Fast optimal thresholding based on between-class variance using mixture of log-normal distribution

Abstract

Image thresholding is a technique used to estimate threshold values for segmenting an input image into distinct regions. The goal of image thresholding is to simplify or change the representation of an image into something that is easier to analyze and is more meaningful. The most famous image thresholding method is Otsu's global automatic image thresholding method which has been widely applied in many fields, especially those with real-time applications. In this paper we propose a new method for segmenting images based on Otsu's method by estimating the threshold using a histogram. Our method is based on between-class variance, and finds the optimal threshold using the first derivative of the BCV relation to obtain iterative equations which then produce the optimal threshold. We apply this method to SAR images, where it gives promising results compared with Otsu's method based on the Gaussian and gamma distributions.