Low Bit Rate Compression for Hyperspectral Data using Sparse Representation

Chengfu Huo, Rong Zhang, Yin Dong, and Anzhou Hu


Hyperspectral Data, Sparse Representation, Compression Technique


Since all the bands of hyperspectral data have the same imaging area, the dictionary that can sparse represent one band may also represent the other bands sparsely. Based on this property, this paper presents a new compression technique for hyperspectral data using sparse representation. First, one band is chosen for training a desired dictionary, and then the other bands are sparse represented over the learned dictionary. The training band is encoded losslessly so as to guarantee that the same dictionary can be re-learned in the decoder. The sparse coefficients of the other bands are quantized at first using fixed number of bits, and then the corresponding position-indices and magnitudes are encoded. The innovation of this paper is that patches having the same spatial location of all bands are restricted to be represented using the same atoms, thus the bit-cost of encoding the position-indices can be reduce significantly. Experimental results based on OMP and K-SVD are provided, which reveal that this proposal has better performance than wavelet based compression technique, in the sense of rate distortion and classification precision, at low bit rates.

Important Links:

Go Back