Compression of multispectral images by address-predictive vector quantization

摘要: Abstract Multispectral images are formed by a large number of component single subject taken in different spectral windows. They often represented tens or even hundreds Mbits data and huge resources required to transmit store them, making some form compression necessary. To obtain high efficiency, exploiting both the spatial dependency, we propose two coding schemes based on vector quantization address prediction, one more suited case strong dependence, other preferable dependence. The performances proposed techniques assessed means numerical experiments compared those known literature. It turns out that for ratios order 25:1 reconstructed almost undistinguishable from original ones, good image quality is still achieved as 40:1.