A post-processing algorithm for compressed digital camera images

Digital cameras generally produce color images by placing color filters on monochrome sensors. The sensor produces one color at each location, and the other two color values for that location are interpolated. A lossy compression stage, using a coder such as JPEG, often follows. For a fixed interpolation algorithm, the color interpolation stage enforces a spatial dependence among pixels. We show how we may exploit this dependence to reduce the compression noise. This is achieved using the algorithm of projection on convex sets. For a fixed compression scheme we can also attempt to find the interpolation method that minimises the compression noise in the original data, and we show an approach to solving this problem.