Objectives  With the enrichment of heterologous image acquisition methods, heterologous image is widely used in many fields, such as change detection, target recognition and disaster assessment. However, matching is the premise of heterologous image fusion application. Simultaneously, due to the differences in imaging mechanisms of different sensors, heterologous images are more sensitive to differences in illumination, contrast, and nonlinear radiation distortion. Therefore, heterologous image matching still faces some problems. There are two main problems, heterologous image feature detection is difficult due to the difference of imaging mechanism, which indirectly increases the difficulty of matching, heterologous image has significant differences in illumination, contrast and nonlinear radiation distortion, which reduces the robustness of feature description and easily leads to matching failure directly.

  Methods  This paper proposes a new matching method considering anisotropic weighted moment and the histogram of the absolute phase orientation. Firstly, anisotropic filtering is used for image nonlinear diffusion. Based on this, the maximum moment and minimum moment of image phase consistency are calculated, and the anisotropic weighted moment equation is constructed to obtain the anisotropic weighted moment map. Then, the phase consistency model is extended to establish the absolute phase consistency orientation gradient. Combined with the log polar description template, a histogram of absolute phase consistency gradients (HAPCG) is established. Finally, the Euclidean distance is used as the matching measure for corresponding point recognition.

  Results  Several groups of heterologous remote sensing images with illumination, contrast, and nonlinear radiation distortion are used as data sources of experiments with scale invariant feature transform(SIFT), position scale orientation-SIFT(PSO-SIFT), Log-Gabor histogram descriptor(LGHD) and radiation-variation insensitive feature transform(RIFT) methods, respectively. The results show that HAPCG method is superior to SIFT, PSO-SIFT and LGHD in the comprehensive matching performance of heterologous remote sensing images, and the average matching number of corresponding points is increased by over 2 times, and the root mean square error is 1.83 pixels. When compared with RIFT method, HAPCG method can achieve higher matching accuracy in the case of similar corresponding points and can realize the robust matching of heterologous remote sensing images.

  Conclusions  The proposed HAPCG method can achieve robust matching performance in heterologous remote sensing images and provide stable data support for multi-source image data fusion and other tasks.