Research On Image Matching And Mosaic Technology Based On Improved SURF Algorithm

Panoramic image is composed of multiple images with different perspectives and overlaps.The main feature is the wide field of view,and all information from multiple angles can be observed in one image.In recent years,panoramic image has been widely used in medical image,remote sensing mapping,virtual reality and other disciplines.Image Mosaic technology as a key technology of panoramic image synthesis is particularly important.The traditional image Mosaic technology has a low matching success rate in the image registration stage,and when there are moving objects in the image,the image duplication problem is easy to occur in the image fusion stage,which seriously affects the quality of panoramic image.In view of the defects in traditional image Mosaic technology,now this thesis proposes relevant optimization schemes,the main jobs of which are as follows:1.Three modules of image data processing subsystem in image Mosaic technology are studied,which are image preprocessing module,image registration module and image fusion module.2.Summarize and analyze the related knowledge of image preprocessing,mainly introduce the image preprocessing from four aspects of image histogram equalization,image denoising,geometric correction and projection transformation.The principle of each step and the methods involved are analyzed in detail,and the results obtained by the experimental simulation are analyzed and compared.3.In image registration module,aimed at accelerating the Robust Features(SURF:Speeded Up Robust Features)algorithm in matching the low accuracy problem in image registration algorithm based on SURF algorithm is put forward.First,SURF algorithm is used to extract the feature points,through the nearest neighbor search(BBF:Best Bin Fast)algorithm to realize Kd-Tree quickly find nearest neighbor feature points,combined with bidirectional uniqueness that matches the image matching method is complete,then under the parallax constraints,using the disparity gradient constraint to the initial feature matching to preprocess,the matching to filter out some of the deflection,finally USES the Random Sample Consensus(RANSAC:the Random Sample Consensus)algorithm for matching of quadratic optimization,and to deal with the noise.Three sets of contrast experiments with different states were set up.The BRISK algorithm,ORB algorithm,SURF algorithm and the algorithm in this paper were respectively used for image matching.The correct rate of image matching of each set of experiments were counted.4.In the image Mosaic and fusion module,aiming at the problem of image ghosting in the image Mosaic of moving objects,based on the improvement of SURF algorithm,this thesis adopts the optimal suture algorithm based on dynamic programming and poisson image fusion algorithm to achieve smooth image Mosaic.Two groups of image mosaics and fusion contrast experiments were set up.Through analysis and comparison,it is concluded that the algorithm used in this paper can effectively eliminate the ghosting phenomenon in the image fusion process,and the mosaics and fusion effect is better.