This paper describes inverse filters that restore imagery with minimal artifacts. Deconvolution kernels are obtained by windowing Wiener-type inverse filters (WTIFs). Windowed WTIFs exhibit good restoration properties. Constraining the size of the WTIF kernel is necessary because of the edge effects associated with discrete convolutions. Spatial windowing is necessary even if processing is performed in the frequency domain. The kernel design procedure is described, and the theory is explained. The performance penalty associated with limiting kernel size is discussed and quantified.
展开▼