Fast combination filtering based on weighted fusion

This paper presents a fast filter based on a modified Lee filter. The smoothing operation in the proposed filter is a linear combination of the center pixel value and the average of pixel values in a window. Based on the new definitions of edge and noise, the local gradient mean is introduced to compute the coefficient of the linear combination. Besides, a multi-scale method is employed to smooth the pixel values near edges. The main advantage of the proposed filter is its computational cost. In addition to some point operations, only several mean filters are taken in this filter. No matter how large the window size is, the time complexity is O(N) (N is the pixel number). Experimental results have shown that the proposed filter can effectively smooth images while keeping edges well, thus can be widely used in computer vision and image processing.