Optimum Thresholding for Medical Brain Images Based on Tsallis Entropy and Bayesian Estimation

Thresholding is a popular technique for image segmentation, specifically in the field of medical image processing. The main challenge for image thresholding is to determine the optimum threshold based on intensity distributions of object and background in the image. In this paper, we propose a new image thresholding method by injecting the Bayesian probability estimation into the classical Tsallis entropy framework. The classical algorithm assumes that the intensity distribution of object does not affect the background pixels, and vice versa. However, the intensity distributions of object and background are essentially crossed. It is possible to estimate the probability of a pixel belonging to object or background by Bayes rule, and use it to update the classical form of Tsallis entropy. The optimum threshold is finally determined by optimizing the information measure function defined with the new form of Tsallis entropy. Extensive experiments conducted over two public datasets of medical brain images have verified the significant superiority of the proposed method.

I. Introduction

Segmentation is one of the most challenging and popular tasks in computer vision for its wide range of applications in recent years. Image segmentation refers to partitioning non-overlapping regions with similar properties in an image, such as texture or grayscale. Currently, many segmentation methods have achieved impressive results.