一种基于索引约束矢量量化的脆弱音频水印算法*

与传统矢量量化不同，索引约束矢量量化在量化过程中通过约束码字索引二进制形式中某一位的值来限定码字的搜索范围。本文利用其特殊的码字搜索方法提出了一种在音频信号中嵌入水印的方法。将原始音频信号分段，每段进行DCT变换并提取若干中频系数构成矢量。水印嵌入时根据水印比特信息和预先设定的索引约束位的值找到匹配码字修改各段DCT中频系数。水印提取时利用传统矢量量化方法得到各量化索引值后，提取出各索引值中与嵌入端相同位的比特值即为水印信息。该方法在量化过程中嵌入水印信息，有很好的实时性。实验结果表明，利用该方法嵌入的水印为一种脆弱水印，可用于认证。

Fragile watermarking scheme based on index constrained vector quantization

Unlike normal vector quantization, index constrained vector quantization can limit the range of searching codewords by select an constrained position in the index in the process of quantization. This paper makes use of this special searching method to propose an audio watermarking scheme. The original audio signal is first segmented into several frames, DCT transform is performed on every frame and the middle-frequnency coefficients are extracted to form vectors. The embedding procedure can be described as finding matching codewords to modify the DCT middle-frequency coefficient vectors in accordance with the watermarking bits and the constrained bits. Normal vector quantization is adopted in the extracting procedure. The constrained bits in the same position are extracted as the watermarking information. This scheme is real-time because the watermarks can be embeded in the quantization process. The experimental results show that the watermarking is fragile and can be used for authentication.