Application of an Improved Orb Algorithm in Image Matching

An improved ORB feature matching algorithm is proposed to solve the problem of high mismatching rate of non-uniform distribution of feature points extracted by traditional ORB algorithms. First of all, to extract feature points of image block, and then calculate each block information entropy, the entropy value dynamically setting the threshold of each block to extract the feature points, make to extract the feature points of distribution more uniform, improve the subsequent registration accuracy. Secondly, for the problem of high mismatch rate caused by insufficient use of local information by the rBRIEF descriptor in the ORB algorithm, on the basis of introducing gradient information to improve the LBP algorithm, an improved descriptor combining rBRIEF and LBP is proposed. The 128-bit descriptor generated by the improved LBP algorithm is used to replace the 128-bit descriptor with the smallest variance in the original rBRIEF to improve the stability of the feature point description. Finally, RANSAC was used to delete the mismatched points. Results Compared with the traditional ORB algorithm, the improved algorithm reduces the time overhead by 8% and the average mismatch rate by about 10%. The improved orb algorithm has lower mismatch rate and higher registration speed.