The Research Of Image Matching Method And Application Based On Local Feature Detection

In recent years computer vision get great development, image matching technique as a core technology in the field of computer vision, getting more and more attention of academic circles at home and abroad. At present stage image matching is widely used in the application of target recognition, image retrieval,3D reconstruction and other aspects. Therefore, image matching has very important research significance and application value.Through the researchers’ constant innovation, image matching technology obtained significant development and emerged a large number of excellent image matching algorithm, and obtained better result. However, due to the complex image information and changing environment, no matter what kind of image matching technology has its limitations, can not adapt to all images matching problems. Especially when the image suffers great changes such as noise interference or affine deformation and so on, the earliest image matching algorithm based on gray value can hardly adapt to image matching requirements of such images.Currently, the image matching algorithm based on feature is widely used in many ways due to its high robustness, the algorithm mainly consists of two aspects:feature detection and feature description, and as the focus of this paper. For the robustness of the algorithm and the speed of operation, this paper puts forward the relative improvement algorithms through the study of the current classic image matching algorithms. Specific work is as follows:(1) Studied the Hessian-Affine operator in the application of local feature detection, a DCT domain local feature descriptor based on the Hessian-Affine operator is proposed. First image preprocessing is adopted to get ideal image for matching. Then using the Hessian-Affine operator to detect the stability feature regions, the feature regions are normalized in polar coordinates for grid sampling, then the formed sampling matrix make DCT transform and through the ZigZag scanning,finally generating a compact feature vector. According to the nature of the target image, select the dimensions of the feature vector flexibility to generate a local feature descriptor with high robustness.(2) Studied the SURF algorithm in the application of fast image matching, a remote sensing image matching algorithm based on discrete cosine transform(DCT) and speed-up robust feature features(SURF) is proposed. It combines the de-correlation, energy compression and separability of DCT algorithm and the rapidity of SURF algorithm.Firstly,the DCT cofficient reduction matrix is constructed. On this basis, the reconstructed image is as the input of SURF algorithm to get the pre-matching results.Finally,the parameters of the transform model are solved by pre-matching results,and the mismatching is eliminates by using the random sample consensus algorithm.A large number of experimental results demonstrate that the proposed algorithm can improve the matching speed considerably under the premise of remote sensing image matching quality, meet the requirement of real-time.