Research of SVM-Based Edge Detection Algorithm Optimization
Zheng Sheng; Liu Jian; Tiari Jin-wen
State Education Commission Key Laboratory for Image Processing and Intelligent Control, Inst. for Patt. Recog. and Artif. Intel., Huazhong Univ. of Sci. and Tech., Wuhan 430074 China
Abstract:In this paper, the image intensity surface for the neighborhood of every pixel is well-fitted by the Least Squares Support Vector Machine (LSSVM), and the gradient and the zero-crossing operators are deduced from the LSSVM with the Radial Basis Function (RBF) kernel function, as an example. The decision is made whether a pixel is an edge or not based on the combination results of the gradient and the zero-crossings. One method using the edge detection evaluating merit figure to optimize the LSSVM parameters is proposed. The optimal configuration of parameters (σ2,γ) for the LSSVM with RBF kernel is (7, 1). With the selected parameters, the computer edge detection experiments are carried out. The experimental results demonstrate the proposed algorithm is reliable and efficient.