A New Algorithm Of Color Image Segmentation Based On Statistical Modeling

In the field of computer vision, image segmentation is still a key technology, and its purpose is to segment the special significant area in the image. Image segmentation plays a crucial role as a preliminary step for high-level image processing, for example, objective localization or recognition, data compression, image retrieval and so on. In the past few years, many image segmentation methods have been developed, such as threshold segmentation, segmentation based on region, etc. So far, because of the complexity of image there are no standard segmentation methods suitable for all different kinds of images. Therefore, image segmentation technology is still one of the hot researches at present.Recently, image segmentation combined with all kinds of theory has been applied to different kinds of images and made good results. This paper around the field of image segmentation studies carefully the theory of support vector machine and statistical modeling and then proposes my own segmentation algorithms, which can be completed as follows:1. This paper studies the theory of support vector machine and then combines Fuzzy C means or Arimoto entropy threshold with support vector machine for image segmentation by using the extracted color and texture features. This algorithm supplies the training sample for support vector machine, which not only is suitable for the human visual perception characteristics but also saves time and obtains good segmentation results.2. When describing the features, this paper is no longer based on the original features but uses the new features obtained by Gaussian mixture model for initial segmentation, and then gets the final segmentation results by classification of support vector machine. This algorithm makes full use of the spatial information of image to make up for the shortcomings of complex background of Gaussian mixture model which only uses the time domain information, and improves the accuracy of segmentation effectively.3. In this paper, an image segmentation algorithm with PDTDFB transform and enhanced hidden Markov tree model is proposed. First extract the features of image, and then transform the features with PDTDFB, finally combine PDTDFB coefficients with enhanced hidden Markov tree model for image segmentation.