Threshold image segmentation based on maximum scatter difference discriminant criterion

Previous research results show that threshold obtained by maximum between-class variance method (i.e. Otsu method) is biased when the area of object and background differs significantly and may lead to failure segmentation. A new image segmentation method based on maximum scatter difference is proposed. Maximum scatter difference uses generalized scatter difference, i.e., the difference of between-class scatter difference and C times of withinclass scatter difference, as the discriminant measure. Maximum scatter difference considers simultaneously the function of discrimination of betweenclass scatter difference within-class scatter difference. The proposed method can prevents the threshold biasing from maximum between-class variance method. Experimental results show that the proposed method can obtain better segmentation result than otsu method by appropriately selecting parameter C.