Research On Improved Support Vector Machines Algorithm And Its Application In Image Segmentation

Support Vector Machine(SVM) is a novel powerful machine learning method developed in the framework of Statistical Learning Theory(SLT). SVM solves practical problems such as small samples, nonlinearity, over learning, high dimension and local minima, which exit in most of learning methods, and has high generalization. Being the optimal learning theory for small samples, SLT and SVM is attracting more and more researcher and becoming a new active area in the field of artificial intelligent and machine learning. SVM has been well applied in many areas, however, the theory research and application of SVM are not very mature, and there are still many problems to be solved.Image segmentation is the precondition of image processing, and it is also the classic problem of image processing. Image segmentation is of partitioning the image into several areas with the similar characteristics (Like gray level, spectrum, texture, etc) and finding the interested part. Therefore, image segmentation can be seen as the process of classifying the pixels with certain characteristics essentially. SVM is an excellent method of learning and classification in both theory and application. SVM algorithm with high accuracy, fast computational speed, characteristic of robustness and strong generalization ability make the SVM suitable for image segmentation.This dissertation research some improved SVM algorithms and its application to image segmentation. The main works of the dissertation can be organized as follows:1. Firstly, summarize the research progress of SVM in resent years and its applications and drawbacks. Introduce the basic concept and methods of image segmentation.2. Present the basic theory of SLT and SVM in detail. And review the basic algorithm and improved algorithm of SVM. Then summarize the characteristics of SVM method.3. Study on SVM multi-class algorithm based on sample pre-selected. Based on membership function and rough set (RS) theory pretreatment, a new method of fuzzy support vector machine (FSVM) multi-class algorithm is researched. Validate the availability of algorithms with several test databases.4. An improved least squares support vector machine based on sparseness method and hierarchical clustering algorithm is researched. Compared with several basic methods on image segmentation experiment, results indicate the algorithm can reduce the time of training and testing and improve the effect of image segmentation. Use Particle Swarm Optimization (PSO) algorithm of a powerful global search capabilities to optimize Fuzzy C-Means clustering(FCM)clustering centers. Then use improved FCM to obtain feature vectors and labels of images for training SVM. Then automatic image segmentation is performed using FSVM classifier. Compared with the traditional method, the algorithm achieves better effect.