A new image thresholding method based on relative entropy

This paper analyses the superiority of choosing the two dimensional histogram of an image as the feature space instead of the co-occurrence matrix of an image. Based on the two dimensional histogram of an image, relative entropy is used to measure the difference between an image and its thresholded image. The vector which minimizes the relative entropy is selected as the threshold vector. Experimental results show that compared with other thresholding method, our method can achieve better thresholded results and is of strong robustness.