贝叶斯多分类器组合模型可以表示图像间的条件概率和限定语义,并依此预测用户查询和序列图像间的相似度,是解决视频序列图像信息检索的有效手段。为了提高视频人脸取证系统中人脸图像匹配速度和准确度,提出和设计了一种基于贝叶斯多分类器的视频人脸跟踪识别取证系统。给出了系统的总体结构、主要模块的设计,以及采用的关键技术。实验中训练数据为有限的静态人脸图像,测试数据为视频图像序列。实验表明,该系统识别效率高、取证能力强、容错性好,在视频人脸分类中具有较好的效果,为动静结合人脸图像分类提供了一定的依据。

Bayesian multi-classifier model are the suitable model to deal with the problem of vector sequence images information retrieval, because they are the appropriate tools to store the conditional probabilities and limited meanings among terms and compute the similarity between user query and sequence images. In order to improve the speed and precision of human face image retrieval in the forensics system, a forensics system of vector human face based on Bayesian multi-classifier is proposed and designed. The overall structure, the main module design, and key technologies are given. In the experiment, trained data are generated from the limited static human face images, and tested data are originated from the video frequency image sequence. The experimental results show that the system has both quick recognition speed and high forensics capacity with the good capability of fault tolerance, and the method gains the good performance on vector face classification, and it provides a certain basis for classifying human face image both dynamically and statically.