The Study And Application Of Evolutionary Clustering Algorithm Based On Manifold Distance And Kernel Function

Clustering is an important data analysis method, and it is widely applied to many domains, such as Computer Vision, Information Retrieval, Data Mining and Pattern Recognition, and so on. As a global optimization technique, Evolutionary Computation has been used for clustering by many scholars now. The clustering algorithm always uses the Euclidean distance as a similarity measure. Although it is greater than the conventional K-means algorithm based gradient descent in global optimization performance, it performs better only on the spatial distribution of the spherical or super-sphere of the data, and bad on data with a complex distribution structure. It is a certain result cause by Euclidean distance. Therefore, using a proper similarity measure function to make the algorithm get more accurate results for complex distribution of data is important.In this paper, firstly we propose a novel evolutionary clustering algorithm based on kernel function. By using kernel functions, we can map the data in the original space to a high-dimensional feature space in which we can perform clustering efficiently, and we call it Kernel Evolutionary Clustering Algorithm (KEAC). It can identify and extract the useful features better by using non-linear mapping, and more accurate in clustering. In the failure of classical clustering, KEAC also can have a good perform.For data with a complex distribution structure, we propose a novel evolutionary clustering algorithm based on mixed measure (MMECA). This method uses the idea of classifying roughly first and classifying precisely then. We use the evolution clustering algorithm that measured by Euclidian distance for rough classification of data sets first, and use the evolution clustering algorithm which measured by manifold distance for precise classification. Experimental results on non-supervised classification problems with a complex spatial distribution show that MMECA has a high accuracy in clustering and have a good robustness.Combining with morphological method, a novel image segmentation algorithm based on Watershed and Kernel Evolutionary Clustering Algorithm is proposed. Firstly an improved watershed algorithm, marker driven watershed transform was used to segment image into many small regions. Then, the Image Characteristics of every small region is calculated. Secondly, by using kernel functions, we can map the Image Characteristics in the original space to a high-dimensional feature space in which we can perform clustering efficiently for the next image analysis. Finally, based on these two segmentation results, final results are obtained. The new algorithm is used to solve different image segmentation tasks, including natural image, texture image and synthetic aperture radar image. The experimental results show that WKEAC is competent for segmenting multiple images with high quality.