On Codebook Redundancy of Vector Quantizer and Its Application

• Main Authors: Kuang-Tai Chen, 陳光泰
• Other Authors: Shaou-Gang Miaou
• Format: Others
• Language: zh-TW
• Published: 2000
• Online Access: http://ndltd.ncl.edu.tw/handle/09808926779642692029

碩士 === 中原大學 === 電子工程學系 === 88 === Signal compression is useful for saving and transmitting large amount of signals. Many signal compression methods have been proposed. Among which, vector quantization (VQ) receives great attention due to its high compression efficiency and wide application diversity. It can be used in a direct method to encode the signal directly or used in a transformation method to encode the transformed coefficients. This thesis discusses the codebook redundancy problem of VQ for quasi-periodic signals. A quantitative redundancy measure of the codebook is proposed. Furthermore, we present two algorithms to reduce the redundancy. Both algorithms use a mixed codebook structure containing one and two dimensional codebooks. The first algorithm is applied to the case where only a LBG-trained codebook is available and the second algorithm uses a LBG-like algorithm to reduce codebook redundancy when a set of codebook training vectors is available. With MIT/BIH electrocardiogram (ECG) database, the experiment results show that both algorithms can reduce the codebook redundancy effectively with essentially no loss of signal quality. When the codebook size and codevector dimension of the mean-shape VQ (MSVQ) are 1024 and 40, respectively, the resulting percent of root-mean-square difference (PRD) is 10.7%. By using the first algorithm, we can reduce the codebook storage space to 60% of MSVQ codebook for 0.1% increasing in PRD. The second algorithm can reduce the codebook storage space to 25% of MSVQ codebook for 0.5% decreasing in PRD.