Topological median filters for pattern recognition

This thesis demonstrates the use of the topological median filter for improvement of pattern recognition processes. The method used in the thesis relies on the edge preservation property of the topological median filter. Topological median filters do not have the same noise removal capability of the conventional median filters, but they better preserve the edges. This edge preservation property is very suitable for image prefiltering prior to the edge detection. The results without prefiltering and with prefiltering by conventional median filters and topological median filters are compared on edge detection performance and edge accuracy. The results show the advantages of the topological median filters over conventional median filters in edge preservation and edge position accuracy. Finally, processing speed is compared for MATLAB m-files and the C code implementation of MATLAB mex-files of the topological median "light" filter.