结合CS-LBP和DBN的非受控人脸识别  

Face recognition under uncontained environment based on CS-LBP combined with DBN

作　　者：李琛 王延杰 梁梦媞 LI Chen;WANG Yan-jie;LIANG Meng-ti(School of Computer Science,North China University of Technology,Beijing 100144,China)

机构地区：[1]北方工业大学计算机学院,北京100144

出　　处：《计算机工程与设计》2019年第5期1430-1434,1439,共6页Computer Engineering and Design

基　　金：北京市教委科研计划基金项目(SQKM201810009005);国家自然科学基金项目(61503005);北京市自然基金项目(4162022);北京市优秀人才青年拔尖个人基金项目(2015000026833ZK04);北京市教委科技创新服务能力建设基金项目(PXM2017-014212-000002)

摘　　要：针对当前现有的局部特征描述子在非受控环境下人脸识别的性能受光照、背景、遮挡等因素影响较大的问题,提出基于Retinex理论的小波融合预处理方法并融合CS-LBP和DBN对图像进行高级特征提取的算法,该算法获得的特征既有局部纹理特征对光照鲁棒的特性又有深层特征的特性。相较于直接使用原始图像,使用局部纹理特征的图像作为DBN的输入,明显提升了计算效率。在FERET、Yale-B和CMU-PIE人脸数据库上的实验对比结果表明,该算法较其它算法识别率有明显提高,对光照的变化尤为鲁棒,验证了其优越性。To solve the problem that the face recognition performance of the existing local feature descriptors is greatly affected by lighting,background,shielding and other factors under uncontrolled environment performance,the advanced features were extracted by wavelet fusion pretreatment based on Retinex theory and the integration of CS-LBP and DBN was proposed.This algorithm included both the robustness characteristics of the local texture features to illumination and deep essential characteristics.Compared with the direct use of the original image,the image using local texture features as input of DBN greatly improves the computational efficiency.Experiments on FERET,Yale-B and CMU-PIE face dataset show that the recognition rate of the proposed method is obviously improved,compared with other well-known algorithms,it indicates that this algorithm is particularly robust to illumination and superior.

关 键 词：小波融合 光照预处理 局部纹理特征 深度信念网络 人脸识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]