The Research Of Meaningful Region Segmentation Technology

Color image segmentation for meaningful region is one of the most important step in computer vision. Useful information will be supplied for contend-based image retrieval and object analysis by proper method. Thus, real time image understanding and image analysis could become possible.Centred on how to extract the meaningful region of color image, thesis made detailed research. By summing up the existing algorithms, and connecting with the application backgrounds, take homogeneity of pixels as preexisting, two kinds of meaningful region segmentation technologies was put forward.Main work of this paper is mainly focused on these two aspects:1. A new improved fuzzy C-means algorithm is proposed to apply in extracting meaningful region of color image. When operating the new algorithm, gaussian mixture density model was used to compute the centre (mean value) of a color image; the theory of cross-entropy distance was introduced to compute the distance . Experimental results show that, the new algorithm not only improves the initial clustering center, but also enhances the segmentation precision and the universality principle.2. We adopt a number of classical powerful algorithms (mean shift clustering, edge detection and region growing) to extract the meaningful regions adds spatial information. The experiments indicate that the proposed method can avoid the phenomenon of un-homogeneity and the results can be easily accepted by human eyes, also, edges and boundaries connected and closed.The algorithm is proposed to extract the color image, but the essential theory is also the same with other types'image or datas. It can be conveniently applied in other image by modifying a small quantity of parameters.