Lossless Hyperspectral Image Compression Using Binary Tree Based Decomposition

A Hyperspectral (HS) image provides observational powers beyond human vision capability but represents more than 100 times data compared to a traditional image. To transmit and store the huge volume of an HS image, we argue that a fundamental shift is required from the existing "original pixel intensity"-based coding approaches using traditional image coders (e.g. JPEG) to the "residual" based approaches using a predictive coder exploiting band-wise correlation for better compression performance. Moreover, as HS images are used in detection or classification they need to be in original form; lossy schemes can trim off uninteresting data along with compression, which can be important to specific analysis purposes. A modified lossless HS coder is required to exploit spatial-spectral redundancy using predictive residual coding. Every spectral band of an HS image can be treated like they are the individual frame of a video to impose inter band prediction. In this paper, we propose a binary tree based lossless predictive HS coding scheme that arranges the residual frame into integer residual bitmap. High spatial correlation in HS residual frame is exploited by creating large homogeneous blocks of adaptive size, which are then coded as a unit using context based arithmetic coding. On the standard HS data set, the proposed lossless predictive coding has achieved compression ratio in the range of 1.92 to 7.94. In this paper, we compare the proposed method with mainstream lossless coders (JPEG-LS and lossless HEVC). For JPEG-LS, HEVCIntra and HEVCMain, proposed technique has reduced bit-rate by 35%, 40% and 6.79% respectively by exploiting spatial correlation in predicted HS residuals.