Abstract : New generation of imaging spectrometers are emerging in the field of space exploration by adding an additional view of measurement, the angular dimension. Multi-angle imaging spectroscopy is conceived to provide a more accurate characterization of planetary materials and a higher success in separating the signals coming from the atmosphere and the surface. The Compact Reconnaissance Imaging Spectrometer for Mars (CRISM) aboard the Mars Reconnaissance Orbiter is a hyperspectral camera that operates systematically in multi-angle mode from space. Nonetheless, multi-angle hyperspectral images are related to problems of manipulation, visualization and analysis because of their size and complexity. In this framework, this PhD thesis proposes robust statistical and physical algorithms to analyze images acquired by the CRISM instrument in an efficient manner. First, I propose a tailor-made data pipeline aimed at improving the radiometric quality of CRISM data and generating advanced products, the latter data being devised to perform fine analysis of the planet Mars. Second, I address the atmospheric correction of CRISM imagery by exploiting the multi-angle capabilities of this instrument. An innovative physically-based algorithm compensating for atmospheric effects is put forward in order to retrieve surface reflectance. This approach is particularly used in this thesis to infer the photometric properties of the materials coexisting in a specific site of Mars, the Gusev crater. Third, I perform an intercomparison of a selection of state-of-the-art techniques aimed at performing spectral unmixing of hyperspectral data acquired by the CRISM instrument. These statistical techniques are proved to be useful when analyzing hyperspectral images in an unsupervised manner, that is, without any a priori on the scene. An original strategy is proposed to discriminate the most suitable techniques for the exploration of Mars based on ground truth data built from independent high resolution imagery.