Registration of MWIR-LWIR band hyperspectral images

Abstract

Previously proposed hyperspectral image registration methods mostly focus on the registration of the images including overlapping bands in VNIR and SWIR range. In contrary to previous methods, we investigate the registration of hyperspectral images with no-overlapping bands in MWIR and LWIR range in this paper. The proposed approach achieves the image registration over 2D maps extracted from 3D hyperspectral data cubes. Considering that the main component of the captured signal in MWIR-LWIR range is thermal radiation, we first propose to use the brightness-temperature estimate of hyperspectral pixels to form the 2D image. In addition, hyperspectral pixel energy, average emissivity and the first three components of principal component analysis (PCA) transform are also utilized and tested for 3D-2D conversion. The performance of the methods are evaluated by the matching ratio of the interest points and by generating mosaic images from the given maps. The experimental results indicate that brightness-temperature estimate, pixel energy and first principal component gives comparable results for image matching. The emissivity maps and the remaining principal components are found to be not successful for image registration as these features do not form a common base for different band signals.