Research On Shape Recognition Method Based On Skeleton Features

Based on shape recognition objects has become an important research direction ofcomputer pattern recognition, which has been widely applied in many fields. Such asimage retrieval, robot navigation, medical image analysis, path planning, characterrecognition and other areas. Although many scholars have already done a lot of researchwork on the shape recognition, there are still many unresolved problems, such as theshape of mutual occlusion or the really goals exist flexible change in the complicatedscene. We mainly aim on the shape description and matching method when they arepartial occlusion or flexible change. The main works and achievements are as follows:1. Several key technologies and methods of shape recognition are introduced,including the shape description and feature matching. As for shape description, wedivide the description methods into two kinds which are based on the contour shapemethod and based on contour region method; For feature matching, two commonmethods are introduced which contain based on distance measure method and based onoptimization algorithm method. At last, we introduce the invariant features of an image,the invariant features of image is skeleton descriptor.2. A new algorithm based on skeleton is proposed to recognize objects withflexibility change and partial occlusion or broken. Firstly, we define a center point andsort the skeleton endpoints counterclockwise according to the defined center point. Thenwe search the skeleton path of the two adjacent endpoints by using A*algorithm.According to skeleton path, the feature invariant is constructed to describe the skeletonendpoints. Finally, graphics matrix is built to measure similarity. As for the descriptor ofendpoint is base on partial skeleton in this paper, which is a local descriptor, so it caneffectively overcome flexibility change and partial occlusion or partial broken.Furthermore, time complexity has proved to be reduced due to the feature invariant oflocal skeleton path is extracted. Theoretical analysis and experimental results show thatthe algorithm can effectively identify objects of flexibility change and partial occlusionor partial broken.3. A new method for recognition flexibility change and partial occlusion objectscombined skeleton and histogram is proposed. First, the combine-points of skeleton aredetected, and the shape of objects is divided into several regions based oncombine-points; second, the invariant of sub-region is established and construct the invariant features in a histogram; Then a new similarity measure function is built for thehistogram; Finally, the similarity of both model and target combine-points are computedto judge whether two shapes are match. In this paper, skeleton combine-points areextracted, compared with skeleton end-points, our method is more stable and thematching accuracy has also improved. Furthermore, our algorithm uses sub-regiondescription, which can effectively recognize objects of flexibility change and partialocclusion. Theoretical analysis and experimental results show that the algorithm iseffective.