Research On Extraction And Matching Of Linear In Image Sequences

Using some features to link the two image of the same scene is a basic task of theimage matching. In real life, linear features often appear as the main features of thevisual image of human, such as runway and object contours. Therefore, the matching ofsequence image often uses linear features as an important analysis of primitives. Thispaper concentrated on theories and methods how to extract the straight line from theimage and how to match for it from image sequence. The study includes:(1)Using the different sensors to obtain image sequence, it often brings the noise.So it is necessary to filter the image to get a better quality image. The paper discussedseveral image filtering methods, and gave a suitable method for our subject. Also itintroduced and compared with several classical edge detection algorithms in the paper.(2) Hough transform is one of the classical methods of straight line extraction. Inthis paper, it discussed and analyzed Hough transform for linear detection, and wemainly discussed randomized Hough transform for extracting straight line. To solve theproblem of inefficient sampling and accumulation in RHT, an improved Random Houghtransform algorithm is proposed as follows: Firstly, it randomly took two points fromimage space, and then judged whether the two points were content with a certaindistance so as to obtain the parameters space, at last, determined whether the line wasexists or not. The theoretical analysis and experiment show that the proposed algorithmhas low memory and fast speed.(3)In matching line segment, this paper discussed how to describe the straight lineand how to description the straight line. And it studied the method based on geometricproperties of straight-line. Firstly, it used linear gradient direction, gradient size, linesegment length and midpoint location for rough matching, and then it extracted the linesegment midpoint of the local gradient information to create an Mid-point descriptor，soas to match the line segment precise. This did the experiment with artificial images andreal images at last. The experimental results show that the proposed algorithm has goodmatching performance.