An optimization method for edge-detector parameter tuning based on visual perception

An optimization method for tuning the parameters of edge detection algorithms based on visual perception is proposed and applied to the Sobel, Laplacian of Gaussian (LoG), and Canny detectors. The method uses human visual quality perception to compare edge maps generated by different edge detector parameter values and iteratively reduces the parameter search space by means of a coordinate search. The method is applicable to any parameterized edge detector. Examples demonstrate that use of the Canny detector yields the most visually appealing edge maps. However, it requires 2.5 and 1.8 times the number of iterations required by the Sobel and LoG detectors, respectively.