A wavelet-based fast search algorithm for vector quantization

For complexity in the process of vector quantization image encoding, a new fast codebook search algorithm based on wavelet transform is presented in this paper. Squared Euclidean distance is used as distortion measurement. In the algorithm, a reasonable initial codeword is chosen for the input vector firstly. The relationship between the input vector and codeword, as well as relationship among codewords and characteristics of vectors in wavelet domain, are then utilized to reject impossible matched codewords. The best-matched codeword for the input vector is finally obtained by reducing the search space gradually. The experimental results show the proposed algorithm significantly reduces the complexity of VQ encoding without compromising the encoded image quality with little preprocessing and memory cost.