The paradoxical evolution of sensor SNR over timeSunday November 16 2008 Sensor Insight
We have seen that SNR has decreased by 1dB in the last five years. However, it is probably clear to all photographers that overall image quality has significantly increased during the same period. Why is this? Simply put, the number of pixels has doubled during the same time period, which means that when viewing at full screen or when producing a print, twice as many pixels will be averaged to achieve the screen or print resolution, resulting in a greatly improved SNR.
Let’s normalize the SNR for a 20x30cm print at 300dpi, which is roughly equivalent to an 8Mpix photo. The following graph shows the normalized SNR as a function of the camera release date. Cameras with a full field sensor format (24x26mm) are displayed in blue, and cameras with an APS-C sensor format (around 15x22.5mm) are displayed in green. (Intermediate formats such as 19x29mm—e.g., that of the Canon 1DMIII—have not been plotted.)
normalized for a 20x30cm print
We see that the full-frame sensors (blue dots) have a clear advantage over the APS-C sensors (green dots). This is because full-frame sensors are collecting twice as much light as APS sensors. It is easy to see that the average difference of SNR is about 3dB (a one-stop gain),4 which is exactly what is expected when doubling the amount of light.
We also see a greater improvement over time for full-frame sensors than for APS-Cs. In five years, full-frame sensors have gained 5dB, which is a gain of nearly two f-stops! For the same period (roughly from 2003 to 2005), APS sensors have gained only 2 dB. However, there has been only limited progress in improving SNR over the past three years or so.
Based on most photographers’ experience, we would probably expect much more improvement on the APS side — after all, the quality of a 2008-model camera at ISO400 is much greater than the quality of a 2003-model camera at ISO200! Why isn't this the case?
The answer lies in the fact that we are not looking directly at RAW images, but rather at RGB images after RAW conversion. Sensors are not the only element of image quality that have advanced over these past few years. RAW converter algorithms, whether embedded in the camera or as software in a PC (e.g., Adobe Photoshop, DxO Optics Pro, etc.) have also made large advances. For example, over the past two years, DxO Optics Pro RAW conversion (with its denoising algorithm) showed a gain of +4dB in the SNR, a much larger improvement than that achieved for sensor technology.
These advances in RAW conversion algorithms will continue, underscoring the advantages of shooting in RAW, as RAW images can always be reprocessed with every new release of a RAW converter. Of course, however advanced RAW converters become, the best final image will still rely on shooting the best RAW image possible.
4This also means that for a 20x30cm print, the image quality of an APS-C camera at ISO200 will be on average equal to the image quality of a full-frame camera at ISO400.