Dynamic Window Search and Probability Dominated Algorithm for Fast Vector Quantization Encoding

Vector quantization (VQ), which has been widely applied in speech and image coding, provides an efficient technique for signal compression due to its excellent rate-distortion performance. However, it requires expensive encoding time to find the closest codeword to the input vector. This paper proposes a fast encoding method to speed up the closet codeword algorithm by accounting the occurrence of each codeword in the codebook for every input vector before practical encoding. Then, based on the occurrences, the search order and search window size are determined. The whole VQ system is constituted by two major tables; one is the codebook and the other is table recoding the information of search order and search range which is called SOSR table here. For any input vector x, it is unnecessary to search the whole codebook to find the best match codeword. We just search part of the whole codebook according to the SOSR table. In the pre-processing procedure, we use full search algorithm to find all the best match codewords’ indices during the generation of SOSR table for every input vector x. In the meanwhile, we also calculate the occurrence of the entire searched best match indexes. However, the input vector x is 16 dimension; the memory used to store SOSR will not be tolerant. Therefore, we use the mean value of input vector x to be the key to find its corresponding SOSR information. The data structure of SOSR is given as follows: struct SOSR_type { int search_window_length; pointer next_codeword; }; struct SOSR_type SOSR_table[256];