Code word searching method in vector quantification

Abstract

The invention discloses a code word searching method in vector quantification. The code word searching method comprises the following steps of: (1) carrying out Hadamard conversion on a generated original code book to obtain a converted code book; (2) generating a reference point set by using the code book; (3) appending a variance value relative to the reference point set for code words in the code book; (4) carrying out the Hadamard conversion on an input vector to be quantified; (5) appending the variance value relative to the reference point set for the converted vector; (6) deleting the code words which do not satisfy the filtering conditions in the code book and searching the matched code words; and (7) outputting the corresponding serial number of the matched code words in the code book by a system. In the invention, by adopting the mode of multi-reference-point set, more redundant code words can be filtered, therefore, the time complexity of calculating the matched code words is reduced, and the searching efficiency is improved. The invention can effectively reduce data access volume, greatly reduce the visit number of a disc, not only adapt the circumstance of higher dimensions, but also satisfy the encoding of the large size code books.

Description

Technical field [0001] The invention belongs to the field of information technology, and in particular relates to a code word search method in the vector quantization process in the field of computer data compression. Background technique [0002] With the rapid development of multimedia application technology, data compression technology has been widely used, among which the vector quantization method is widely used as an effective data compression method, for example, the invention patent "vector quantization method and device" applied by Huawei (Patent application number 200610164455.2, publication number CN 101198041A). This patent application first proposes a vector quantization method, which reduces the complexity of vector quantization through a codebook adjustment method and a simple vector quantization device. This patent application quantizes the input vector by using N codebooks and the adjustment amount of each codebook, thereby reducing the storage capacity of ...

AI Technical Summary

Problems solved by technology

The main disadvantage of this patent application is: in actual use, vector quantization usually exists in the form of high-dimensional vectors, and the technical solution provided by the above patent application still has a relatively high time complexity

The algorithm uses the mean value only combined with a single reference point (origin) to search the code word, which causes the disadvantage of the algorithm: using a single reference point to search the code word, because the code word filter condition is too single, there is still search efficiency The problem of low and frequent disk access times is particularly prominent in high-dimensional situations and large codebook sizes

Images