Sort-scan predictive vector quantization on multispectral satellite images

This study proposes a novel idea for a sort-scan model that converts 2-D image blocks into 1-D vectors as an effective preprocessing for vector quantization (VQ). With the sorting of pixel values, samples in the vector space will be highly clustered with fewer distortions when approximated by a certain number of representatives (i.e., the codebook). The sort-scan model can be applied to predictive VQ schemes without extra overhead for the sorting information, Experiments to compress multispectral satellite images were performed and compared using both fixed and dynamic types of scan models. The results show that our sort-scan model is capable of gaining an improvement of 1.4 dB on PSNR relative to the traditional raster-scan model, and even outperforms the previously issued MAW model