Graph Theory Based Image Segmentation

Image segmentation is an important and critical image analysis technique. In imageengineering, image segmentation is not only a critical step between image processing and imageanalysis but also the foundation of further understanding of image. Generally, only the interestedpart of image which is commonly particular and have unique characteristics is paid attention towhen we study images or put them into use. In order to analyze the interested target, first of all, theimage is divided into regions which have their own characteristics, and the interested target isextracted from the original separation; then, subsequent steps are done, such as measuring, featureextraction. We call this process as image segmentation.Image segmentation has been a hotspot in the image engineering. There have been thousandsof algorithms so far. In this paper, the method based on graph theory is chosen to research, researchcontents are as follows: In the first place, research background, significance and research status ofimage segmentation is introduced; In the second place, image preprocessing methods before imagesegmentation are presented, as the result of image pre-processing has directly effect on the laterimage segmentation quality, improved methods are highlighted in order to improve the quality ofpretreatment in this paper; In the last place, the proposed methods are described in details: firstly, amulti-layer pyramid model diagram is constructed based on image preprocessing. Secondly,construct association model based on semi-supervised learning approach and obtain the similaritymatrix. Finally, using normalized segmentation criteria complete image segmentation.In this paper, the proposed image segmentation framework is experimented on Berkeley imagedatabase and MSRC image database. The results show that segmentation quality of this method hasa certain improvement compared with some traditional, classical methods.