Study Of Shape Description And Matching Based On Chain Code And The Shape Context

With the development of science and technology, We are surrounded by largeinformation, the digital image carries a lot of information as the visualization systeminformation.In the age that computer has an irreplaceable role, how to use computer toidentification and automatic retrieval digital image so to finish the computer vision,which became the focus of the researchers. During the matching of digital images, wealways describe the image features as the number or symbol (as descriptiors). Thenthrough the certain similarity rules, to automatic matching and retrieving image.The paper studing the method and the matching of image description, especially onthe shape contour, the main research contents include the following directions:First, introduce the process and the basic properties of shape matching, at the sametime; introduce the commonly used algorithm in each shape matching step.Second, based on the commonly used Freeman chain code; we find a way toimprove the Freeman chain code which using the statistical features of Freeman chaincode and the characteristics of block entropy.We present a new approach: fusion theMinimum Sum Statistical Chain Code and Entropy Matrix Singular Value to describeshape contours. The method combines the contour and region-based description, whichuaing the advantage of chain code and singular values of entropy matrix. It is geometricinvarianceat, at the same time; describes the distribution characteristics of theshapespace. Then apply the described algorithm to MPEG-7standard database, theresult is good.Third, take the shape contours as a contour points set; we study the shape contextalgorithm; in shape context, it is using the position relation.Then we find a newapproach to adaptive selection context feature points.The new approach overcomes theshortcomings of shape context when chose points. The shape context is time-consuming. When the shape is simple we could select little points and when it is complex, wecouble select many points. By the end, we use the new approach in the MPEG-7standard shape library.The retrieval result proves this algorithm is good.Chain code and shape context descriptor is very typical of the contour shapedescriptors; they have a practical significance in the image retrieval system.