SOFM网络在矢量量化的应用

Application of the SOFM Neural Network in Vector Quantization

矢量量化作为一种高效的数据压缩技术,在语音和图像的编码、传输中都有广泛的应用,其关键在于码书设计。码书的好坏直接影响语音、图像的编码质量。本文针对经典LBG算法对初始码书敏感及整体训练时间较长这两个缺陷,着重研究SOFM算法在这两方面的性质和特点,结果证实SOFM算法相对于LBG算法训练时间较短,且利用SOFM网络设计的码书受初始码书的影响小,具有很强的自适应性,很好的改善了LBG算法在这两方面的缺陷。

Vector quantization as a highly efficient data compression technology has been widely used in voice and image compression coding and transmission. The key problem of VQ is codebook design, because codebook has direct impacts on voice and video encoding quality. There are two serious shortcomings about the classic method LBG algorithm. It is sensitive to the initial codebook and training time is long. To solve these two problems, the text mainly research SOFM algorithm property and point of these two aspects, the result confirms that the codebook designed by SOFM network suffers small impact from the initial code book, and it can self-organized proceed study discipline, and have very strong adaptability. So we can see it well improved LBG algorithm＇s shortcomings in these two aspects.