Research Of Image Segmentation Based On Fuzzy Clustering And Active Contour Model

With the advance of electronic computer technology, digital image processing as a new discipline has become an indispensable tool in the information society. Image segmentation is a fundamental topic in image processing, computer vision, object tracking, and medical imaging. Its main purpose is to partition an image into a series of sub-regions with homogeneous properties (intensity, color or texture, etc.), and thereby separate the components of interest from the background. Over the last few decades, researchers have made great efforts to solve the problems of image segmentation and have proposed many algorithms. However, image segmentation remains to be challenging due to the presence of noise, complex background, low contrast and intensity inhomogeneity. To improve the performance of image segmentation algorithm, scholars are still exploring and developing new image segmentation algorithm and theory for segmentation results with better versatility and more accuracy. Meanwhile, here lies the significance of this thesis topic.Fuzzy c-means (FCM) utilizes the least square error to measure similarity between sample points and cluster center. Iterative method is used to optimize the objective function, to achieve optimal clustering of image data. Because fuzzy belongingness is successfully introduced into clustering method, FCM can retain more information of the original image. The active contour model, based on partial diffenrential equation, is highly favored by many researchers for its free topology and flexible structure. By combination of low-level image information and high-level comprehension mechanism, it can obtain accurate segmentation results, better robust and practicality.This dissertation mainly discusses fuzzy clustering algorithm and active contour model algorithm in the field of image segmentation, and through various means to improve the original algorithm. Results obtained in this dissertation are as follows:(1) A novel fuzzy c-means algorithm based on the local coefficient of variation for image segmentation is presented. Firstly, to enhance the noise immunity of image segmentation, the local gray similarity matrix is modified by replacing each pixel with the median of the neighboring pixel values. Then, the algorithm reconstructs a new local similarity measure between pixels by introducing the local coefficient of variation, to control the weight between the center pixel and each of the neighboring pixels more accurately. Finally, the fast clustering is adopted to make the segmenting process only dependent on the gray levels of the image, which can improve the efficiency greatly. Compared with other segmentation algorithms. the proposed algorithm enhances the accuracy of image segmentation and has stronger anti-noise ability.(2) A level set image segmentation algorithm using local correntropy-based fuzzy c-means and its simplified model are proposed. Firstly, because the correntropy criterion has significant advantages in dealing with noise, we replace it with a squared error criterion to reconstruct the objective function of FCM_S algorithm, which can adaptively emphasize or decrease the weights of samples that are close to their corresponding cluster centers. Then, the proposed clustering algorithms are incorporated into a variational level set formulation, so that it can accurately classify pixels. Finally, the iteratively re-weighted algorithm and gradient descent flow method are adopted to solve our models. Experimental results show that our methods have the capacity of extracting weak edge and objects with intensity inhomogeneity and weakening the influence of noise.(3) In this paper, we propose a novel active contour model driven by local signed difference and local Gaussian distribution fitting energies. The evolution force of the level set function is defined as a liner combination of the local signed difference term based on local entropy and the local Gaussian distribution fitting term. Then, the contour can be driven to the objective boundaries by adopting gradient descent method to minimize the energy functional. The experiment results show that the proposed model can segment images with intensity inhomogeneity and is less sensitive to the size, location and shape of the initial contour when compared with the classical active contour models.(4) A level set method with improved initialization using region growing for hippocampus segmentation is developed. Firstly, adaptive region growing algorithm is used to gain a rough region of hippocampus. Then, morphological operations are applied to the results to eliminate internal spots. Thus, sequential hippocampal contour curve could be achieved through the application of contour tracking operator. Finally, with the contour as prior information, it can be driven to approach the target and stop at hippocampal boundary by using the improved level set method. Experimental results demonstrate that the proposed method can ideally extract the contours of the hippocampus that are very similar to manual segmentation drawn by specialists and have better segmentation accuracy and efficiency.