DxO Labs technicians strictly follow a precise set of testing protocols:
 |  |
DxO Noise chart |
Gretag Macbeth Color Checker |
All measurements are performed using DxO Analyzer, which utilizes either the DxO Noise chart or a Gretag Macbeth Color Checker.
The Gretag Macbeth Color Checker features 18 uniform color patches and 6 gray level patches. The chart is placed within a reflective light box that delivers a uniform field of light onto all patches of the Color Checker.
The DxO Noise chart is a transmission chart consisting of a set of circular holes equipped with neutral density filters ranging from D=0 to D=4. The density filters are positioned on a circle and the chart itself is placed against a light table delivering a uniform lighting across the whole chart. When shooting such a chart with the camera being tested, the sensor sees a wide range of light levels with a 1/10,000 ratio from minimum to maximum.
Noise is measured for each color channel: R, Gr, Gb, B. The mean gray level and noise characteristics are computed for each patch. These characteristics are interpolated for all gray levels to form the basis for the full SNR calculation. DxO Analyzer then computes the dynamic range, tonal range, and color sensitivity of the system.
Measuring in RAW matters
DxO Analyzer is able to measure noise both on RAW and RGB images. The presentation and the meaning of the results are the same. However, noise measurements on RAW and RGB are complementary and describe different aspects of the camera. Noise measured for RAW measures the noise of the sensor, but since RAW conversion changes the distribution of the noise, classical noise measurement on RGB images leads to different results.
Depending on color filter spectral response in particular, it may be necessary to apply a white balance with a strong amplification (up to three or four times) on one or several channels. Sensors are commonly not very sensitive to blue wavelengths, although the blue channel is strongly amplified. It is also normal that the spectral responses of the red, green, and blue channels overlap significantly, which leads to a color rendering with a strong chrominance amplification. The combination of both white balance and color rendering may lead to a strong amplification of the noise (up to 8 to 10 times). These adjustments and amplications are revealed through Color Sensitivity Measurement.
The demosaicing algorithm has to mix information from different color channels, which leads to a combination of noises as well. Sharpening filters may also increase noise, although Image Signal Processors (ISPs) usually include denoising filtering. Compression also creates quantization noise that may be structured by the compression itself (as with the 8x8 JPEG blocking effect). This said, very strong compression may use a quantization larger than the noise standard deviation, with the result that the noise may be completely absorbed by the quantization.
The bottom line is that whatever the RAW converter, a better RAW file delivers a better image. Photographers shooting in RAW need to know what RAW delivers, as future technical advances to RAW converters can be applied to RAW images.
Backlinks
On the dxomark web site, we evaluate and rank many types of digital cameras with image sensors that vary widely in pixel count, pixel size, and digital signal processing. To ensure that sensor performance comparisons between cameras are fair, it is very important both to test under identical shooting conditions and to take viewing conditions into account.
Read more