Thresholding Algorithms for Image Segmentation -Entropy Based Comparison

Abstract

Image segmentation relates to the process of labelling of pixels in any image. This concept of image segmentation is generally used to find the Region of Interest – ROI in images, thus it is used very frequently in the field of computer vision. Among the various techniques of image segmentation, Thresholding is quite common, this technique is quite simple and efficient. Thresholding can be done locally as well as globally, and this selection of suitable thresholding technique is a critical factor for image segmentation. The Otsu’s method and K-means algorithm are commonly used techniques for thresholding, the Otsu’s method works on Global thresholding and the K-means works on Local thresholding. But both methods i.e. Otsu’s method and K-means algorithm, explores the criteria of minimizing the within-class variance, to yield better segmentation results. But among the two, which one is better? The work performed in this paper relates to the comparison of the Otsu’s method and K-means algorithm, to find the better among the two – by using entropy as the key comparison parameter.