Multisensor Fusion of Landsat Images for High-Resolution Thermal Infrared Images Using Sparse Representations

Land surface temperature (LST) is an important parameter in the analysis of climate and human-environment interactions. Landsat Earth observation satellite data including a thermal band have been used for environmental research and applications; however, the spatial resolution of this thermal band is relatively low. This study investigates an efficient method of fusing Landsat panchromatic and thermal infrared images using a sparse representation (SR) technique. The application of SR is used for the estimation of missing details of the available thermal infrared (TIR) image to enhance its spatial features. First, we propose a method of building a proper dictionary considering the spatial resolution of the original thermal image. Second, a sparse representation relation between low- and high-resolution images is constructed in terms of the Landsat spectral response. We then compare the fused images created with different sampling factors and patch sizes. The results of both qualitative and quantitative evaluation show that the proposed method improves spatial resolution and preserves the thermal properties of basic LST data for use with environmental problems.