ASM姿态矫正结合字典学习优化的人脸识别  

Face posture recognition method using dictionary learning and ASM posture correction

作　　者：钟小莉[1] ZHONG Xiao-li(School of Computer Science,Qinghai Nationalities University,Xining 810007,China)

机构地区：[1]青海民族大学计算机学院,青海西宁810007

出　　处：《计算机工程与设计》2018年第11期3538-3543,3583,共7页Computer Engineering and Design

基　　金：国家自然科学基金项目(61462072)

摘　　要：针对人脸图像中姿态变化而导致识别率降低的问题,提出一种基于主动轮廓模型(ASM)姿态矫正结合字典学习优化的人脸识别方法。利用ASM提取人脸图像局部特征,对人脸进行矫正对齐;将人脸图像进行Gabor小波变换以提取初始特征,执行KPCA获得最终特征空间,利用非约束字典学习进行优化;利用人脸样本特征空间构造稀疏字典,形成稀疏表示分类器。在构建分类器时,使用错误分类的人脸图像更新训练基向量,提高分类器的分类精度。实验结果表明,该方法在LFW人脸数据库上识别一幅人脸图像仅需1.05s,对姿态变化、低分辨率具有很好的鲁棒性。相比其它几种方法,其取得了更高的识别率。For the issue that the recognition accuracy decreases for pose variations in face images,a face gesture recognition algorithm based on active shape model(ASM)posture correction and dictionary learning was proposed.ASM was used to extra extract the local features of face image,and the correcting alignment of faces was done.Gabor wavelet transform was performed on the face image and the initial features were extracted.The kernel PCA was performed to obtain the final feature space,and the non-restraint dictionary learning was used to do optimization.The sparse dictionary was constructed using the feature space of face samples to form a sparse representation classifier.In the construction of the classifier,the face images with error classification were used to update the training base vector,to improve the classification accuracy.Experimental results show that the proposed method needs only 1.05 seconds for recognizing a face image from LFW database.It can accurately identify the human face attitude.And it has good robustness to pose variations and low resolution.The proposed method has higher recognition accuracy than other pose variations methods.

关 键 词：人脸识别 ASM姿态矫正 字典学习 核主成分分析 稀疏表示 GABOR小波变换 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]