Optimizations of Canny Edge Detection in Ghost Imaging

Abstract

An optimization of the Canny edge detector’s application in ghost imaging is presented. Based on the pseudo-thermal light ghost imaging scheme with a binary object, a thin and accurate edge map can be extracted by using a Gaussian-filtering-optimized Canny edge detector. The scale of the Gaussian filter in Canny edge detection algorithm is the dominate factor in the performance of the edge detector, and can be evaluated by the bit error rate of reconstructed binary image based on the edge map. Simulation results indicate the optimal window size of Gaussian filter for ghost imaging is proportional to the full width at half maximum of the self-correlation function in the idler arm samples without any priori knowledge of the object. Experimental results show that, with an appropriate Gaussian filter, the reconstructed binary image can approach the original binary object with the minimum bit error rate, which means the edge detection result is optimal.