Maximum Inter Class Variance Segmentation Algorithm Based on Decision Tree

Introduction

Image segmentation is an important and complicated technique with many applications in image processing and analysis, such as computer vision, pattern recognition, medical image processing. The aim of image segmentation is to extract regions of interest from complex scenes.

During the past decades, many different kinds of image segmentation approaches have been proposed. Generally, image segmentation methods can be divided into four types: image thresholding, image boundary based, image region based and mixing segmentation technology. Among these segmentation approaches, the thresholding method is widely used due to its simplicity and ease of implementation (Goh, Basah, Yazid, Safar, & Saad, 2018; Mittal & Saraswat, 2018). Its main idea is choosing a threshold which can distinguish the image background and target in the image. And the maximum inter-class variance segmentation algorithm (Otsu, 1978) is one of the classical image thresholding segmentation algorithms. It is a kind of global automatic nonparametric unsupervised algorithm and widely used, which takes the maximum inter class variance as measure criterion. While there is still a false targets problem in the maximum inter class variance segmentation algorithm. And the study of this paper is based on the segmented image which are segmented by the maximum inter class variance segmentation algorithm. After segmented by the maximum inter class variance segmentation algorithm, the segmented image is obtained which is composed of many independent regions (Li & Feng, 2016). Each region in the segmented image corresponds to a target. These targets contain not only the targets of interest, but also the false targets. And these false targets are not caused by noise but exist in the original image. It is difficult to split these false targets out effectively because their grayscale values are similar to the grayscale values of the regions of interest. For the segmented image, these false targets are interference, and the removal of false targets plays a very important role for image segmentation.