Noise in RAW Images

I collected RAW images from the web in order to characterize the noise levels of various cameras. The results were surprising. Most of the sampled RAW images had noise levels between 1 and 3 on a scale of 0-255 (black to white). In terms of bit-depth, most cameras had noise in the 7th or 8th bits. One camera, a very expensive DSLR, had noise in the 9th bit (< 1 on the scale 0-255).

Method of Measurement
Ideally one should use a photo of a monotone mid-gray surface. The ISO level should be "normal", around 100, since higher speeds increase noise. The surface being photographed must be absolutely uniform, of course. One way to accomplish this is to make an out-of-focus image, so that surface irregularities are blurred over many pixels. Another way is to photograph a gray overcast sky. The photo must not be too bright, since saturation of the camera sensor eliminates noise (all pixels at maximum). Given a suitable photo, noise can be measured as the difference between each pixel and the average of its immediate neighborhood. I used a program (Fotocx) which does the following: using a clicked spot on the image, measure the noise for each RGB color for each pixel within a radius of 10, average these values to get an overall noise level per RGB color.

Using images from the web for this project was not ideal, since a uniform gray area was not always available. I was able to find 21 images that I judged suitable after an hour or so of searching. I used out-of-focus areas when possible. I measured many spots and used the lowest noise values found. I inspected the images at 400% size to be sure I was measuring noise and not surface texture.

This image is typical. An out-of-focus area above the dog's ear was measured.

Others have done similar projects and reached similar conclusions. Search the web for "RAW image noise" and you can find these (most hits will be about using Photoshop and other tools to reduce noise).

JPEG images produced by the camera have less noise than their corresponding RAW images. The reason is that JPEG images are processed inside the camera to reduce noise. Others have written about this: search the web for "camera jpeg noise".