Study Of Vector Quantization Codebook Design Algorithms

Vector Quantization (VQ) is a significant technique in image compression. VQ research includes codebook design, codebook search and codebook index assignment. And codebook design is a key technique of VQ.In this paper, empty voronoi problems, ant colony codebook design algorithm, particle swarm optimization codebook design algorithm and split vector quantization based on Local Cosine Transform (LCT) will be discussed.First, Hypo-adjacent technique is a way to solve the empty voronoi problems. To overcome the limitation of the technique, an improved technique that is based on the technique is proposed. The improved technique combines the largest voronoi splitting technique, and adopts the way to only cluster the Hypo-adjacent vector with the smallest distortion measure to deal with the empty voronoi. As a result, total distortion measure is decreased, and the performance of vector quantization is improved.Second, in ant colony codebook design algorithm research, frequency sensitivity measure is adopted in the original basic ant colony algorithm. Since the scheme decreases the probability of choosing the same cell by increasing the distortion measure, the randomization of choosing cell is increased. The scheme can avoid receiving a local optimization solution, and enhance the global search ability.Third, in particle swarm optimization codebook design algorithm research, the global best position updating condition in basic particle swarm optimization algorithm is modified. Random probability condition is adopted to update the global best position, extending the global search area and avoiding the premature phenomenon.Finally, in split vector quantization based on LCT research, split vector quantization is adopted to quantize LCT transform coefficients and to design the codebook. The availability of improved split vector quantization algorithms is validated from the objective index and the subjective index, respectively.