Line Feature Extraction From Image Based On Stochastic Point Process

Line feature is line segment in image, which can sketch out the linear objectives, contour or structural information of the object in image, and during image recognition, that is the important property extracted from images. Line feature extraction is to extract the significant lines or useful contour in image, which in order to strengthen the objectives concerned by people and make it possible to distinguish the content and some properties of the objectives. In the process of image analysis and image understanding, people mainly relying on the information of the objectives to recognize and judge the objective. Therefore, the studying of image line feature extraction has important theoretical significance and practical application value.Based on stochastic point process, this thesis studies the line feature extraction algorithm of image which regards the whole network segment as an object and the line feature in image is composed of network segments. This method makes full use of the line features' gray information and the structural relationship among segments, and is not sensitive to noise. Initial status doesn't affect the final extraction results. If the initial temperature is high enough, it also can get the global optimal solution. The main content includes:(1) An improved model of the line feature extraction is proposed. Candy model is a line feature extraction model based on stochastic point process. The relationship of the line feature connection and adjacent are redefined, Candy model is improved and the quality of the line feature extraction is improved based on the mechanism analysis of candy model line feature extraction.(2) The model algorithm RJMCMC is studied. Birth-death proposition kernel based on connection, proposition kernel based on segment network after initialization and the line segment network initialization based on Hough transform are given, which improve mobile rules of the line segment. Sometimes, the parameters'selection problem of the model is discussed in this thesis. Experimental results show that the improved model is a better method which can extract the line features of the image and accelerate the computing speed.