Digital images are represented by a stream, currently of 8-bit or 10-bit values representing the luminance and chrominance information, or a stream of 8-bit or 10-bit values representing the R', G', and B' information. Image processing sometimes involves multiplication of each digital word by: its proportional contribution to the processed image, a vector to relocate the pixel, an algorithm to change overall image size. To control these processes, additional information may be carried in the alpha channel synchronised to the image. As an example of the process, if an 8-bit sample is multiplied by an 8-bit factor, the product becomes a 16-bit word. At some point, this may have to be rounded or truncated back to 8 bits for the next operation. This introduces slight discrepancies in the result which may be visible as lagged edges, colour bleeding, etc. If successive truncations are performed during a sequence of image processing steps, the artefacts frequently become increasingly visible.

Good practise calls for maintaining some or all of the “extra bits” throughout as much of the image processing as the facilities permit.

Experience has shown that digital image processing provides the fewest distracting artefacts when the R', G', B' signals are first converted to the linear R, G, B. For complex image processing, and for critical results, the 8-bit encoding may be replaced by 10 bits (or more if that can be accommodated).