Nonlinear operators for edge detection and line scratch removal

A nonlinear edge detection and line scratch removal method is proposed. The nonlinear operation is performed on the differences and sums of four neighbor pixels. For edge detection, the first derivative of the image brightness function is approximated by computing the maximum horizontal and vertical differences along the vertical and horizontal directions, respectively. The edge-detected result appears to be similar to the one obtained using Robert's operator. The method can be used to smooth out a line scratch that manifests itself as a narrow, bright or dark, vertical line. The minimum (maximum) of two sums between horizontal neighbors is selected for bright (dark) vertical line removal. Several frames from a motion picture with line scratches have been processed using this method, and visually pleasing restoration results have been achieved with very little smoothing.