Design of Computationally Efficient Vector Quantization Scheme for Image Compression

Abstract

Vector Quantization (VQ) is a conventional image compression technique that uses a simple encoding and decoding process to achieve high compression. The generation of a codebook is an important aspect of VQ design because it has a direct impact on the computational cost and the quality of the reconstructed image. Linde-Buzo-Gray (LBG) is a state-of-the-art codebook design technique that employs the k-mean clustering algorithm. The Bat Algorithm (BA), Particle Swarm Optimization (PSO), Honey Bee Mating Optimization (HBMO), Cuckoo Search Optimization (CS) Algorithm and Fire y Algorithm (FF) are among the optimization techniques used to nd the best codebook. Due to the lack of an optimal solution in the search space, these algorithms have a high time consumption. To predict image patterns for codebook design, this study proposes a novel approach in which the histogram's peak values are applied to prede ned pattern based masks. The patterns obtained are used to initialize the codewords of the initial codebook. In rst iteration the codewords are matched with the training vector using closest match function. A sorting function is applied to obtain the best codewords having the minimum distance with training vectors. The second iteration is performed to get the index and to nilize the codebook. According to the experimental results, the proposed pattern based masking (PBM) algorithm takes less iterations and converges faster than other algorithms, especially at bitrates 0.375, without sacri cing peak signal to noise ratio (PSNR) or structural similarity index measure (SSIM).