Research On Image Matching And Retrieval Algorithms Based On Shape Feature

Shape matching technology is a hot topic in computer vision, it have a very wide range of applications in shape-based object recognition, graphic stitching, content-based image retrieval and other areas. In recent years, researchers has made many remarkable achievements in the shape matching technology, but there are still exist many problems better to resolved, such as affine transform between shapes, nonlinear variability and shape of the deformation or occlusion. For these reason, how to choose effective shape description and matching method is the key to solve the problem. In this paper, we put forward some new methods based on deeply studying traditional methods to solve these problems.our works are as follows:Sum up the general flow of shape matching, and do a thorough analysis and research from the general process. Firstly, we select filtering algorithm to smooth the input image in the progress of pretreatment to remove noise. Then, Using the edge detection method to extract the planar curve of target. to shape description, which can be divided into contour-based descriptor and region-based descriptor. and we also have done a detail analyze about their advantages and drawbacks.For partial deformation or occlusion problem, A affine-invariant shape matching algorithm based on LCS sequence method and local invariant is presented in this chapter. The algorithm includes two steps, namely rough matching and exact matching.As for rough matching, we use the LCS sequence method to find the longest common sequence of feature points. The sequence of feature points are corresponded. Then,using the features points to divide the target curve. for exact matching, A new affine-invariant local feature descriptor is constructed, which is used to describe the curve segments. And their matches are measured through their similarity. At last, The experiments results shows that our algorithm is effective on affine target recognition.For nonlinear variability problem, a shape retrieval algorithm based on contour and BOW method is presented in this chapter. Firstly, the contour fragments to describe the local object feature are obtained by splitting the object contour with contour curvature.During the period, we use the gaussian function denoising and delete the wrong corner via the character of curvature. Then, the contour fragments are described based on the centroid of shape context. Finally, BOW method is combined to construct the shape description for retrieval purpose. The shape description is better for grasp the disciplinenonlinear variability. So our method has a better retrieval performance than other methods.