The Research On Fuzzy Clustering Alogrithms With Noise Immunity And Its Application

Depending on the difference between the degree of abstraction and processing methods, image technology can be divided into three levels:image processing, analysis, and understanding. Image processing is an operation of the lower layer and mostly operating on the image pixel level. Image noise reduction, image coding and image segmentation is more representative in image processing techniques.Image segmentation occupies a very important position in the image project. Its essence is refers to segment the image into different regions, each region has its unique characteristics, then extract the target of interest. Application of image segmentation is widely used to deal with practical problems such as machine vision, pattern recognition and traffic control systems. Because of the image is affected by every%ind of objective factors in the process of imaging, the image itself exists varieties of uncertain problems. Fuzzy theory can deal with the uncertainty problem. In recent years, based on the theory of fuzzy image segmentation methods obtained lots of attention and research.As a basis for subsequent image analysis and image understanding, Image segmentation is a very important technology of image processing. However, image segmentation techniques are one of the difficult problems of the development of image theory all the time. At the same time, with the increasing of image segmentation field, the study on the image segmentation algorithm has more significance.For the existing fuzzy clustering algorithms, This paper is mainly analyzed the existing problems of these algorithms, and obtained improved image segmentation algorithms by researched and experimented. The main work is as follows:1. In order to improve the two-dimensional histogram fuzzy C-means clustering segmentation algorithm noise immunity and universal, dimensional histogram weighted fuzzy C-means clustering segmentation method is proposed. For two-dimensional histogram fuzzy C-means clustering segmentation algorithm exists a threshold parameter selection leading poor noise immunity, this paper introduced the properties weighting to two-dimensional histogram weighted fuzzy C-means clustering segmentation method, an effective solution for each dimension attribute poly problem class contribution. Compared to the two-dimensional histogram algorithm, salt and pepper and Gaussian noise immunity performance of the proposed algorithm increases average 2 to 3db; at the same time, compared to the fuzzy local information C-means clustering segmentation algorithm, salt and pepper and Gaussian noise immunity performance increase average 2 to 3db and anti-Gaussian noise less about 1 db, but the speed of the fuzzy local information C-means clustering segmentation algorithm is about 40 times of the proposed algorithm. The results show that:compared to existing two-dimensional histogram fuzzy C-means clustering algorithm more suitable for noisy image segmentation needs. At the same time, compared to the local fuzzy C-means clustering algorithm is more conducive to real-time requirements of occasions target tracking and identification needs. At the same time, a large number of tests proved that the proposed algorithm is suitable for the synthetic images, intelligent traffic images and remote sensing image.2. For the robust local fuzzy C-means clustering algorithms proposed by Krinidis etc, the iterative formulas of their clustering centers lack rigorous mathematical theory. To solve this problem, by rigorous mathematical analysis and deduce to obtain membership and clustering center expressions. The new robust local kernelized fuzzy C-means clustering segmentation algorithm is presented.Experimental results indicate that the algorithm is effective and suitable to segment complex remote sensing image.3. A fast fuzzy local information C-means clustering algorithm was proposed because the fuzzy local information C-means clustering algorithm was time-consuming. In this algorithm, co-occurrence matrix was introduced which constituted by target pixel and its neighboring pixels, the new cluster membership and the cluster center expressions were obtained. To improve the noise immunity of algorithm, filter processing to neighborhood pixels membership when pixel was classificated. The experimental results demonstrated that the proposed algorithm meets the needs of the effectiveness of image segmentation, Compared with fuzzy local information C-means clustering algorithm, the proposed algorithm has advantages of segmentation performance and save time.4. For Fuzzy c-means clustering with local information and kernel metric for image segmentation algorithm time complexity too large to fit in real-time occasion required, a fast nuclear space algorithm of fuzzy local information c-means clustering segmentation is proposed. Firstly, using the space distance between pixels neighborhood pixels with information and gray variance information construct a weighted co-occurrence matrix, Secondly combined one-dimensional histogram with two-dimensional histogram between the pixels and neighborhood pixels constructed objective function of the new algorithm.Thirdly, to improve the algorithm of noise immunity, filter processing to neighborhood pixels membership when pixel classification. The experimental results shows that compared with KWFLICM, the proposed algorithm has advantages of better performance and save time.