Image binarization focusing on objects

Abstract In this paper, an image binarization method based on a new discriminant criterion is proposed. The criterion emphasizes much the homogeneity of the object gray level distribution and while intentionally de-emphasizes the heterogeneity of the background such that the new binarizing or thresholding method can overcome some shortcomings of famous Otsu's method. Experimental results on the three real images show that compared to both Otsu's and recent Kwon's binarizing methods, the proposed method has not only visually better or comparable segmentation effect but also, more favorably, removal ability for noise.