Multi-Orientation Edge-Based Satellite Image Matching Method for Optical and Sar Images

Abstract

Abstract. Multi-sensor image matching is a key technology to fully exploit complementary information on images from multiple sources. In particular, the nonlinear radiance distortion between optical and SAR images makes image matching very difficult. Focusing on satellite optical and SAR image matching, this paper proposed an oriented edge-based template image matching, namely Multiple Orientation Edge Feature Correlation (MOEFC). Firstly, the edge feature was extracted by the Sobel operator to construct oriented gradient channels on optical and SAR images. Then spatial filtering was performed separately by a two-dimensional and a one-dimensional kernel. Finally, the similarity was determined by stacking the normalized cross correlation (NCC) of each oriented gradient channel. To validate the new method, it was compared with other four edge-based template image matching methods detected by Sobel, Canny, Laplacian and phase congruency-based methods on satellite optical-SAR image pairs, which were tested in rural and urban areas. And the results show that the proposed MOEFC method can obtain the largest number of correct matching (NCM) with the least root mean square error (RMSE) in both areas. In the rural area, the NCM of the MOEFC method is higher than the other four methods, and the improvement ranges from 54.1% to 74.3%. And the RMSE of the MOEFC method is 0.645 pixel. In the urban area, the NCM of the MOEFC method can also be improved by 3.6% to 46.4%. And the RMSE of the MOEFC method equals 0.489 pixel.