Finite-State Vector Quantization Techniques for Image Compression

Srijati Agrawal
2020 International research journal of innovations in engineering and technology  
Different techniques of Finite-State Vector Quantization, in the image coding framework are investigated in this paper. Introduced in this paper are two FSVQ designs, conventional vector quantization and sidematch vector quantization. Convention Vector Quantization divides an image into K-dimensional blocks where each vector x is mapped to its corresponding code vector from a codebook of size N. It exploits the correlation between adjacent pixels within a block of pixels. The encoder of a
more » ... encoder of a vector quantizer uses the previously encoded blocks to make a selection from a family of codebooks. Although vector quantization provides an acceptable visual quality of the compressed image, it does it at the cost of increased bit rate. The Side Match Vector Quantization reduces the bit rate required for storage and transmission of the image and provides a good visual quality image. SMVQ requires creating its own state codebook for each block for encoding as well as decoding. SMVQ takes advantage of the spatial contiguity of pixel vectors by exploiting the correlations of the nearest neighboring blocks. They try to minimize the granular noise that causes the annoying effect of visible pixel block boundaries in conventional VQ. Since the conventional method for the generation of state codebook for SMVQ is time consuming, the method of rapid generation of state codebook proposes a fast method for codebook generation. In this paper, different image compression techniques of vector quantization, side-match vector quantization and rapid generation of state codebook method will be implemented to evaluate the best possible method. Although each technique is an improvement over the other, the proposed method for rapid generation of state codebook is faster than the others and without any loss of perpetual visual quality. The Linde-Buzo-Gray algorithm, also known as LBG algorithm, is used for the generation of codebooks.
doi:10.47001/irjiet/2020.407001 fatcat:o6poey3kdjfadnq4ftkakslikm