BRIFT:一种基于二值描述符的多模态图像匹配方法

BRIFT:A Binary Descriptor for Multi-Modal Image Registration

针对使用辐射变化不敏感特征变换(RIFT)在最大索引图(MIM)上计算特征描述符、进行特征匹配耗费时间长的问题,提出了基于RIFT的二值描述符算法(BRIFT)。首先计算图像的相位一致性并得到MIM,然后在MIM上通过快速算法计算特征描述符并将其二值化,最后利用Hamming距离作为距离测度进行匹配。将存在多种几何畸变和辐射度变化的异源遥感图像作为测试数据,将BRIFT算法分别与SIFT,BRISK,BRIEF,RIFT等方法进行对比,结果表明,在略微损失匹配精度的情况下,所提BRIFT算法能节省约80%~90%的特征描述符计算时间,节省约50%的特征匹配时间,实现异源遥感图像的快速鲁棒匹配。

Using radiation-variation insensitivity feature transform(RIFT)to calculate feature descriptors and perform feature matching on the maximum index map(MIM)is time-consuming.To solve this problem,an algorithm named binary descriptor for RIFT(BRIFT)is proposed.First,the phase consistency of the image is calculated and the MIM is obtained.Then,the feature descriptor is calculated and binarized through fast algorithm on MIM.Finally,the descriptor of each feature point is matched based on Hamming distance.Taking the remote sensing images with various geometric distortions and non-linear radiation distortions as the test data,the BRIFT algorithm is compared with other feature matching algorithms including SIFT,BRISK,BRIEF,RIFT.The comparison result shows that the proposed BRIFT algorithm can save about 80%to 90%of time consumption of feature descriptor evaluation and about 50%of feature matching time consumption with a slight loss of matching accuracy,and achieve fast and robust matching of remote sensing images from different sources.