Frequency selective signal extrapolation with applications to error concealment in image communication

Signal extrapolation is an important topic in image communication where it can be used for various prediction purposes such as concealment of image data corrupted by transmission errors. This contribution outlines a method for extrapolating a signal beyond a limited number of known samples. The known signal samples are approximated by a set of basis functions which are defined over an area covering known as well as unknown samples. By minimizing a suitable error criterion and successively selecting the most dominant basis functions, a non band-limited signal extrapolation can be obtained. It is shown that this extrapolation can successfully be used for concealment of transmission errors in video communication as well as elimination of defective pixels in X-ray imaging.