一种基于ORB特征和运动一致性的图像校正算法

Image correction algorithm based on ORB feature and motion consensus

目前图像校正算法使用较多的是SIFT特征点和随机采样一致性(RANSAC)算法,但是在提取SIFT特征和用RANSAC算法剔除误匹配会消耗大量的时间,剔除误匹配后仍然存在少量错误.因此提出一种基于ORB特征和运动一致性算法实现图像校正.首先提取ORB特征,使用汉明距离进行初始匹配,然后采用运动一致性算法剔除错误匹配,再使用拓扑约束项对顽固误匹配进行剔除,最后利用RANSAC算法计算出变换矩阵H,从而完成图像的校正.实验结果表明,与SIFT算法相比,该算法不仅特征检测和匹配速度较快,计算资源消耗较少,且具有很好的鲁棒性.

The SIFT feature points and RANSAC(random sampling consensus) algorithm are frequently used in existing image correction algorithm,but it takes a lot of time to extract SIFT features and eliminate false matches with RANSAC algo- rithm,and there are still a few errors after eliminating false matches. Therefore,an image correction algorithm based on ORB feature and motion consensus is proposed,in which the ORB feature is extracted,and initial matching is implemented by means of Hamming Distance. Next,the motion consensus algorithm is used to eliminate false matches,the topological constraint term is used to eliminate the stubborn false matches,and then the transformational matrix H is calculated by means of the RANSAC algorithm to complete the image correction. The experimental results show that,in comparison with SIFT algorithm,the algo- rithm not only has faster speed for feature detection and matching,and less resource consumption for computation,but also has good robustness.