Research And Implementation Of Shape Matching Based On Shape Context

Shape matching which is a basic problem in computer vision and pattern matching field is according to some measures to measure the similarity of shapes.Algorithm based on the shape context is proposed based on shape contour this year.It has been applied to many fields,such as the shape matching algorithm in the character recognition, face recognition, content based image retrieval, intelligent video surveillance.The traditional shape context algorithm can match only simple shape and has the following shortcomings:1. It is very sensitive to the contour of the shape because of noise influence.2.It is uniform or random for the boundary extraction of contour points and not very good in representation of shape and existence of redundancy.3. It is very complex and time consuming for calculation of polar histogram in the center point as the center and so is he method of calculating the centroid points. This article in view of the above targeted improved and studyed the algorithm. The improved algorithm uses median filtering method to overcome the influence of noise, fuzzy shape silhouette firstly in the image pretreatment process. Secondly, it improves boundary extraction method to obtain the boundary point and retain important features and remain a few. Thirdly, it reduces the complexity of algorithm because it calculates the centroid coordinates method to obtain the boundary pointcoordinates.The improved algorithm can effectively match shape after tested and analysised. It can effectively match shape by programming and simulation of the improved algorithm.The modified shape context algorithm is very time consuming in image processing median filtering, polar histogram statistics and matrix calculation, which limite the ability of its mass ratio on image. It has discussed the possibility of the three part based on GPU accelerated parallel computing in this paper.It shows that using GPU accelerated effectively reduces the algorithm processing time and obtains good speedup ratio by programming the simulation results.And it makes the mass ratio of image become possible.