A simple algorithm for restoration of images corrupted by streaks

A nonlinear algorithm based on a detection-estimation strategy for the restoration of images corrupted by streaks is presented. Only those pixels of the image which are detected as belonging to streaks are replaced with an estimation of the true values using neighborhood information. The removal of streaks is achieved without destroying important image features such as edges and details. First, a simple method is described for the case of images corrupted by horizontal streaks affecting entire rows in the image. The approach is then generalized for applications involving a wide variety of streaks of arbitrary sizes, angles, curved streaks, etc. Streaks which occur naturally in the image are however preserved. Computer simulations indicate that the proposed method provides significant improvement in performance over median filters, both perceptually and under MSE criteria. Illustrative examples are included.