一种基于视觉词袋模型的人脸识别方法  

Face Recognition Method Based on Bags of Visual Words

作　　者：王玲[1] 吕江靖 程诚[1] 周曦[1] 

机构地区：[1]中国科学院重庆绿色智能技术研究院智能多媒体中心,重庆400714

出　　处：《电视技术》2015年第17期112-115,120,共5页Video Engineering

基　　金：中国科学院战略性先导科技专项基金项目(XDA06040103);国家自然科学基金项目(61472386);重庆市科委科技攻关重大项目(cstc2012gg-sfgc4001)

摘　　要：针对人脸图像因受表情、光照、角度等因素影响,导致人脸识别率较低的状况,提出了一种基于视觉词袋模型的人脸识别方法。该方法首先对图像进行分块并提取局部特征,其次利用训练样本的所有局部特征训练全局的混合高斯模型,然后以此为初始化训练单张图像的混合高斯模型,生成该图像全局特征向量,最后用PLDA进行人脸识别。通过在LFW数据库上进行实验,结果显示该方法的识别率高于传统的特征提取方法,证明其具有更强的识别性能。In order to improve the state of low recognition rates because face images are affected by the factors of expression, light, pose, et al, a new encoding method based on bags of visual words is proposed. Firstly, all training images are sampled into local ar- eas in a dense and local appearance feature descriptors are gained from the local areas. Next, global GMM is obtained from local feature descriptors of all training images. Then, using global GMM as initialization of every image from the test database, image- specified GMM is gotten from which the face feature vector generates. Finally, PLDA is used for face recognition. The experiments are carried on the databases of LFW and the resuhs show that the recognition rates of the proposed method are higher than the other traditional feature extracting method, it proves that the proposed method has a stronger recognition performance.

关 键 词：视觉词袋模型 人脸识别 混合高斯模型 特征提取 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]