Research And Application Of Image Segmentation Based On Tensor

With the development of Data Acquired Technology, so many data set manifold distribute in computer vision field, tensor and tensor subspace technique are applied in many kinds of segmentation methods for get better results. The main job of this paper as follows:At first, this article presents a local- and global-statistics-based geometry active contour model for image segmentation. We first propose a energy functional with a local Gaussian distribution fitting term and an global intensity fitting term. A weighting function that varies dynamically with the location of the image is applied to adjust the weight of the two terms dynamically. At the same time, a new termination criterion for curve evolution is proposed to forbid more useless iterations.To avoid active contour curve oscillation,(we)choose small time step for active contour curve when it evolutes to the edge of the targets. As a result of the introduction of the Split Bregman iteration scheme,we could use a rather large time step while maintaining the stability of the evolution process. The split Bregman iteration scheme can accelerate the evolution of the curve and significantly reduce the number of iterations.Active contour model is based entirely on mean value of local intensity,which is difficult to segment the texture image with full of structure informatio. In this paper, we combine gray information in structure tensor, as this way, the image teature is not only with grey information but also structure information. Weaken ability in image segmentation technique is a difficult task that the researchers have to survive. In this paper, we extract tensor feature then divide them into two categories by Kmean (k=2): one with the stronger discriminative ability and the other with weaker ability. The second category is optimized by Principal Component Analysis to better distinguish the objects with low contrast. The image feature combines with the stronger discriminative ability and the optimized by PCA.The article presents a semi-supervised fuzzy clustering algorithm based on harmony search algorithm and hybrid difference algorithm. It improve the optimizing performance by correcting improvisation in harmony search algorithm, leading in variation, intersect and selecting operation of difference algorithm. Secondly, we present a new Objective function for optimizing the center of clustering. Thirdly, we define a new distance function to optimize the initial center of clustering. The experimental results demonstrated the validity of our method. Moreover, it shows that the proposed algorithm avoids the problems of slow convergence and degeneracy to classical FCM algorithm when applied to real world data clustering with exiguous labeled data, and presents its effectiveness for the application in interactive segmentation of images with a small amount of labeled data points given by usera symmetric tensor field is used to identify the global and local image structure. The orientation tensor field is used to control the size, shape and orientation of the filter. The behavior of the filter can be enhanced, which is accordance with the orientation of tensor in every small local region. On contrary, the behavior of the filter can be weaken, which is not accordance with the orientation of tensor in every small local region. It is applied in 2D and 3D image with noise.The experimental results demonstrated the validity of our method.Structure tensor is used to extract the image feature for it is a good descriptor for local neighborhood information of the image. So we improve cost function based on structure tensor in Live wire algorithm. At the same time, we find the shortest path by harmony search algorithm and hybrid difference algorithm which make the computing speed and accuracy improved.