Lossless Compression for Hyperspectral Images Using Cascaded Prediction

In the paper, a three-level cascaded prediction based lossless compression method for hyperspectral imagery (HSI) is proposed. This method contains three main steps: preprocessing, prediction and coding. Firstly, k-means algorithm is employed to classify spectral vectors in preprocessing stage. In the prediction stage, a three-level cascaded predictor consisting of a similar neighborhood mean (SNM) predictor, a low-order recursive least squares (RLS) predictor and a high-order least mean square (LMS) predictor is adopted. SNM predictor is used to eliminate spatial redundancy, RLS predictor is used to eliminate spectral redundancy, and LMS predictor is used to obtain final predicted value. Finally, Huffman encoder is employed to encode the prediction residuals. The experimental results demonstrate that this method combines the advantages of the three predictors, significantly reduces the computational complexity and improves the compression effect.