Research On Image Registration And Application Based On SIFT Algorithm

Image registration is one of the basic tasks of image processing. It is used tomatch the images which were obtained from different conditions of the same objects orscene. In order to improve the accuracy of the image registration, many researchersstudied this problem and proposed a lot of registration algorithms, such as registrationalgorithm based on gray level information, optimization strategy and characteristic.One of the most important algorithms is Scale Invariant Feature Transform(SIFT)registration algorithm, it was proposed by Lowe DG in2004. Although this algorithmcan achieve good effect, there are still some problems, due to many redundant keypoints which need great storage space. This problem may lead to mismatch and costmore the registration time. Therefore, we research on SIFT algorithm and the mainworks of this paper are as follows:(1) We propose an improved SIFT registration algorithm with regional extractionmethod. The number of targets in an image is limited, so if we can find out the sametarget area before matching, the SIFT feature points will be decreased. This way alsocan improve the efficiency of the registration. In this paper we use the traditional edgedetection algorithm to extract the target area at first, and then use SIFT registrationalgorithm to match the target. Experimental results show that the proposed algorithmcan improve efficiency of the traditional method and guarantee registration precision.(2) We propose a new SIFT algorithm based on the visual attention model. Themost significant area of the image can be found by the visual attention mechanism.Comparing with the traditional edge detection algorithm, this algorithm extracts the areawhich corresponds to the important goal in the image. The area also can reflect the maininformation of the image. It excludes the interference of the background, so that thefeature points which are extracted by SIFT have a clear meaning. After we using thevisual attention model to extract significant goals，we match the corresponding target.The experimental results show that this method can reduce the complex backgroundinterference. It improves the registration precision and efficiency. In addition, thealgorithm still has good robustness.(3) We apply this algorithm to image mosaic. An improved image fusion algorithmis proposed. This new fusion algorithm solves the gap between the overlap regioneffectively and makes the image transition smoothly. The improved registration method has higher precision, so that it can get better effect with the improved fusion method.