The Research Of Object Matching And Recognition Based On Shape Context

Shape is the main basis for us to distinguish what kind an object is, and the shape of the object is the basis for visual analysis and object recognition too. In digital image processing,the shape of an object is defined sa the contour of an object. Object recognition is an important branch of computer vision research. The research of object recognition is based on analyzing the shape feature extracted and shape matching. This thesise adopts the shape context object matching and recognition.Shape context is a strong descriptor and widely used in video surveillance, character recognition, image retrieval and so on. Its key advantage is the global descriptor, applicable to all general deformation. In this thesis, the shape context algorithm is studied, this thesis mainly aims at the following three aspects:(1) Firstly, it aims at shortcoming of shape context algorithm, by calculating the centroid to determine the selection of the contour points. Make given specific points processing,attain to reduce the amount of computation(2) Then it aims at the selected contour points, ignore the factor of the shape complexity and contour points distribution does not always reflect the real shape in the shape context algorithm. This thesis introduce fineness ratio to describe the shape complexity, using the given contour points and the fineness ratio of product as a practical selection of contour points to achieve adaptive select point.In order to make these points can reflect the real contour,this paper sets the contour points of the distance according to perimeter of shape average contour points.and in order to make up for the actual contour points too little and improve the matching correctness,this thesise paper to extract the corners and contour points together to participate in the match.(3) The shortcoming of traditional median filter fixed template is improved, then by using edge detection, contour tracking and other steps to extract shape features,and then conducting shape matching.Using the TPS transform model to calculate the distance between the minimum bending energy and the shape of the shape is used as the judgment condition,then the object recognition is accomplished.Using MPEG database verify the method in this thesise. the experimental result is good,Based on the analysis of the data, the method is applied to the MPEG database image retrieval by setting threshold, the retrieval precision is high.