Research And Engineering Application Of Image Segmentation Based On Probabilistic Graph Model

Image segmentation is a basic research task in the aspect of an image processing and a computer vision, and it is the key element to analyze and understand an image.Image segmentation methods based on probabilistic graphical model provide a solid theoretical basis for solving the problem of a lot of uncertainty existing in the image information processing. It provides a very comprehensive theoretical framework,therefore, so that a mass of scholars have studied deeply and made it been developed rapidly. It has been widely used in the fields: medicine, computer vision and remote sensing image processing etc. The early image segmentation methods based on the low level features of the image can not accurately represent the target object. However, the image segmentation methods based on probability graph model can combine the high level semantic information and the low-level features of the image, and it describes perfectly the characteristics of a target object. But the structure of the models are more and more complex, the number of potential energy function items and the variable and parameter is increasing respectively. It makes parameter training more and more difficult and the efficiency of minimizing the energy function lower and lower. To solve the problem, image segmentation methods based on probability graph model are systematically studied in this paper.The main contents and innovations of this paper are as follows:(1) The research status and existing problems of the image segmentation methods based on probabilistic graph model are summarized and analyzed in detail. The basic theories of the image segmentation methods based on probabilistic graph model and two kinds of probabilistic graph models are introduced. The design steps of the image segmentation methods based on probabilistic graph model are also introduced.(2) An efficient image segmentation algorithm based on the robustnP Potts high order CRFs model is proposed. The first step of the improved image segmentation algorithm is the calculation of the local optimal solution of the image, which provides an initialization for the ?-expansion algorithm for the complex structure and many parameters of the robustnP Potts high order CRFs model. Then, the graph is updated dynamically during each iteration of the ?-expansion algorithm. It not only solves the problem of low efficiency of image segmentation due to the complexity of the model,but also effectively avoids the over segmentation of the image.(3) An image segmentation method based on the deep higher order CRFs model is proposed. It consists of pixels, edges of adjacent pixels and segmentation blocks(super pixels). The model based on robustnP Potts higher order CRFs model introduces the TextonBoost potential for unary energy function which makes unary energy function can betterly represent the features of the image, which also uses the cooperative cut potential function as pairwise energy function. The model gets more precise object segmentation whether it's in the case of two labels or multiple labels.(4) The production and processing of areca is the application background. The image segmentation methods based on probabilistic graph model is applied to the production and the processing of irregular objects. It realizes the adjustment of the position of the irregular object, so as to achieve the effect of automatic processing and improve the automatic level of machine in the production and processing of irregular objects.