Appraise of similarity measures in image compression using vector quantization codebooks of Thepade's Cosine and Kekre's Error Vector Rotation

Vector Quantization (VQ) is an efficient lossy image compression technique used in codebook generation. The selection of similarity measure for codebook generation affects the distortion between original and reconstructed image. This paper presents comparison of twelve assorted similarity measures for image compression using Kekre's Error Vector Rotation (KEVR) and Thepade's Cosine Error Vector Rotation (TCEVR)code book generation algorithm. The codebook of four different sizes 128, 256, 512 and 1024 are generated for 20 images using 12 different similarity measures. These similarity measures are chosen from five different families such as Minkowski, L1, Squared L2, Intersection and Fidelity. The images are reconstructed from respective codebooks and compared with original image with respect to Mean Square Error (MSE) and Peak Signal-to-Noise Ratio (PSNR). The results have shown that similarity measures belonging to SquaredL2and Fidelity family such as Squared, Euclidean and Square-Chord give lower MSE and higher PSNR values as compared to others. The TCEVR gives better compression over KEVR for all similarity measures except Fidelity family.