Lossless compression of hyperspectral images using three-stage prediction

We propose an enhancement to the method for lossless compression of hyperspectral images using two-stage prediction (IP3-BPS). The original IP3-BPS method used a two-stage predictor. The stage-1 predictor took advantage of spatial data correlation and formulated the derivation of a spectral domain predictor as a process of Wiener filtering. The stage-2 predictor took the prediction from the stage-1 predictor as an initial value and conducted a backward pixel search (BPS) scheme on the current band for the final prediction value. We propose an optimal bidirectional pixel search (OBPS) scheme and take it as the stage-2 predictor; the previous stage-2 predictor becomes the stage-3 one in turn. The use of the OBPS scheme improves the efficiency of exploiting the calibration-induced data correlation. Experimental results carried out on the AVIRIS'97 images show the proposed method outperforms previous methods; a 1.5% improvement in bits per pixel was observed compared to the original IP3-BPS method.