Vector quantization (VQ) is a well-known signal compression method. In the framework of VQ, fast search method is one of the key issues because it is the time bottleneck for practical VQ applications. By introducing the-law-of-cosines and directly using the angular information to reject a candidate codeword, the previous work [Chung, K.L., Lai, J.Y., 2004. Pattern Recognition Lett. 25 (14), 1613-1617] has proposed a very efficient fast search method for VQ encoding. However, there still exist two problems in this work as (1) a complicated arccosine function (i.e., COS-1) is used and (2) the reference vector for a given input vector is fixedly selected as the initial best-matched codeword in terms of the minimum difference between L2 norm and no updating to the reference vector is conducted during the whole search process. This paper aims at improving these two problems further by (1) avoiding using the COS-1 function completely and (2) updating the reference vector whenever it is possible so as to be able to set up a better reference vector for the following search process. Because a better reference vector always helps to reject a candidate codeword, it in principle can speed up the search process. Experimental results confirmed that the proposed method outperforms the previous work [Chung, K.L., Lai, J.Y., 2004. Pattern Recognition Lett. 25 (14), 1613-1617] obviously.
ASJC Scopus subject areas
- コンピュータ ビジョンおよびパターン認識