Cross-Spectral Image Patch Matching by Learning Features of the Spatially Connected Patches in a Shared Space

Abstract

Cross-spectral image patch matching is a challenging problem due to the significant difference between two images caused by different imaging mechanisms. We consider cross-spectral image patches can be matched because there exists a shared semantic feature space among them, in which the semantic features from different spectral images will be more independent of the spectral domains. To learn this shared feature space, we propose a progressive comparison of spatially connected feature metric learning with a feature discrimination constrain (SCFDM). The progressive comparison of spatially connected feature network keeps the property of each spectral domain in its corresponding low-level feature space, and interacts the cross-spectral features in a high-level feature space. The feature discrimination constrain enforces the framework to refine the shared semantic space for feature extraction. Extensive experiments shows that SCFDM outperforms the state-of-the-art methods on the cross-spectral dataset in terms of FPR95 and the training convergence. Meanwhile, it also demonstrates a better generalizability on a single spectral dataset.