The Method Of Fast Image Mosaicking Based On Area Blocking And Feature Point Matching

With the rapid development of computer vision,the requirements for its image and video become more and more critical.Therefore,panoramic image with high resolution and wide perspective has become the daily necessity.Panoramic image stitch technology can satisfy people's visual demands for video and image by synthesizing the collected sequence images into a seamless panoramic image.Because of its wide application prospect,it has attracted many scholars to conduct in-depth researches.Image stitch technology has been divided into two categories: based on characteristics on the one hand and on area on the other.The former has been featured by small amount of calculation and high real-time performance,which has been the current research hotspot.Hence,in this thesis,the image stitch technology based on characteristics had been focused so as to improve the real-time performance of image registration and enhance the quality of stitch results.Meanwhile,the traditional panoramic image stitch algorithm had been advanced,and the block-based double threshold FAST(Features from accelerated segment test)matching algorithm had been proposed.Moreover,the Laplacian multi-resolution image fusion algorithm had been improved,and the real-time performance and the accuracy of the results of stitch were enhanced.Besides,the efficiency and practicability of the algorithm had been verified by experiments.The main research contents are as follows:1.To traditional image stitch algorithm has been for the entire image,had done not detect or match feature points in two most similar area,which reduces the real-time performance of image registration and the accuracy of feature point matching.In order to improve the accuracy rate of feature point matching and the real-time performance of image registration,a dual-threshold FAST feature point matching algorithm based on block has been adopted to first calculate the most similar region and then detect and match feature points in the most similar regions.The proposed algorithm had enhanced the realtime performance of image registration and the robustness of feature point matching.2.In the process of image acquisition,vertical direction of camera tilt or jitter may cause the deformation of stitch result.At the same time,while multiple sequence image stitch,cumulative error easily leads to the distortion and deformation of the panorama.In order to improve the quality of panoramic image stitch,the image alignment algorithm based on camera calibration is put forward,by the transformation of several coordinate axes,which weakens the distortion and deformation of panoramic image.3.When there is moving object in the overlapping area of two images,artifacts and incomplete image will appear.To handle this,Laplacian multiresolution fusion algorithm based on Min-Cut/Max-flow has been proposed,which ensures the fusion efficiency,improves the quality of the fusion,a stitch line has been calculated by weight value in the image coincidence area.Experimental verification that the fusion result can realize natural transition,and eliminates artifacts and incomplete of object.