摘要
自然场景下的人脸图像数据通常分布在非线性的高维空间中,因此,传统线性特征提取算法难以获得鲁棒的特征。针对上述问题,提出一种基于非线性提取的人脸识别算法。该算法将非线性特征提取算法引入到人脸识别的过程中,对人脸特征匹配阈值进行预处理,将模拟遗传退火算法和深度信念网络相融合,先利用模拟遗传退火算法优化深度信念的网络连接权值,在此基础上对预处理人脸特征匹配阈值进行寻优,增强了传统算法对于天气、光照、形态等多种外界因素的鲁棒性。实验仿真证明,该算法提取特征的稳定性强,能有效的识别人脸图像,精度较高。
Face image data in natural scenes is usually distributed in nonlinear high-dimensional space. Therefore, traditional linear feature extraction algorithms are difficult to obtain robust features. Aiming at the above problems, a face recognition algorithm based on nonlinear extraction is proposed. The algorithm introduces the nonlinear feature extraction algorithm into the process of face recognition, pre-processes the face feature matching threshold, fuses the simulated genetic annealing algorithm with the deep belief network, and first optimizes the deep belief network by using the simulated genetic annealing algorithm. Based on the weight of the connection, the pre-processing facial feature matching threshold is optimized, and the robustness of the traditional algorithm to various external factors such as weather, illumination and shape is enhanced. The experimental simulation proves that the proposed algorithm has strong stability and can effectively recognize face images with high precision.
作者
江诚
石雄
JIANG Cheng;SHI Xiong(School of Electrical and Electronic Engineering , Wuhan Polytechnic University, Wuhan 430023,China)
出处
《武汉轻工大学学报》
2019年第2期35-39,共5页
Journal of Wuhan Polytechnic University
关键词
非线性特征提取
阈值
人脸识别
鲁棒性
nonlinear feature extraction
threshold
face recognition
robustness