基于概率神经网络的近红外人脸朝向识别方法  

Near-infrared face orientation recognition method based on probabilistic neural network

作　　者：梁剑波 柴群 周长敏 LIANG Jianbo;CHAI Qun;ZHOU Changmin(Big Data Engineering College,KaiLi University,KaiLi guizhou 556011,China)

机构地区：[1]凯里学院大数据工程学院,贵州凯里556011

出　　处：《激光杂志》2023年第7期172-177,共6页Laser Journal

基　　金：黔东南州科技计划项目(No.黔东南科合J字{2021}46号);贵州省教育厅自然科学研究青年人才项目(No.黔教合KY字[2020]187号)。

摘　　要：近红外人脸图像受到光照干扰的影响,导致近红外图像噪声较大且清晰度较低,人脸朝向识别难度较大,为此提出基于概率神经网络的近红外人脸朝向识别方法。首先结合多级小波分解和样条插值法对光照干扰下的近红外人脸图像加以补偿,然后通过二维主元分析法提取图像特征向量,再建立概率神经网络并利用模拟退火粒子群算法优化网络平滑因子,最后将待识别近红外人脸图像输入至训练后神经网络之中,实现人脸朝向识别。实验结果表明,所提方法的近红外人脸图像预处理效果更好,人脸朝向识别pitch角、yam角和roll角误差值容许范围分别为97%、97%、92%,MAE、STD和RMSE分别为3.49、3.45、5.00,均优于文献对比方法,人脸朝向识别精度较高。The near-infrared face image is affected by the light interference,which leads to the high noise and low definition of the near-infrared image,and the face orientation recognition is difficult.Therefore,a near-infrared face orientation recognition method based on probabilistic neural network is proposed.First,combine multi-level wavelet decomposition and spline interpolation method to compensate the near-infrared face image under illumination interference,then extract the image feature vector by two-dimensional principal component analysis method,and then establish a probabilistic neural network and use simulated annealing particle swarm optimization algorithm to optimize network smoothing factor,and finally input the near-infrared face image to be recognized into the neural network after training to realize face orientation recognition.The experimental results show that the proposed method has better preprocessing effect on the near-infrared face image,and the allowable error values of the pitch angle,yam angle and roll angle of face orientation recognition are 97%,97%,and 92%,respectively.MAE,STD and The RMSEs are 3.49,3.45,and 5.00,respectively,which are better than the literature comparison methods,and the face orientation recognition accuracy is high.

关 键 词：概率神经网络 近红外人脸朝向识别 多级小波分解 样条插值法 模拟退火粒子群算法 

分 类 号：TN911[电子电信—通信与信息系统]