Error-free compression of multispectral image data using linear vector prediction

A method for error-free (loss-less) compression of digitized multispectral image data is investigated. The compressor applies the linear prediction technique to the pixel vectors in the spectral direction to exploit and remove both the inter-band and the intra-band redundancy in the multispectral image data set. The method is error-free in that no information is lost in the reconstructed image data compared with the original data. Also, the compressed file is designed so that the partial decoding is possible directly from the compressed file. Thus any rectangular portion in the image frame of any spectral band in the multispectral data set is able to be extracted without decoding of other portions. This provides a great convenience to the data user and saves a great deal of temporary storing space in the decoding procedure. Comparisons with other common data compression software are given, from which the advantage of the proposed method in the compression rate is obviously seen.<<ETX>>