Research On Compression Algorithm Of Multispectral Remote-Sensing Images

With the development of multispectral remote-sensing technology, compression of multispectral images is of more and more importance. As the original data is needed sometimes, the lossless/nearlossless compression of multispectral images is necessary. In this paper, some theories and methods of compression of multispectral images are reviewed and researches on multispectral remote-sensing image compression by Karhunen-Loeve Transformation(KLT) and neural network are discussed.At first, theories and methods about coding are systematically reviewed from the viewpoint of information theory and the arithmetic coding is discussed in detail. Secondly KLT and reversible integer-to-integer wavelet transform for image compression are discussed, and a method of nearlossless compression of multispectral images is given, which combines KLT and integer wavelet transform together. The spatial redundancy in the images is removed by KLT and the inter-band redundancy is removed by the integer-to-integer wavelet transform. The experiment results from practical multispectral images have shown that this algorithm is efficient. If the original image is reconstructed by five eigen subimages, the nearlossless compression ratio is above 11 for the data used in this paper and the PSNR is more than 45dB. This method is easy in controlling MSE by choosing the number of reservation eigen values. In the end, the PCA algorithm based on neural network is proposed. The experiment results show that this PCA algorithm can get the same compression ratio as the KLT algorithm with low complexity.