Skip block based distributed source coding for hyperspectral image compression

Due to the involvement of massive amount of information in processing and transmission operations, reduction of encoder complexity is a key requirement for lossy compression methods of hyperspectral images. Distributed Source Coding (DSC) is the enabling technology which reverses the encoder-decoder complexity and provides error resilience. In this work, for complexity reduction, the adaptability of block based design to spatially varying characteristics is exploited and combined with DSC. Blocks of size 8x8 with Mean Absolute Error (MeanAE) less than 3.0 and Maximum Absolute Error (MaxAE) less than 4 are identified and skipped from coding, and the remaining portion has been 2D (2-Dimensional) DCT (Discrete Cosine Transform)/SPIHT (Set Partitioning in Hierarchical Trees) coded. Skip block based DSC scheme results in variable source statistics and this is handled with rate adaptive Low Density Parity Check (LDPC) codes. This new block based algorithm, together with rate adaptive codes results in better compression with reduced coding complexity. Our experimental results show that skip block based DSC coded scheme, in addition to being very flexible, retains all the desirable features of compared well-known algorithms. This is also highly competitive to 3D-SPIHT, and better than 2D SPIHT both in terms of compression efficiency and classification.