LAUSR

Due to different imaging mechanisms bringing significant nonlinear radiation differences (NRDs), traditional feature matching methods are difficult to obtain satisfactory results for multispectral images. The key to multispectral image matching is eliminating the NRDs and extracting more robust features. This letter proposed a novel descriptor combining the phase consistency (PC) gradient and log-polar coordinates to solve the NRDs of multispectral images. First, the nonlinear weighted moment (NWM) is constructed for detecting feature points. Then, the PC model is extended to construct the absolute PC orientation gradient. Combined with the log-polar description, the histogram of PC gradients (HPCG) is established. Finally, a purification algorithm with dynamic adaptive Euclidean distance and fast sample consensus (FSC) constraints is proposed to eliminate mismatches while retaining the correct matches. The proposed HPCG descriptor was evaluated using six groups of multispectral images. The experimental results show that our method is better than the other seven descriptors [i.e., scale-invariant feature transform (SIFT), edge-oriented histogram (EOH), edge histogram descriptor (EHD), log-Gabor histogram descriptor (LGHD), PC edge histogram descriptor (PCEHD), optical-to-SAR-SIFT (OSSIFT), and histogram of oriented self-similarity (HOSS)], which have strong adaptability and robustness.