Research And Improvement Of A Hybrid Image Filtering Based On Edge Detection Theory

Mean Filter Algorithm is critical in Linear Filter while Median Filter Algorithm is the critical one in Nonlinear Filter in image filtering area. However, there are different defects lie in both Algorithms. After discussing them, the dissertation introduces some algorithms in Hybrid Filter which is comparatively better. Each kind of Hybrid Filter Algorithm has its own advantage and flaw, so, the dissertation compares the simulation result and the theory of each one.Based on image filter theory and edge detection theory, the dissertation brings forward a kind of new Hybrid Image Filter Algorithm. To get rid of the defects of both Linear Filter and Nonlinear Filter, this algorithm judge if pixel point is edge point to do different kind of filtering by using edge detection theory. If pixel point is edge point, the algorithm uses a specific Median Filter algorithm the process, and if else, it uses a specific Mean Filter algorithm. The algorithm uses a specific method to find the edge point. Firstly, it finds the gray value tangential direction of each pixel point, and then divides the moving window into two parts, secondly, it compares the average gray value of each part to verify if pixel point is the edge point.The dissertation combines the method based on direction information and the method based on contrast to determine the gray value tangential direction; it also uses triangle module operator to join Minimum Total Error Probability (MTEP) Criterion and Neyman-Pearson (N-P) criterion to find the judgment threshold.Two simulation results are shown to verify the performance of this algorithm. One is about the comparison of three version of this algorithm, and the first two versions are lack of one critical part or two, so that we can make sure that each part of the algorithm is important. The other one is about the comparison between the hybrid filter based on edge detection theory and other hybrid filter. With the simulation result, we can clearly find that the algorithm of this dissertation is better to some extend actually. It effectively reduces the error probability and gives a new research direction for image filtering.