An Efficient and Highly Parallel Hyperspectral Imagery Compression Scheme Based on Distributed Source Coding

This paper extends our previous work on hyperspectral imagery compression based on distributed source coding (DSC). We apply DSC principles to facilitate efficient parallel encoder implementations with moderate memory requirement. Based on our previously proposed wavelet-based DSC framework, we propose a novel adaptive coding scheme that judiciously combines DSC and intra coding tools, taking into account the source statistics and inter-band correlation, as well as the coding gains and limitations imposed by tools. Bits extracted from wavelet coefficients tend to have different statistics and inter-band correlation at different significance levels and in different wavelet subbands. Therefore, it is non-trivial to determine the optimal coding strategy. Toward this we propose modeling techniques to estimate the performance of DSC/intra coding under different bits extraction scenarios. This model is used to define adaptive coding strategies that can optimally incorporate different bit- extraction techniques with DSC/intra coding tools according to the bit significance levels and wavelet subbands. Experimental results demonstrate that adaptive coding can achieve up to 4dB improvement over a non-adaptive system, and the improved DSC-based system is comparable to some 3D wavelet system in terms of coding performance. While we focus on hyperspectral images in this paper, many of the proposed techniques are applicable to other wavelet- based DSC image and video applications.