A Noble Image Segmentation Using Local Area Splitting and Merging Method based on Intensity Change

Abstract

This paper proposes a new image segmentation algorithm that involves local area splitting and merging based on intensity change. Most image segmentation algorithms take advantage of features such as pixel intensity and edge to split or merge an image. Therefore, in addition to susceptibility to noise, the latter algorithms have a problem in that they achieve different results depending on the initially selected seed location. The proposed method adaptively changes pixel intensity during the process of region segmentation to the representative intensity of the adjacent sub-area of high homogeneity. Therefore, this method is not affected by the initial seed location, and it also eliminates pre-process, such as noise removal, because the pixel intensity is progressively stabilized to the average value of object. In addition, this method preserves the edges of segmented objects and reduces the phenomenon of excessive region merger by determining the direction of the next merger upon splitting a local area into small sub-areas. Our experimental results demonstrated that the proposed method accurately segments images higher credibility than the existing image segmentation algorithms.