增强边缘梯度二值卷积神经网络的人脸姿态识别

Enhanced Edge Gradient Binary Convolutional Neural Network for Face Recognition Across Pose

为了提高人脸姿态识别的识别精度,设计了一种增强边缘梯度二值卷积神经网络用于识别.首先,提出ROILBC(Region of Interest Local Binary Convolution)在人脸姿态图像上提取二值特征并归类,根据二值特征图谱和原像的对比情况选择人脸姿态图像ROI(Region of Interest)以供后续网络学习.其次,提出DR-MGPC(Dimensionality Reduced Modified Gradient Pattern Convolution)提取图像边缘梯度二值特征,在此基础上,提出Enhanced DR-LDPC(Enhanced Dimensionality Reduced Local Directional Pattern Convolution)提取图像增强边缘梯度方向特征.网络采用直方图相似度、卡方检验、常态分布比对的巴氏距离法作为测量依据来进行识别;实验在FERET和CAS-PEAL-R1数据集上进行,相比其他人脸姿态识别方法,提出的二值模式卷积神经网络在识别精度和计算效率上更优异.

In order to improve the accuracy of face recognition across pose,an enhanced edge gradient binary convolutional neural network is designed for recognition.Firstly,this paper proposed ROILBC(Region of Interest Local Binary Convolution)to extract binary features on face image and classify the features,ROI(Region of Interest)of the face image is selected which based on the comparison between binary feature map and the original image for subsequent network learning.Next,the DR-MGPC(Dimensionality Reduced Modified Gradient Pattern Convolution)is proposed to extract the edge gradient binary features,then,the Enhanced DR-LDPC(Enhanced Dimensionality Reduced Local Directional Pattern Convolution)is proposed to extract the enhanced edge gradient direction features.The network adopted Histogram similarity,chi-square test,and Bhattacharyya distance as judgments to discriminate different face pose.Extensive experiments conduct on FERET and CAS-PEAL-R1databases,the experimental results show that our method significantly outperforms other approaches on accuracy and computational efficiency.