基于SVM和归一化技术的音视频特征融合身份识别

针对噪声环境下人脸识别率和说话人识别率低的问题,在研究特征层融合的基础上,结合归一化技术和SVM理论,提出了一种融合人脸和语音的多生物特征识别模型。首先采用离散余弦变换和局部保持投影算法提取人脸特征及SVM方法提取语音特征,在特征层进行融合得到融合特征后,计算测试身份与模板间的距离,为了减少计算量和提高识别性能,对匹配距离进行归一化处理,最后输入到SVM进行识别。仿真结果表明,在噪声环境下,当信噪比降低时,融合识别率要明显高于单个系统的识别率,达到了身份识别的目的。

Audio-visual Feature Fusion Person Identification Based on SVM and Score Normalization

In order to solve the problem of low recognition rate of face recognition and speech recognition under the wicked noise conditions. Based on the studies of feature level fusion theory and combined with Normalization and SVM theory, a novel model for face features and speech features fusion recognition is presented in this paper. First, we extract the face features and speech features correspondingly, then we fuse the two features on the feature level in order to obtain the fusion feature, after the calculation of the distance between the test people and template people we normalize the matching distance so as to reduce the computational and to improve the recognition accuracy. At the last, we put the normalization matching distance into SVM can we obtain the recognition result. The experiment show that the fusion system performs well both in response time and system accuracy espeeially in noisy background.