Abstract: A self-adaptive HVS-optimized textures compression algorithm based on Vector Quantization (VQ) is presented. Utilizing the property of Human Visual System (HVS), a function judging the similarity between blocks is designed instead of using Euclid distance between pixels in block. Correlated threshold in the judgment is computed using the property of image. With the novel quantizer, different resolution images can be handled automatically. In addition, a self-adaptive threshold adjustment during the compression is designed to improve the reconstruct quality for textures with different regional correlation. To enhance the efficiency of the code-words, lateral association is used through the compression process. Experiment on various resolution images indicates that the algorithm can achieve satisfied compression rate and reconstruct quality at the same time. Furthermore, the compression and decompression process is speed up with the usage of GPU, on account of their parallelism.