Image thresholding based on edge information analysis

Binarization (i.e., image thresholding) is widely used as a preprocess algorithm in image analysis and understanding. In this paper, a novel image thresholding method is proposed, which is characterized by its simple procedure and strong ability of preserving detailed information. The proposed technique includes three steps. Firstly, the edges of an image are detected using the Canny method. Secondly, the gray-level transition range of each edge point (i.e., the maximum and minimum intensity value in a 3*3 block surrounding the current edge point) is calculated and then a new histogram of cumulative edge gray-level transition range is obtained. Finally, the peak of the new histogram is set as the optimal threshold for image binarization. Experimental results show that the proposed algorithm is substantially more robust and more reliable than the classic Otsu method.