Affine LBG for Codebook Training of Univariate Linear Representation

Tiannan Dong, Jianji Wang, Meng Yang, Kai Yi, Nanning Zheng

Abstract

LBG algorithm is a simple and effective method to train code-book for vector quantization. Since LBG was proposed, several interesting algorithms have been published to improve the effectiveness and efficiency of LBG. Univariate linear representation is another important data compression method, which approximates a target vector by a linear transformation of a selected codeword from codebook. Many applications also use LBG or K-means algorithm to train the codebook of univariate linear representation. In this paper, we propose an improved LBG algorithm called the affine LBG algorithm to train the codebook for univariate linear representation. The experimental results show that the affine LBG algorithm can derive a more effective codebook than LBG algorithm for univariate linear representation. Moreover, the affine LBG algorithm is more efficient than LBG algorithm.
BibTeX
@inproceedings{dong2018affine,
  author = {Tiannan Dong and Jianji Wang and Meng Yang and Kai Yi and Nanning Zheng},
  title = {{Affine LBG for Codebook Training of Univariate Linear Representation}},
  booktitle = {GlobalSIP},
  year = {2018},
  doi = {10.1109/globalsip.2018.8646389},
}

Loading related content...

Figure