Research On Splicing Algorithm Of Large Parallax Image Based On Local Feature Matching

Image matching and splicing is a technique that first matches two or more narrow-angle images with overlapping regions and then mosaics them into a wide-angle,high-resolution image,which has been widely used in medical images,remote sensing images,virtual reality and so on.Image matching and image registration fusion are two key research contents in image stitching.Because in the shooting,the ideal shooting conditions are usually not satisfied,that is to say,when shooting multiple photos,the camera should be fixed to a point in space,only doing rotation,so the shot will have a certain parallax.Large parallax image splicing often occurs in the overlapping regions due to low registration accuracy and distortion in the non-overlapping regions.In order to solve this problem,this paper makes a detailed study on the main process of large parallax image mosaic,and optimizes the image matching and registration.The main work of this paper is as follows:(1)Research the image matching algorithm based on local feature points,elaborate and learn the SIFT algorithm,SURF algorithm,ORB algorithm in detail,then propose an improved ORB algorithm to match images,and test the above image matching algorithms on the open data set,Finally count the matching accuracy and matching time of the matching algorithm,which proves that the improved ORB algorithm has not only good robustness but also high matching efficiency.(2)When the random sampling consistency algorithm is used to filter the interior points,if the parameter model is poor,the correct matching point will be misjudged as the outer point,which will seriously affect the accuracy of the image registration algorithm.Experiments show that using the same feature matching algorithm,the vector field consistency algorithm are faster and retain more interior points.(3)An improved APAP algorithm is proposed.Firstly,the image grid is divided according to the distribution of feature points to achieve the purpose of high-precision registration of overlapping regions.Then the global optimal similarity transformation matrix is calculated by using random sampling consistency algorithm,which effectively limits the distortion of non-overlapping regions.Finally,the content of overlapping regions is perceived.The mosaic image is obtained by preserving the low-importance region fusion.In order to verify the superiority of the proposed algorithm,the APAP algorithm,SPAP algorithm and AutoStitching algorithm are used to compare the experiments with the proposed algorithm,and the results are evaluated by visual subjective,quantitative alignment and operational efficiency.