基于数字水印的人脸与声纹融合识别算法

Augmenting remote multimodal person verification by embedding voice characteristics into face images

提出远程多模态的生物特征数字水印算法,将声音特征作为水印加入到人脸图像中.运用文献[1]提出的改进型量化索引调制(QIM)方法,算法加入一个脆弱型的水印用于篡改检测,同时加入一个鲁棒型水印用于隐藏声音的高斯混合模型(GMM)参数.利用人脸、声纹和多模态识别算法,提出的方法能够实现对篡改的检测,对常见的攻击,例如图片缩放、高斯噪声、模糊化、伽马校正和JPEG压缩等具有鲁棒性.在由295人组成的XM2VTS数据库上,该多模态系统能够获得95.93%的识别率,同时获得3.19%的等错误率.

A novel biometric watermarking algorithm was proposed to augment remote multimodal recognition by embedding voice characteristics into face images.Using the modified quantization index modulation（QIM）scheme proposed by reference[1],the algorithm embedded both a fragile watermark for tampering detection,and a robust watermark to represent the Gaussian mixture model（GMM）parameters extracted from voice.Using face,voice and multimodal recognition algorithms,the proposed watermarking scheme can detect tampering,and is robust to watermarking attacks such as resizing,Gaussian noise,blurring,Gamma correction and JPEG compression.On the XM2 VTS database consisting of 295 persons,the multimodal system can obtain recognition rate of 95.93% for identification,and equal error rate of 3.19% for verification.