combination of camera lens, film emulsion and scanner resolution mean that many of the current 5+ megapixel cameras come close to capturing all the detail that's present. (We have a new poster under production, produced from 45 separate digital images, "stitched" into a single 400+ megapixel image, that should comfortably outpace the resolution of portable digital cameras for years to come.)

Why did we choose to shoot a picture of a picture? The idea was to show a typical subject (a house and surrounding foliage) in a way that would be absolutely consistent from camera to camera. Any outdoor subject is going to vary considerably from day to day, as the lighting changes with the weather, atmospheric conditions, and season. Shooting a poster lets us compare images from cameras shot weeks, months, or even years apart, with the sure knowledge that nothing has changed from one shot to another.

Things to look for here are fine detail, as seen in the foliage and tree limbs against the sky, sharpness in the corners, and the preservation of subtle detail in the shaded brick patterns. - Many cameras with overactive noise suppression severely blur the brick patterns that are in shadow.

While the House poster in the shot above provides absolute repeatability from test to test, it doesn't offer the range of brightness (dynamic range) that the original scene had, nor does it contain the nearly infinite range of fine detail found in nature. For these reasons, we still shoot the original house, even though the vagaries of nature mean that no two shots will ever be directly comparable. (In fact, over the eight or so years since we first shot this subject, the trees in front of the house have now grown so large that they obscure much of the subject. - We're unfortunately going to have to switch to a different subject in the near future.)

Things to look for here are how well the camera handles the range of light levels from very bright to quite dark, and how well it renders the very fine detail visible in various parts of the image.

Note though, that because this is shot outdoors, the character of the light is unavoidably going to change quite a bit, depending on the atmospheric humidity and the time of year. - You thus shouldn't rely on it for absolute comparisons between cameras, since it's unlikely that conditions will be identical from one test to the next.

We've discontinued use of the Musicians poster in our testing, due to its low resolution, and the unavailability of any higher-resolution versions of the image.

Because most of its various elements are now contained or represented in the combination of the Still Life and Multi Target shots, we no longer routinely shoot the Davebox by itself, as would normally appear in this space. (We do however, still use it for our flash range and low light tests below.)

Low-Light Tests

Low light photography is an area where there are really enormous differences between digital camera models. This test starts at a light level about equivalent to typical city street lighting at night (one foot-candle), and then progresses down from there, each successive test being at half the light level of the preceding one. You may also see the effect of poor low-light autofocus in some of these shots, although we use a different test setup to check autofocus performance more directly. (The results of which are reported on in the main Test Results section.) Things to look for here include:

• Exposure limit: What's the darkest level a camera can handle at each ISO setting? If the leftmost images are reasonably bright, the camera should do fine with typical city night scenes. If you're shooting in the more dimly-lit suburbs, you'll need a camera capable of producing good images one, two, or three steps to the right of that.
• Autofocus Limit: How dark can you shoot and still get well-focused pictures?
• White Balance: Does the camera's white balance suffer at low light levels? (Many do.)
• Noise Levels: Look at the photos, print them on your own photo printer. How large a print can you make at acceptable quality levels, at various ISO settings and light levels?
• Detail loss to anti-noise processing?: Do details in the white gauze and even in the lettering on the test targets suffer at lower light levels?

(Note: If you'd like to use a light meter to check light levels for subjects you might be interested in shooting, a light level of one foot-candle corresponds to a normal exposure of two seconds at f/2.8 and ISO 100.)

Resolution Target

Because the critical elements of this target now appear as part of the Multi Target above, we no longer shoot standalone images of the Resolution Target.

Viewfinder Accuracy/Flash Uniformity

Viewfinder accuracy is an important parameter, especially for shots where framing is critical. The optical viewfinders on most digital cameras match the (poor) accuracy of those on film cameras, typically showing only about 85% of the actual final frame area. It's likely that this is a deliberate design choice by the camera engineers, to help avoid users accidentally cutting off the heads of their subjects. We disagree with this approach, or at least feel that it should be mitigated a bit, perhaps by increasing the accuracy to 90 to 95%.

Unlike the optical viewfinders, the LCD viewfinders on most digital cameras tend to be quite accurate. There are exceptions though, and it's unfortunately not uncommon to find an LCD monitor that only shows 90% or less of the final frame.

Things to look for on this test chart are:

• Optical and LCD viewfinder accuracy: When we shoot this target in the studio, we line things up so the bold black outline on the target is just visible at the edges of the viewfinder frame. The resulting photo then very directly shows how accurate the viewfinder is.
• Flash Uniformity: The flash units on many digital cameras don't illuminate the scene very evenly, especially at wide angle focal lengths. We shoot the viewfinder target with the flash enabled, so the photos here also show how uniform the flash is.