The Research On Analysis And Computation Of Hierarchic Skeletons

As the key technology of computer vision, artificial intelligence, image processing and so on, image recognition can be applied in many fields of military, industry and lives. Selecting right feature is one of the most important factors to make recognition successful, and shape, which includes lots information of objects, is studied deeply and broad.Firstly, distance transform and thinning are introduced and analyzed. The ridges of distance map are positions of skeletons, and the results of thinning are just the midlines of object. But, skeletons based on distance map can not guarantee the connectivity and skeletons of thinning are sensitive to boundary disturbs. So the continuous and stable skeletons may be computed by combining the distance map and thinning.Then, the human visual system (HVS) is introduced. Based on relative theory of physiology and psychology, some disciplines of recognition are presented. The more space held the more important the object; recognition is a hierarchic progress from the whole to details; the sharp part has more significance. These disciplines are applied into skeletons and expressed by parameters of skeletons. The area model of skeleton pixel is constructed to establish the relationship between skeleton radius and local information. The model of information distribution is also gotten by introducing skeleton angle. The algorithm of multi-scale skeletons will use above two models.Following, a multi-scale skeleton algorithm is presented by combining distance transform and thinning. Different recognition functions based on distance map are used to control the extent of thinning to get skeletons of different scale. Skeletons of the algorithm are topology-preserving and continue. By applying angle the algorithm can describe shape more variedly. Due to the open system, new methods of distance map and thinning can be applied to reduce the algorithm complexity. And a feedback system based on the algorithm is also probable.In addition, the computation of skeletons in three-dimensional space is discussed. A 3D thinning algorithm is realized. The extension of the presented algorithm to 3D objects is discussed too.Finally, conclusion on thesis is made and some ideas for future work are given.