关键词: 图像配准, 特征点匹配, 描述子, 中心对称局部二值模式, 空间序列特征

Abstract:

An image registration algorithm is proposed through the analysis of the objective function. There are two optimal solutions to the function: the biggest number of efficient point pairs and the image transformation matrix with the highest accuracy, which can be solved by two different point matching matrices. The algorithm gets the transformation matrix by the points described in the center symmetric local binary pattern (CS-LBP), and obtains the efficient points by the statistical deflection angle order (SDAO). The SDAO is a sample and accuracy descriptor that integrates the local structure and global information of images. By combining with the advantages of the two description methods, the algorithm achieves a high alignment accuracy and a small computational volume.

Key words: image registration, point matching, descriptor, center symmetric local binary pattern, spatial order property