Research Of Color Image Segmentation Algorithm Based On Wavelet And Graph Theory

Abstract:Image segmentation is an important branch of the field of image processing, It’s result has important meaning to image analysis, image understanding and computer vision, which belongs to the research on fundamental image processing. Graph theory has been widely applied in various fields, such as coding theory, stochastic process, reliability theory, computer programming, economics, genetics, and so on. In recent years, with the research and development of the technique of image segmentation, graph theory has been widely used in the field of image segmentation, a large number of graph based image segmentation algorithm were proposed, and achieved better results.This paper studied the application of graph theory in image segmentation。The main contents and innovations are as follows:Firstly, describing research actuality, then, discussing the application of graph theory in image segmentation, and the same time, deeply studied three kinds of graph based image segmentation algorithm, and analyzed the advantages and disadvantages to the three algorithms. At the same time, theory and application of three kinds of image segmentation evaluation algorithm was introduced, which were VOI, PRI and GCE, for objective evaluation of the image segmentation algorithm.Secondly, the graph theory based on FH(Proposed by Felzenszwal and Huttenloch, so the algorithm is called for short "FH algorithm" in the paper) algorithm for image segmentation is studied. Aiming at the shortcomings of FH algorithm, improved method is proposed:(1)The L*u*v*color space is adopted instead of RGB color space due to color more uniform distribution, and combined with the L*u*v*color space, the weighted function in the FH algorithm is redefined.(2)The idea that wavelet transform and FH algorithm combined is proposed, that is to say, the wavelet transform is used in the L*u*v*color space image, then the low frequency image as a weighted undirected graph of objects in FH algorithm. The effect of L*u*v*color space and wavelet transform’s levels in segmentation results were discussed and analyzed.Finally, three kinds of Algorithm evaluation method (VOI, PRI and GCE) were used to evaluate the improved algorithm. The experiment results show that the improved method depressed over-segmentation, received better segmentation results.