Performance analysis of entropy thresholding for successful image segmentation

Image segmentation refers to a procedure of segmenting the foreground (object of interest) from the background. One of the well-known methods is thresholding based segmentation that segments an image according to a threshold value. Most of the proposed methods either proposing a new algorithm or improvising the algorithm to segment the foreground. However, there is no analysis is carried out to determine the successfulness of the methods under different conditions. This main contribution of this paper is to analyse the entropy thresholding namely the method proposed by Kapur and Li for various parameters which include noise measurement, size of the object, and the difference in intensity between the background and object. In this paper, a few conditions were proposed to ensure successful image segmentation. Based on the experimental result, intensity difference needs to be around 35% and the object size is about 73% for all noise levels for Kapur. For Li entropy, the intensity difference needs to be at a minimum of 44% and 80% for object size. It is demonstrated that the proposed conditions accurately foresee the result of image thresholding based on Kapur and Li entropy.