The Research Of Natural Scene Image Segmentation Based On Graph And Clustering

Massive natural scene images require fast retrieval and analysis. Images self- identification and filter are the very urgent tasks for national defense fields, and image segmentation is the key to complete them well. Natural scene images'indetermination, complex background, non-uniform resolution and target object in natural image is uncertain. Therefore, the natural scene image segmentation is difficult point and hotspot in area of image segmentation. This dissertation takes natural image segmentation as its subject, do researches further on the popular image segmentation based on graph theory and clustering. Mass experiments of image classification and image retrieval were done using regions be segmented. Completed the achievements as follows:1. For N-Cut algorithm is inefficient and Efficient Graph-Based Image Segmentation is sensitive to strong edges in images. These classical graph-based image segmentation algorithms are not robust to segmentation of texture image. In this paper, we propose a novel segmentation method that Graph-Based Color-Texture region Segmentation, which overcame shortcoming of existed graph-based segmentation method. It extract feature vector of blocks using color-texture feature, calculate weight between each block using the neighborhood relationship. The experimental show that new method is robust to segment texture image and strong edges image. And more efficient than N-Cut algorithm.2. Pre-specified number of categories, the initial classification center is one of the issues can not be avoided to clustering-based image segmentation algorithm. In view of this, the paper presents Adaptive Affinity Propagation clustering image segmentation algorithm. It adaptively calculates preference in Affinity Propagation clustering using the integral characteristic of image and applies this method to image segmentation. The results of experiments demonstrated that the new method segment image more accurate than classical K-Means and FCM clustering method.3. In order to apply adaptive AP clustering segmentation method to region-based image retrieval system, improving accuracy of the regional match in image retrieval system, the paper proposes a weighted regional matching. For increasing the weight of important regions, reducing the weight of the secondary regions, this method effectively improves the image retrieval accuracy.4. Clustering-based image segmentation algorithms lack spatial association information and graph-based methods can not deal with the problem of discrete regions. In order to combine advantage of two types of algorithm, overcome their shortcomings, we proposed a combine graph and clustering natural scenes image segmentation method. Experiment results of image segmentation demonstrated that the method is effective to avoid spatial constraints and regions be segmented is more smooth than using single one of method. Experiment of RBIR show that this algorithm is a feasible method for theory and practice.