This paper presents an innovative lossy color image compression based on contrast sensitivity of human visual perception, in which the image contrast sensitivity value (CSV) of each block is determined from Symmetric Gaussian function (SG). Then the outcome of SG is grouped using K-Medoid (KM) cluster to define the adaptive block quantization table (ABQT) for each block in order to provide better quality reconstructed image with good compression ratio (CR). Our experimental result shows that the observed CR is averagely increased by 3 times and visual quality is averagely increased by 9.9% than the existing algorithms namely traditional JPEG and JPEG with contrast sensitivity technique. The tabulated results indicate that proposed methodology out performs very well with the structural content of natural color test images.