A Self-similar and sparse approach for spectral mosaic snapshot recovery

Traditional Hyperspectral Imaging (HSI) architectures face a fundamental trade-off between spatial, spectral and temporal resolution, requiring repetitive scanning of the scene. A new generation of Snapshot Spectral Imagers exploit Spectrally Resolvable Detector Arrays to sample the entire hyperspectral cube from a single frame. However these systems are also limited since only a single band is captured by each pixel. In this work, we propose a novel approach for estimating the missing measurements, by exploiting the self-similarities and the sparsity of representations in appropriate dictionaries, without the need for additional training examples. We demonstrate the high quality reconstruction of the proposed method in cases where we artificially induce the particular sampling pattern, as well as in cases where the frames are acquired by snapshot spectral mosaic sensors.