Image segmentation : finding the optimum threshold using discriminant analysis

Abstract

Digital image processing nowadays became a must. It is used in many domains including medicine, industry, etc…. One of the important techniques in digital image processing is image segmentation where each image is divided into two classes: one for the object and the other for the background. In order to be able to segment an image, we need to find the best threshold for the image. There are many ways to find the best threshold. In this paper, I focused on using discriminant analysis which was proposed by Otsu (1979). So our problem is reduced to either maximizing the between class variance or minimizing the within class variance. The threshold that has the minimum class variance is the optimum threshold and according to this threshold, a new matrix will be generated where each pixel in the matrix of the image will be converted either to 0 or to 255. The resulted matrix will be essential to find the segmented image using MATLAB.