Histogram of the orientation of the weighted phase descriptor for multi-modal remote sensing image matching
Histogram of the orientation of the weighted phase descriptor for multi-modal remote sensing image matching
Abstract
Multi-modal remote sensing image (MRSI) has nonlinear radiation distortion (NRD) and significant contrast differences to which image gradient features are usually sensitive. Although image phase features are more robust against NRD, they might not be much helpful in resolving the problems of directional inversion or phase extreme value mutations that are common in the phase feature calculation. To address these issues, a new MRSI matching method—"histogram of the orientation of weighted phase” (HOWP)—is proposed in this paper. This method distinguishes itself from other methods in three aspects: (1) a feature aggregation strategy is used to optimize feature points by extracting the corner and blob features separately; (2) a novel weighted phase orientation model is established to replace the traditional image gradient orientation features; and (3) a regularization-based log-polar descriptor is constructed to generate robust feature description vectors. To evaluate the performance of the proposed method, we selected 50 sets of typical MRSIs with translation, scale, and rotation differences for comparison with the other four state-of-the-art methods. The results show that our method is more resistant to radiometric distortion and the contrasting differences in MRSIs. It also performs better in tackling the problems of direction reversal and phase extreme value mutation, as evidenced by more, the number of correct matches (NCM). Since the method has improved the average NCM by 1.6–4.5 times, the average success rate by 35.5%, and the average rate of correct matches by 11.1% with an average root of mean-squared error of 1.93 pixels. Moreover, we have put forward an extended version of the HOWP method (Simplified-HOWP) when there is no image rotation, which manifests in an average 0.75 times improvement in NCM of Simplified-HOWP performance over that of the HOWP method. The executable code and test data are linked in https://skyearth.org/publication/project/HOWP/.
Figure 1. 五种算法的对比结果