Research On Skeletonization Algorithms For Graphic Recognition

Skeleton is a representation of an object which has been shown to useful in interrogation, animation, finite element mesh generation, path planning, and feature recognition. The work of this paper is supported by the National Natural Science Foundation of China (No.60273099)——"New method study of 3D graphics recognition based on generalized conditional skeleton". The research about this foundation is mainly focus on skeleton extraction and shape similarity measure of object's skeleton. This paper first summarizes and compares the research about typical skeletonization algorithms, and discusses the influencing fact for skeleton recognition. Then, a novel discrete skeletonization algorithm based on distance map is presented in this paper. In this segment, a Skeleton-growing model is presented. This model chooses a skeleton point as a seed. Based on the criterion of skeleton point in discrete domain, the whole skeleton is depth-first searched out by growing from the original skeleton seed one by one. Meanwhile, the redundant skeleton branches are eliminated by Discrete Curve Evolution model, the visual branches are remained completely, the skeleton from this algorithm satisfy human vision. After that, a new algorithm in continuous domain for Medial Axis Transform of polyhedra is presented. A classification of the Medial Axis points in polyhedra is introduced, and the correlative relation of these Medial Axis points is discussed. Some special Medial Axis points are computed base on the correlative elements by sweep off equation group. All other points,edges and faces of the Medial Axis are lead out according to the special Medial Axis points. The result of the algorithm has proved that, compared with typical methods, the complexity is reduced, and the precision of the result is easy to control. Then,Optimal Discrete Filter's theory is discussed, and a new edge detection algorithm is designed on this theory. This algorithm can achieve multi-scale edge detection for grey image efficiently. In fact, the research of this segment is prior work for Discrete Curve Evolution. The recognition methods based on boundary curve and Shock-graph (skeleton) are introduced in this segment, and the performance of both algorithms are compared and discussed. Finally, the whole work of this paper and the future research have been concluded and discussed.