New linear prediction model for lossless compression of hyperspectral images

A novel adaptive linear prediction model for lossless compression of hyperspectral images has been developed. In this paper the model concepts are discussed. New prediction technique is particularly useful for lossless compression of very high spectral resolution images. The adaptive prediction model is an extended unification of 2- and 4-neighbour pixel context linear prediction schemes. It provides new insight into how to predict every band of the image by maximizing compression ratio. The possibility to recognize the best prediction scheme and then to apply it to particular image band has increases compression ratio of multispectral images significantly. The model has been embedded in a lossless compression algorithm at the prediction phase. The algorithm has been tested on real hyperspectral images. Moreover, the same algorithm has been run with two different prediction techniques and without prediction. The results obtained show advantages of the newly proposed prediction model.