A New Algorithm Of Image Segmentation Based On Local Feature Classification

With the progress of science and technology, the information age has arrived. People get information from image,because image is an important way to obtain information、express infortation,that is the window of people apperceived the world, image processing has been rapidly developed.Image segmentation is a very important center links for image processing,also get attention. In recent years, various image segmentation algorithm has been discovered. But so far, it hasn’t appeared a segmentation algorithm to have good segmentation results for all image. In image processing, image segmentation results will directly affect the quality of image processing. Therefore, image segmentation technology has been widespread concern in the academic community, medical community and industry etc.. So in the computer vision field, the research of image segmentation technology is not only the challenge, but also the opportunity.So far, based on the existing and obtain good segmentation results segmentation method, this paper aimed at image segmentation is proposed the research of quaternion PHT and the twin support vector machine(TWSVM)、Nonsubsampled shear wave and the hidden Markov tree(HMT) model, the major work completed a few points as follows:1. Using quaternion PHT extract image pixel features, and then ACM and FCM combined to select the training samples, The use of twin support vector machine trained on the training sample, and finally the use of twin support vector machine classification. The experimental results show that it is better to keep contact with each other and correlation between components of the image, improve the segmentation accuracy.2. Using the correlation of the non-subsample Shearlet transform coefficient between the scales, between the direction,and in the scale to build hidden Markov tree model for image segmentation.Firstly image by non-subsample Shearlet transform is obtained coefficient three kinds of relations between scales, neighborhood and cousin system, then using the model to describe the statistical correlation between adjacent coefficients, so as to realize the image segmentation. The experimental results show that, by using the hidden Markov tree model coefficient variety relation, can better segmentation for the edge.