Research On Fuzzy Clustering Algorithm In Pattern Classification

Pattern classification is one essential foundation of machine learning, artificial intelligent and many other research areas. Since pattern recognition existed, there are many methods used to solve the problems of it, however, many classification problems have no clear boundaries. Therefore, people introduced fuzzy theory into clustering methods, and fuzzy clustering algorithms are used to solve pattern classification problems successfully. In this paper, we mainly did research on how to improve one common fuzzy clustering algorithm (Fuzzy C Means), and its application.We first introduced some knowledge of fuzzy clustering, and then conclude some usual clustering algorithms. From classifying high dimension data, we compared the cluster results and performance of Hard C Means and Fuzzy C Means (FCM), and found the different and relationship between two algorithms. Hoping to make non-linear cluster problems easier, we introduced kernel based fuzzy clustering method. By compare it with FCM, we conclude its advantages and disadvantages.One of the most popular applications of FCM is image segmentation. Considering image space neighborhood information and different contributions of every pixel, we brought in spatially weighted FCM. Indicating the affect which initial cluster center made to final results, we used histogram of image to improve the method to choose initial cluster centers. From simulation, we found out that improved the algorithm had better segmentation performance than traditional FCM, and also indicate it reduced the calculation quantity for every iteration time and its resistance to noise had developed.According to different segmentation results getting from different color spaces, we chose HSI color space for it is more closed to human eyes. By changing RGB image by some equations, we got H, S, I components. Using clustering result of improved spatially weighted FCM, we got one new color image segmentation feature. The simulation result confirmed the good segmentation performance of this method.We all know that image segmentation is one important basement of machine vision, so we introduced consist of general machine vision system in this paper and built a simple hand vein image capture system. After some preprocessing of hand vein images, we used a combination of fuzzy algorithm segmentation and edge detection to get hand vein feature. At last, we indicated the improvement of method proposed in this paper.