Multispectral-image coding by vector quantization with Kronecker-product representation

This paper proposes a new technique based on vector quantization (VQ) for very low bit-rate encoding of multispectral images. The new algorithm relies on the observation that in high spectral-resolution imagery the shape of a generic spatial block does not change significantly from band to band. Therefore, it is reasonable to represent each 3-D spatial/spectral block as the Kronecker product of a spatial-shape vector and a spectral-gain vector, and to jointly quantize only these representative vectors in place of the original block. Even though such an encoding strategy is suboptimal with respect to full-search VQ, the huge complexity reduction allows one to use much larger blocks and to better exploit the redundancy among close pixels of the image. Numerical experiments carried out on high spectral-resolution images show fully satisfactory results, with compression ratios exceeding 100:1, good image quality and very low encoding complexity