Hyperspectral Images Matching via Saliency Features Map

Hyperspectral image(HSI) is applied in many areas such as disaster rescue, geology exploration and ocean observation. However, owning to man-made devices and natural conditions various, the utility of HSI is still limited. Hyperspectral image matching aims at aligning multi-source information from different sensors or air conditions. So this technology attracts more attentions to improve the HSI effectiveness. This paper proposes a novel scheme for hyperspectral image matching using saliency detection and features map. A saliency detection method uses graph by SLIC [1] and manifold ranking extracting similar candidate regions in various channels. Then, we build a features map by guided filtering edges to enhance the key characters and remove unrelated noise. Finally, we make use of mutual information (MI) [2] frame to match the features maps. Experimental results in real hyperspectral data show that our method provides good performance in island and coastline scenes, and outperforms the state-of-the-art methods for hyperspectral image matching.