Improving Image Segmentation Quality Via Graph Theory

Image segmentation is a fundamental process in many image, video, and computer vision applications. It is very essential and critical to image processing and pattern recognition, and determines the quality of final result of analysis and recognition. This paper presents a semi-supervised strategy to deal with the issue of image segmentation. Each image is first segmented coarsely, and represented as a graph model. Then, a semi-supervised algorithm is utilized to estimate the relevance between labeled nodes and unlabeled nodes to construct a relevance matrix. Finally, a normalized cut criterion is utilized to segment images into meaningful units. The experimental results conducted on Berkeley image databases and MSRC image databases demonstrate the effectiveness of the proposed strategy