Research On Shape Matching And Its Application In Paper-Cut

Image is an effective visual form of information and has its wide application in every field of society. The shape, which can be regarded as the contour of objects in image, is not only a basic intrinsic characteristic of objects, but also the important feature for recognition of objects. As a part of foundation research contents of computer vision and pattern recognition, the shape matching question includes shape description and shape matching. Its technology can be applied to object recognition, content based Retrieval and reconstruction of medical image, etc. Shape matching is an application-oriented subject and most of its theories and technologies are closely related to practice. There are almost no general theories and technologies of shape matching. Research on the technology of shape matching for practice purpose can not only solve the problem, but also enrich the technology of pattern recognition. Therefore, the research of shape matching is an important and significant work..In this paper, based on the research on shape description and shape matching methods, a new shape descriptor and some matching algorithms are proposed and improved. Theoretical analysis and experimental results have proved that the proposed methods are feasible and effective. The application of these methods in paper-cut image retrieval system is also completed. The main work in the thesis are as follows:(1) Research on the representation of new shape descriptor.Shape description is the foundation of shape matching, directly affecting the accuracy of matching results. Through the analysis of classic shape description methods, a new descriptor based on geometry features is proposed. It is the included angle between the line segments connecting center of gravity and two endpoints of major axis. The proof of the new shape descriptor being a invariant to rotation, scale and translation is proposed and the compared experiment results are given. Theoretical analysis and experiments both show that it is efficient and simple, and gains desirable matching results.(2) A paper-cut image matching algorithm based on bounding circle.Considering the simple shape and sharp corners of paper-cut patterns, we propose a bounding circle matching algorithm based on corner feature. The target shape is described by the relative position of boundary corners and region center of gravity. The corresponding relations between corners in two images are determined by whether they fall in the same bounding circle. The similarity measuring method is given at last. The experiment proves that this algorithm is invariant to rotation, scale and translation as well as suitable for the matching and retrieval of simple paper-cut images.(3) A paper-cut image matching algorithm based on geometric hashing.As for the matching of complicated paper-cut image, after the analysis and research on geometric hashing method, we choose two feature points as point-basis and improve the computation of invariant coordinates to reduce computation of method. And we construct hash function and linked hash table to resolve the address conflict problem effectively. The algorithm has good ability to handle the matching of paper-cut images under deformation or only part of them are of similarity.The experiment proves that this algorithm is suitable for the matching and retrieval of complicated paper-cut images and achieves satisfying retrieval results. (4) The design and implementation of paper-cut image retrieval system based on the matching methods proposed.Paper cut-out is a traditional folk art with a long history in China. We build the shape feature data base of paper-cut images, design and implement a paper-cut image retrieval system based on the matching methods proposed. The system has provided many kinds of retrieval types and realized effective management to the paper-cut image data base. Images used in experiments are selected from practice and the paper cut-out images are acquired by scanning the books. Great deals of experiments are made to confirm the validity and usability of this system.The results show that:(1) The new shape descriptor proposed is invariant to rotation, scale and translation, and gains desirable matching results for irregular graphs.(2) The bounding circle matching algorithm proposed is invariant to similarity transformation, producing accuracy and effective matching or retrieval results of paper-cut images under deformation.(3) The computation of improved geometric hashing methods is reduced and the address conflict problem is resolved. It can achieve high accuracy for complicated paper-cut image retrieval.(4) The paper-cut image retrieval system based on the matching methods proposed has strong usability and simple operation. It can realize the fast retrieval of paper-cut images as well as effective management of paper-cut image data base.