Implemention Of Image Stitching Based On Improved ORB Features And Fusion Algorithm

As a widely used image processing technology,image stitching is spliced into an image mainly based on the overlapping area of the images to be spliced,through registration and fusion operations.This paper mainly studies the two key steps of image registration,image registration and image fusion technology.Aiming at the shortcomings of Oriented FAST and Rotated BRIEF algorithm with no scale invariance,a feature extraction algorithm based on improved ORB is proposed.Firstly,the scale invariance was achieved by constructing the Hessian matrix and the image pyramid,and using the BRIEF binary feature descriptor with rotation characteristics to extract the feature descriptor.Secondly,used Hamming distance algorithm to roughly match the feature points,and used Grid-based Motion Statistics algorithm to distinguish between correct and incorrect matching point pairs.Finally,the improved Random Sample Consensus algorithm was used for fine matching,and the error-matching feature point pairs were eliminated in one step,which can effectively improve the correct matching point pairs under scale transformation.The algorithm in this paper was compared with SIFT,SURF and traditional ORB algorithm,and the results showed that the algorithm in this paper had the advantages of high matching quality and strong real-time under the phenomena of image scaling,illumination and blur,and improved the registration accuracy by about 5% on the basis of the strong real-time advantages of traditional ORB algorithm.In the aspect of image fusion,an improved weighted fusion algorithm based on optimal seam-cutting was proposed to solve the problems of unclear fusion area and splicing gap in the existing algorithm.Firstly,this algorithm used the method of dynamic programming in the overlapping area of the image to be fused,and found the optimal position of the suture line by using the suture search criteria.Secondly,seam-cutting were used to divide the overlapped area from the left side to the overlapped area,the overlapped area and the overlapped area to the right side.Finally,the non-sutured area was fused by the method of gradual in and out weighted fusion,and the final fusion image was obtained.The fusion results were evaluated by the three objective indexes of image information entropy,standard deviation and averagegradient,from which it can be seen that the algorithm in this paper can make the image smooth transition,eliminate the problem of exposure difference,solved the ghosting phenomenon at the stitching gap,and improve the stitching quality.In this paper,the improved ORB algorithm and a weighted fusion algorithm for optimal seam-cutting was used to design the image stitching system,the image sequence fusion operation was carried out and the overall effect is verified.The experimental results showed that the proposed algorithm can effectively increase the registration accuracy,improved the stitching efficiency and achieve a better stitching effect.