Lossless coding of hyperspectral images through components ordering by minimum bidirected spanning tree

In this paper, we aim at improving the vector lifting scheme dedicated to lossless coding of hyperspectral images by an optimal components ordering. For this purpose, we propose a three-stage algorithm. The first one consists in classifying the spectral components into separate clusters sharing similar characteristics. Then, within each cluster, the problem of ordering the components for inter-component coding is addressed. It is formulated as the design of a minimum spanning tree whose vertices are the spectral components. Finally, at the last stage, the spatial and spectral redundancies are simultaneously exploited by a vector lifting scheme. Experiments carried out on AVIRIS images indicate the outperformance of the proposed method w.r.t. the state-of-art coders.