Research And Application Of Image Matching Algorithm Based On The Features

Image matching has been widely used in many fields, such as military, network security, industry, medicine and so on. It’s the fundamental technology of image processing, it’s also a challenging and important research for further explorations. The research on image matching has been popular for years and the related algorithms have been proposed by the researchers, from the original image matching which based on gray to the current better image matching algorithms based on the feature. In this thesis,we mainly study the image matching algorithms based on the feature.Firstly, we introduce the concept of image matching, the general process and several important factors of image matching, meanwhile, briefly describing the basic theory of scale space. Then, we study several classical algorithms of image matching,such as SIFT, PCA-SIFT, SURF and color-based SIFT. We have a detailed study of the principle and steps of the SIFT algorithm, the extracted features by SIFT own several advantages, such as the invariance to scaling, translation and rotation, robust to view angle and the luminance. The extracted feature points by SIFT have the problem of correlation and matching efficiency, which lead to the emergence of PCA-SIFT and SURF. PCA-SIFT uses the principal component analysis to extract a small number of features to replace the more features of the SIFT. SURF implements the acceleration and parallels by proposing Hessian(Matrix). Color-SIFT combines the color information and the features which extracted by SIFT to accomplish the image matching.We also perform an image matching experiment using the algorithm in this thesis. Since the matching rate of generated descriptor isn’t high enough and matching effect between the original images and the symmetric images isn’t good, presenting a new method which utilizes the SMC framework in search strategy and modifies SIFT descriptors to solve these problems.Search strategy is an important bottleneck of the image matching performance improvement. Under the basis of SIFT, we introduce the SMC framework into the search strategy. SMC optimizes the search result by the combination of resampling and potential constraints. Experiments show that the improved algorithm has a significant improvement accuracy and speed during image matching. Reversing the process of the SIFT descriptors, makes it have the mirror image feature, so it can leads to a satisfactoryresult during image matching process. At the end of this thesis, an application of image classification which based on image matching has been proposed. Based on the SIFT algorithm, BoF algorithm is introduced into the image classification, which mainly clusters the feature points which extracted by SIFT to improve the through the image classification effect.