Image segmentation based on multiple means using class division method

Abstract

Image segmentation is the process of partitioning a digital image into multiple segments. Image segmentation is typically used to locate objects and boundaries (points, lines, curves, etc.) in images. Segmentation is an important part of many automated image recognition systems. The goal of segmentation is to simplify and/or change the representation of an image into units that are easier to analyze. It is the process of assigning a label to every pixel in an image such that pixels with the same label share certain visual characteristics. A histogram based image segmentation method is presented which divides the whole image into sub regions based on two mean values. The means are computed based on a preset threshold value calculated using Otsu's method, which is dependent on the intra class variance and class probability. Based on the mean values and the threshold, the image histogram is divided into three classes, LowIntensity, HighIntensity and unknown class. The unknown class can be further subdivided to these three classes based on modified means calculated using modified threshold‥ This process continues until no further subdivision based on mean values is possible. Finally, all the respective class regions are merged. The method better identifies fine structures of objects in complex images. The computational cost is found to be low for this method as the iterative method consumes reduced computational time.