Thresholding Method Based on the Relative Homogeneity Between the Classes

Image binarization method focusing on the objects emphasized much the homogeneity of the object gray level distribution, can overcome some shortcomings of famous Otsu’s method. In this paper, the specific information of image is considered, the gray and neighborhood average histogram are applied, a more detailed description of the threshold calculation is presented. A new threshold discriminant criterion is proposed with relative homogeneity information both of foreground and background. Using valley point neighborhood histogram information, a modified discriminant analysis is constructed. The experimental results on images with a minimal distribution difference between classes show that, compared to both Otsu’s and focusing on the objects methods, the proposed method has not only better segmentation accuracy, but also better adaptability for images.