Multichannel compression for sequential image retrieval

The paper presents a compression method for multichannel remote sensing images. The method is basically a DPCM scheme that uses a new stochastic model adapted for sets of correlated images. The model was tested on 2 sets of data having different spectral characteristics: multispectral/multitemporal data (SPOT-XS) and hyperspectral data (AVIRIS). The compression rates are derived from the Rate Distortion Function. They are compared, on the same basis, with the compression rates obtained by using the classical AR model for images. The new stochastic model is definitely superior when the image set is spectrally correlated. The compression scheme is given in 2 versions: a random order compression algorithm for data retrieval from image archives and a sequential order compression algorithm for image transmission. Both schemes are developed for lossy versions but they can easily be adapted for lossless compression too.