IMAGE QUALITY: Canon EOS 50D vs. Olympus EVOLT E-30

Color Reproduction
Color accuracy is the first measure of a digital camera – lifelike color is vital in most kinds of photography. We look at two measures of color accuracy: saturation and color shift. In our tests, the Canon 50D takes the prize for accurate saturation, but the Olympus E-30 performs significantly better on color shift – our testing software calls it “color accuracy.” Of the two problems, color accuracy is the more difficult to fix in post-processing.

We evaluate color accuracy with a lab test: we shoot a GretagMacbeth color target under controlled lighting, and run the resulting images through Imatest, the leading software for digital image evaluation. In addition to delivering numerical data, Imatest plots each GretagMacbeth color swatch on a gamut chart. The colors at the edge of the gamut chart are highly saturated, and at the middle, they are completely unsaturated. Black, white and gray swatches appear at the center of the chart. Other colors are scattered around the gamut. The small squares show the ideal chart position for each color, and the circles show the camera's reproduction. The distance between each pair of square and circle shows the amount and type of color inaccuracy. If the circle is further from the center than the square is, the camera has oversaturated that color. The distance between a circle and square rotating around the center shows color shift or color error – tinting blues green or purple, for instance. The difficult problem with color error is that in most cameras, the errors vary from color to color, so improving one color can make the other ones worse.

Olympus E-30

Canon EOS 50D

In our tests, comparing shots taken with custom white balances, the E-30 delivered saturation figures of 104.4 percent at ISO 100 and over 105 precent at ISOs from 200 to 800. At 1600, the E-30 dropped slightly to 103.1 precent. It dropped to 99.16 precent at ISO 3200. In its “Standard” picture mode, the Canon 50D turned in scores between 99.4 and 99.61 percent, except for an outlier of 97.36 at ISO 1600. In its “Faithful” mode, the 50D turned in saturation levels under 90 percent, suggesting that Faithful doesn't deliver on its implied promise. As far as saturation goes, more is not better – numbers over 100 typically lead to blocked-up channels and a loss of detail. The 50D has a clear advantage in this measure.

Canon EOS 50D	Olympus E-30

The E-30's most significant advantage over the 50D is in color accuracy. With a custom white balance and noise reduction off, the E-30 managed to keep its error below 7.0 at every ISO, with a low score of 6.21 at ISO 200. In contrast, the 50D's scores in Standard mode clustered around 8.25. In Faithful mode, the 50D managed one score below 7 – 6.89 at ISO 100.

Noise
Image noise is pixel-to-pixel variations in either color or luminance that are artifacts of the cameras' electronics. Noise can obscure the image, hiding texture and detail, or contaminating color. Many components and processes in a digital camera can contribute to image noise – the individual photoreceptors on the sensor, the sensor architecture, the conversion of analog to digital signal, image processing and compression, and so on. As a general rule, the size of the photoreceptors on a camera's sensor is a significant factor in noise. Bigger receptors deliver more powerful signals, which can overwhelm system noise.

With roughly comparable pixel counts, but very different sensor sizes – the E-30's 4/3 sensor is much smaller than the 50D's APS-C sensor – the two cameras have very different receptor sizes, and therefore, different noise performance. That's the way it is. In-camera noise reduction can significantly improve noise performance, but the E-30's reduction system would have to be much better than the 50D's to overcome the sensor size difference.

With the camera's noise reduction systems in their standard modes, the Canon 50D has very good noise scores that run from 0.53 at ISO 200 to 1.64 at 3200, while the E-30 runs from a decent 0.8 at ISO 100 to a bad 2.59 at ISO 3200. With noise reduction set to “High,” the E-30 delivers a score of 2.01 at ISO 3200, significantly worse than the 50D's 1.28 score with noise reduction set to high. Apparently, the E-30 can really turn off noise reduction, because its worst score was a really awful 6.99 at ISO 3200 with noise reduction off. That's roughly comparable to a free cellphone's performance. The differences between the Canon 50D's best and worst scores at ISO 3200 – 1.28 to 2.0 – aren't nearly as radical as the E-30's. It seems as though the 50D always does at least some noise reduction, even when the user turns the feature off.

It's also important to note that proper exposure makes a big difference in noise scores. It only makes sense – cutting exposure reduces the strength of image data, while leaving the spurious sources untouched. After bracketing a shot, we ran the tests on underexposed images from both cameras. A two-thirds underexposure on either camera yielded noise increases on the order of 30 to 40 percent.

Noise Reduction

We photograph Editor-in-Chief Alex Burack's lustrous brunette wig with every camera we test. Hair is useful to look at when evaluating noise and the tradeoffs in using noise reduction. In a typical portrait, individual hairs should be distinguishable, and they're about a pixel wide. So, a noisy pixel – one that is off in color or luminosity – can break up the line of a hair.

Noise reduction algorithms have trouble with hair, because they work by comparing nearby pixels, and making the odd ones closer to the average color for the area. With patches of blue sky, that works well, because the great majority of pixels are about the same color and tone. It's easy to find the bad ones, and easy to decide what color to make them. With hair, noise algorithms have more trouble, because hair naturally has very local variations in color and brightness. Single hairs will throw unique highlights because they bend at a unique angle. Cameras are challenged in distinguishing between noise and signal in a lone bright pixel, if the neighboring pixels vary enough – the system can't tell what the average is, or what's an outlier.

The hair shots from both cameras show the two types of noise distinctly. Color and luminosity show up very differently. Color noise appears in blotches that get larger and larger with higher ISO. Luminosity noise doesn't enlarge with ISO. Instead, more small areas of bright and dark show up. In our hair shots, the Canon 50D shows less-pronounced color and luminosity noise than the Olympus E-30 at every ISO level. Comparing images taken with the cameras' noise reduction systems set to their various strengths, we see that both cameras increase the amount of noise reduction as ISO increases. Though the 50D keeps noise pretty well in check up to ISO 800 when it's set to Standard or High, we see a steady loss of definition in the individual hairs. At 1600 and 3200, noise increases significantly, even with reduction set to High. As we look at the screen, the increased noise looks a bit like improved definition. The Olympus E-30 shows the other option in the noise/definition tradeoff: the E-30 shows steady increases at every ISO step, and at each noise reduction setting, but it preserves definition pretty well -- more or less until the noise itself erodes the fine details.

 

Canon 50D Noise Reduction Off (ISO 400)	Olympus E-30 Noise Filter Off (ISO 400)
Canon 50D Noise Reduction Standard (ISO 400)	Olympus E-30 Noise Reduction Standard (ISO 400)
Canon 50D Noise Reduction Strong (ISO 400)	Olympus E-30 Noise Reduction High (ISO 400)

 

Canon 50D Noise Reduction Standard (ISO 800)	Olympus E-30 Noise Reduction Standard (ISO 800)
Canon 50D Noise Reduction Strong (ISO 800)	Olympus E-30 Noise Reduction High (ISO 800)
Canon 50D Noise Reduction Off (ISO 800)	Olympus E-30 Noise Reduction Off (ISO 800)

 

Dynamic Range

Dynamic Range indicates a camera's ability to render detail in both highlights and shadows. Noise performance influences dynamic range significantly, so the E-30's disadvantage in noise performance translates directly into a disadvantage here. We report results for dynamic range at a low ISO. Readers should extrapolate poorer performance at higher ISO, roughly in line with noise performance.

The Olympus E-30 notched scores of 7.51 stops of dynamic range at high quality, and 9.9 stops at low quality. The Canon 50D delivered 8.18 and 11.3, respectively. We expect the differences at both high and low quality will be visible in prints. The target markets for these cameras include middle-of-the-road professionals such as the typical wedding photographer or non-elite photojournalist, and wedding shooters will have an easier time holding detail in white dresses and black tuxes with the 50D than with with E-30. The difference will be more pronounced at higher ISOs.

We test dynamic range by shooting a Stouffer step chart, which is a piece of conventional silver film processed to show more than 13 stops of brightness when backlit with a light box. We evaluate the images with Imatest. Our “high quality” score shows the brightness range at a noise threshold that's acceptable for the subject, rather than the background. “Low quality” includes more noise, and indicates the range that preserves shadow and highlight textures. Tones outside the low quality range are lost in the camera's background noise. In highlights, they're blown out, and in shadows, they're dead black.

White Balance
Taking custom white balances is a bit like flossing. It's not a barrel of laughs, but you should do it. The temptation to skip it fades significantly for anyone who has to recover from the consequences of poor practice.

We found that both the Canon 50D and the Olympus E-30 had pleasing daylight presets, and handled typical daylight snapshots well with auto white balance. The E-30 renders daylight a little warmer than life both with its daylight preset and in auto mode. The 50D and the E-30 produced pleasing results, but the Canon was more accurate.

The cameras were easily fooled in shade and most artificial light, and could easily go astray in auto mode, neutralizing scenes where a particular color predominates. Both the 50D and the E-30 have problematic presets for artificial light as well. Both produced images that were unpleasantly cool with their flash presets. The fluorescent presets were wildly inaccurate – Canon and Olympus seem to fight like hell to avoid the standard, ugly green tint, so they go for a tungsten-toned orange error. Both cameras' tungsten presets are very warm, which is pleasing in many cases, but not at all accurate.

We find it reassuring to note that the cameras' custom white balance systems agreed closely. When we shot on a tripod and used identical targets and lighting to set the balance, the 50D and the E-30 matched. We looked at RGB values for patches of an off-white wall, and the relative differences between red, green and blue values were within two percent for the two cameras.

Resolution
Resolution describes the level of fine detail a camera can render in an image. Again, we use Imatest software to evaluate shots from each camera. For this test, we photograph a high-contrast test target which shows a black bar against a white background. Imatest analyzes the transition from black to white in the image. A perfectly sharp image would show a border of pure black pixels against a pure white background. For a number of reasons, cameras don't deliver that. On all digital cameras except ones with Foveon sensors, each pixel is made up of information from red, green and blue receptors which are next to each other, instead of in the exact same spot. (Foveons put the pixels on top of each other.) That means that each pixel is a composite of the data from each receptor, and in some ways a compromise. The receptors also sit behind an “antialiasing” filter, which blurs the image to prevent color fringing where the three color's data disagree, and also to avoid a sharp stairstep pattern on borders that aren't perfectly vertical or horizontal.

Given all that, it's merely surprising rather than baffling or gob-smacking that the Olympus E-30, which has about 20 percent fewer pixels than the Canon 50D, comes out on top. The measure we use for resolution is line widths per picture height, which is theoretically the maximum number of horizontal lines an image can show. Our best result with the Canon 50D was 2057 lw/ph in a shot taken at ISO 100 at f/6.3. Imatest indicates that the image was undersharpened by 15.5 percent. The E-30 scored 2133 lw/ph at ISO 100 and f/11, with 12.9 percent undersharpening. The E-30 is only a little bit better, but it's still better.
 

At the risk of sounding like Canon apologists, we'll note that lenses are a significant part of the system. Canon sent us a low-end one, while Olympus sent us an excellent optic. With an L series lens, the 50D would have done better. Otherwise, we set out as even a playing field as possible: the E-30 and 50D were mounted on the same very heavy tripod, pointing at the same target under the same lighting. Both were in their standard modes, with noise reduction low. Imatest's measurement of sharpening doesn't necessarily follow the camera setting – it measures how much additional sharpening could be done to an image without creating artifacts.