Abstract: In this article a new hyperspectral image compression method is proposed. Since the three-dimensional wavelet transform compression cannot efficiently utilize the properties of a certain class of images, a hybrid predictive coding and transform coding scheme for hyperspectral images is designed. First, a general expression is derived from the high spectral correlation of hyperspectral images. Then, a new band prediction mode is built in accordance with the expression. Based on the mode all the spectral bands of a hyperspectral image can be reconstructed from a reference band compressed by wavelet transform coding. To improve the precision of prediction, the selector of the reference band and the predictor are established according to the statistical regularity of hyperspectral images and the property of similarity that exists among the shapes of histograms of most bands in the image. Finally, deviation data of the band prediction are compressed by three-dimensional wavelet transform coding. Results indicate that the proposed method can provide competitive performance in comparison with the state-of-the-art technique and it possesses such advantages as good scalability, high flexibility and good progressive transmitting capability.

Key words: hyperspectral image, image compression, predictive coding, transform coding, wavelet transform