Design Of Target Recognition System Based On Spatial Relationship Of Contour Segments

In this thesis,the target recognition based on contour segment spatial relationship is studied.The main contents are divided into image preprocessing and contour extraction,rough matching based on image contour and target recognition based on contour segment spatial relationship.First of all,this thesis studies and experiments on the extraction technique of target contours.In this chapter,the target contour extraction is divided into two stages:preprocessing and contour detection.The algorithms of gray level,filtering,.edge extraction,expansion and hole filling are introduced in detail.Finally,the extracted contour is redrawn,and the outer contour of the image target is extracted.The extraction result can meet the needs of subsequent work.Secondly,the target matching technique based on contour shape features is studied.This chapter mainly studies the target matching based on Hu image invariant moment and shape matching based on shape context.The Hu image invariant moment is mainly calculated by calculating the Hu invariant moment of the target contour region,and then calculating the geometric distance between the invariant moments of the two images,and finally obtaining the matching result.The shape context mainly obtains a matching matrix by matching the two shape description matrices,and solves an optimal solution.This optimal solution is the matching result of the shape context.Finally,the results of the two matching algorithms are compared and analyzed.Thirdly,the image matching based on the contour segment spatial relationship is studied.Because discrete contour segments have certain matching errors and disadvantages when calculating the above descriptors,the spatial relationship of discrete contour segments is studied.Different methods are adopted to determine the spatial relationship parameters for the two different image formats encountered.:For the image of the captured occlusion,only the contour segment of the image can be obtained after image preprocessing,so the centroid position of the contour segment is used to express its spatial relationship parameter.For the complete image in the database,using the method of obtaining the image skeleton,the key position points on the image skeleton are used to represent the spatial position parameters of the image.Finally,the spatial relationship parameter constraint method is used to obtain the final matching result.Then the above algorithm is integrated,and the graphical interface programming is completed by C++ combined with Qt,and the target recognition system is designed to be transplanted and used.Finally,the work completed in this topic is summarized,and the content of the follow-up work is prospected.