Adaptive Vector Quantization for Improved Coding Efficiency

In this paper, we propose a novel method of improving the initial codebook for Vector Quantization (VQ) to compress still images. VQ is a simple and efficient compression technique which comprises of three phases: 1. Codebook Generation 2. Index Map Generation and 3. Image Reconstruction. Default codebook is generated first and is improved by refining the representative vectors called the code vectors. The codebook optimization technique proposed in this paper improves the quality of reconstructed images to a greater extent where the average bpp is decreased to a value of 0.79 which is a significant improvement. Benchmark images such as Lena, Kush, Cameraman, Barbara are tested with the proposed technique and this technique produces better results.