Multiscale gap statistic for edge detection based on evidence theory

To solve the conflict between image denoising and edge localization, a novel edge detection algorithm based on multiscale gap statistic information fusion through DS evidence theory was proposed. Gap statistic of gray-scale image was defined. Detection uncertainty was introduced in fusion procession. Mass functions (or basic probability assignment functions) were designed based on the relationship between Gap statistic response and detection threshold. Detection uncertainty reaches maximum value at the detection threshold. And mass functions corresponding to each Gap statistic at different scales were combined by Dempster's combination rule. Pixels were classified into edge or off-edge according to the joint mass function. Tex experimental results show that this proposed method is effective in reducing detection uncertainty and better than classical edge detectors.