Research On Image Matching And Identification Algorithms Based On Shape Feature

Shape matching and recognition is an important bifurcation problem in computer vision.It has important applications in object recognition,image retrieval,image registration and object tracking.In recent years,experts and scholars in shape matching and recognition has made the outstanding and effective research achievements,but there are still many problems unsolved in this area,such as: linear transformation of target's recognition,the target of local deformation and screening's recognition,the target of non-rigid transform's recognition and so on.These differences of the shape make it hard to identify.In order to solve these problems,it is necessary to study the shape descriptor and the descriptor matching algorithm.Based on the in-depth study of traditional shape matching recognition algorithm,some new solutions are proposed in this paper.The main contents of this paper are as follows:1?The general process of shape matching and recognition is summarized and the corresponding algorithms of shape matching and recognition are introduced and analyzed by the steps of shape matching and recognition.The general process of shape matching and recognition algorithm includes: shape extraction,shape representation,shape matching and recognition.In this paper,we analyze the domestic and foreign research status of these areas,and introduce the typical methods.2?In order to improve the computation speed and the discrimination ability of descriptor in the process of shape matching,a affine shape matching method using the projection area which calculate by contour is proposed.The algorithm can be divided into coarse and fine match two stages.In the coarse matching stage.Firstly,statistics contour projection area as feature point descriptor.And then,ant colony algorithm is employed in match the public feature point sequence in two pictures.Finally,divide the target curve by the public feature point sequence.In precise match stage,using wavelet transform coefficient construction wavelet local invariant describes the target curve segment to matching the 5% target with minimum cost of the first step,so as to achieve the recognition of affine target.The average retrieval speed of this algorithm is higher than traditional shape projection distribution descriptor 44.3%.Retrieval result on the MPEG-7 image library is 98.65%,The comprehensive evaluation index on the MPEG-7 affine image library is higher than traditional shape projection distribution descriptor 3.1%,25% higher than the shape context.The algorithm proposed in this paper has good matching effect,high efficiency and strong noise immunity,and can be applied to affine shape matching and retrieval effectively.3?The object shape recognition of non-rigid transform and local deformation is a difficult problem in shape recognition.A shape recognition algorithm based on the curvature word bag model is proposed to solve that problem.Firstly,the approximate polygon of object contour is obtained by using Discrete Contour Evolution algorithm,and the shape contour is decomposed into the contour fragments by the vertices of the polygon.Then the bag of contour feature model is used to represent the shape contour fragment.Finally,a linear support vector machine is used to classify the shape feature descriptors.The steps of bag of contour feature model represents the shape contour fragment: 1)Sampling the contour fragment with the same curvature integral and calculating the curvature characteristic,and obtaining the curvature description of the curve;2)Coding each contour fragment with Locality-constrained Linear Coding algorithm;3)The shape feature description was obtained by using the max pooling method to pooling the contour fragments of different curvature integral levels.The experimental results show that the recognition rate of the proposed algorithm in mpeg-7 database can reach 98.21%.The highest recognition rate of the Swedish Leaf and Tools database was 97.23% and 97.14% respectively,The recognition rate of the basic feature descriptor was increased by nearly 10%.The proposed algorithm has high recognition rate and good robustness,which can be applied to the target shape recognition field of non-rigid transform and local deformation.