稀疏表示人脸识别算法的研究与改进

摘要

为了提高稀疏表示人脸识别技术在对姿态、光照等情况下的识别率和鲁棒性,在GSRC算法的理论基础上提出使用向量总变差模型取代最小l1范数进行稀疏求解.在扩展YaleB人脸数据库和ORL人脸数据库上的数值实验结果中,改进方法在识别率和鲁棒性上都得到了提高,尤其在低维观测数据下,具有较大的优势.表明使用向量总变差模型进行稀疏求解在稀疏表示入脸识别率更具有优势.%In order to improve the recognition rate and robustness of Sparse Representation Face Recognition for the attitude and illumination,in this paper,based on the GSRC algorithm,a vector-valued total variation model was raised to instead of l1-norm for solving the optimization problem.The numerical experiment results on the extended YaleB face database and ORL face database show that the recognition rate and robustness are improved by using the proposed method,and the advantages obtained under the low dimensional observation data are especially greater.