Abstract: By fully exploiting the abundant edge features and the strong interband structural correlation of hyperspectral images, a spatial-spectral lossless compression algorithm for hyperspectral images is proposed using local edge based prediction. Based on the coding framework of spectral oriented least square (SLSQ), the algorithm presents a three-modes predictor, which adds a third prediction mode (no prediction) in addition to the original intraband prediction and interband prediction modes. Therefore, the proposed algorithm is more accordant with the correlation property of hyperspectral images. Considering that local diagonal edges generally exist in images, an improved diagonal edge based predictor is adopted for intraband prediction by introducing diagonal edge detection into the median predictor. For interband prediction, a simple but effective strategy for selecting the prediction context is first presented through the analysis of the property of the context when an edge exists in a local context window, followed by an interband predictor based on local edge structural similarity is used to select the optimal prediction context adaptively within the context window. Experimental results show that the proposed algorithm can better remove both intraband and interband correlations, improve the prediction performance and lossless compression ratio.

Key words: Information processing, hyperspectral images, lossless compression, spatial-spectral, local edge based prediction