Research On Shape Feature Extraction And Matching Based On Skeleton

Shape matching is one of the hotspots in computer vision and pattern recognition, skeleton as an excellent representation method of shape topology, it represents the main visual characteristic shape of an object, and simplifies the shape of the object representation, shape matching based on skeleton is widely used in different areas, such as target recognition, shape retrieval, medical image processing and augmented reality. But the current skeleton methods exist some problems, such as poor immunity, large computation, match results still further improved and so on.To solve the problem of connectivity and single-pixel, the paper uses the skeleton growth model to extract shape skeleton, and to reduce time complexity by improving the skeletal growth model. For the problem of the current skeleton extraction methods are susceptible to boundary noise and deformation, the paper proposes a method for pruning the skeleton, based on building hierarchy skeleton, evaluating the significance measures and pruning strategy. In order to improve the skeleton recognition performance, the paper presents a skeleton matching method by first matching the skeleton end points between two skeleton graph, then matching skeleton branch points on the basis of the matching end points, thereby obtaining the similarity skeleton.Experimental results show that the proposed pruning method can effectively remove redundant branches, for a same threshold value, the method has better adaptability and stability than the current major pruning method, however, the method also has the problem of shortened skeletal branches trimmed, which will affect the shape reconstruction based on skeleton. The proposed skeleton matching method has good stability in the process of non-rigid body shape matching with the deformation and contour noise, and the proposed method compared against other methods on the recognition performance has good results.