A Fast Edge-preserving Filter And Its Applications

As a front-end preprocessing technique, image filtering is the simplest and also the most important technology in image processing. Tradi tional filter methods pretend to blur the edge details while reducing the noise, but it will produce obvious artificial trace in deep application. Therefore, many researchers pay more attention on the edge-preserving filter.The edge-preserving filter based on global optimization, can achieve better effect in most of the image processing cases, relative to the various local algorithm, but generally the global optimization algorithm needs to solve a large sparse matrix equation, with the disadvantage of large operation cost and real-time applications. This topic proposed the Separated Weighted Least Squares Algorithm, SWLS for short, is an analog algorithm to accelerate the WLS algorithm. First, get the expected edge of the x direction and y direction by one dimensional WLS which can do a Gaussian elimination with time complexity O(N) to triangular matrix linear equation to each line and each column of the image matrix. Then, get the final edge-preserving result by solving the constant weighted gradient equation with the original image, the expected edge of the x direction and y direction. Compared to WLS,our algorithm can achieve more than 8 times on computational efficiency with comparative edge-preserving effect and better real-time applications. At the same time, our subject attempts to apply SWLS accelerate framework to a more general framework weighted gradient field equation. Then, take robust enhancement and sparse data interpolation for example, the two gradient field equation fo rms by dense input and sparse input is analyzed, achieve a similar effect with the original framework of the robust enhancement, and also analyze the reason of poor performance of sparse data interpolation.