Research On Image Segmentation Algorithm Based On Clustering

Image segmentation is an important research topic in image processing. It decides the final result and quality of image analysis and image understanding. Because of the importance of image segmentation, many scholars at home and abroad have done a lot of researches on image segmentation and all kinds of different image segmentation algorithms are proposed. However, many of the proposed image segmentation algorithms are for the specific study object. The versatile segmentation theory has not been proposed by far. Therefore, people still continue to explore new segmentation algorithm and new segmentation theory.Fuzzy C means algorithm is an algorithm which has the most perfect theory and the most extensive application in clustering segmentation algorithm based on objective function. It can retain more of the original image information because the fuzzy concepts are successfully introduced into the pixels of membership. Although fuzzy C means algorithm is applied and developed widely due to its advantages, it is needed to determine the number of classification beforehand, is noise-sensitive and is easy in local optimum when it is used for image segmentation.As an effective clustering algorithm, mean shift algorithm does not need any priori knowledge and is analyzed entirely depending on the samples in feature space. In recent years, mean shift algorithm has been widely applied in computer vision field such as image segmentation and tracking. However, its convergence is too slow to be practical and the selection of bandwidth has great impact on the performance of mean shift algorithm.In view of the above consideration, the noise immunity performance of conventional fuzzy C means algorithm is improved by introducing context information into conventional fuzzy C means algorithm, and conventional fuzzy C means algorithm is improved by using the swarm intelligence algorithm. Mean shift algorithm is improved by using the nearest neighbor algorithm, correlation comparison algorithm, hybrid PSO algorithm and conjugate gradient algorithm respectively. Finally, the fuzzy C means algorithm and mean shift algorithm are studied and improved from the angle of transform domain. Details are as follows: Fuzzy C means algorithm based on space constraints is proposed by introducing context information into traditional fuzzy C means algorithm. The noise immunity performance of traditional algorithm was improved by amending the membership function and improving the distance measure. The velocity of convergence of fuzzy C means algorithm based on space constraints algorithm is improved by amending the value of membership.Swarm intelligence based fuzzy C means algorithm is proposed by use of global and robustness of particle swarm optimization algorithm and differential evolution algorithm, which avoids the traditional fuzzy C means algorithm sinking into local extreme. Moreover, the problem of setting initial value and providing the number of categories is resolved by the swarm intelligence algorithm.The application of mean shift algorithm in image segmentation is explored and conjugate gradient based accelerated mean shift algorithm has been proposed. Conjugate gradient algorithm is characterized by simple and better speed of convergence. The properties of conjugate gradient algorithm are used by conjugate gradient based accelerated mean shift algorithm. The speed of convergence of mean shift algorithm is improved by interleaved execution of mean shift algorithm and conjugate gradient algorithm.The adaptive mean shift algorithm has been proposed by use of nearest neighbor algorithm, correlation comparison algorithm and hybrid particle swarm optimization algorithm respectively which is used for solving the problem of bandwidth selecting. Mean shift, like other gradient ascent optimization methods, is susceptible to local maxima, and hence often fails to find the desired global maximum. In order to avoid sinking into local extreme, the mean shift vector is firstly optimized using hybridized particle swarm optimization algorithm. Then, the optimal mean shift vector is updated using mean shift procedure.Finally, the noise immunity performance of traditional clustering algorithm is improved from another angle by using Contourlet transform. The image containing noise is firstly denoised by Contourlet transform, secondly, the denoising image is segmented by clustering algorithm. Experimental results show the validity of new algorithm.The image segmentation algorithm based on clustering is studied and explored in this paper. The traditional algorithm is improved by using various means. Experimental results on different images show the efficiency of the proposed algorithms.