Lossless Compression of Hyperspectral Images Using Adaptive Prediction and Backward Search Schemes

摘要

In this paper, an effective lossless compression scheme for hyperspectral images is presented. The proposed scheme is based on a table look-up approach in prediction and employs two novel measures to improve the compression performance. The first measure takes advantage of the spatial data correlation and formulates the derivation of a spectral domain predictor as a process of Wiener filtering. The derived predictor is considered statistically optimal provided that the data within a small context window are stationary. This property holds in most cases due to spatial data correlation. Under the Wiener filtering framework, the proposed predictor can be extended from one-tap to multi-tap prediction to further improve performance. In the second measure, a backward search scheme is used instead of look-up tables, which reduces the memory storage requirement drastically and achieves performance equivalent to that obtained using multiple look-up tables. The search effort is greatly reduced using the quantization index approach. Simulations on parameter settings and refinements on entropy coding are conducted to fine-tune performance. Experiments on 5 sequences of AVIRIS images show that the proposed scheme can yield an average compression ratio of as high as 3.85.