摘要
人脸识别是一项实用新技术,在使用过程中有很高的要求,目前对于人脸识别的研究多种多样,但在识别速率和准确度上很难达到人类视觉的效果。文章研究了在输入网络之前首先通过Sobel算子对人脸图片进行预处理,然后在卷积神经网络的基础上对网络结构和参数进行了改进,并采用SVM作为分类器,利用CUDA(Computer Unified Device Architecture)进行加速,使得网络的速度和识别率有了很大提高。最后将改进的网络与PCA、BP神经网络和传统CNN的人脸识别方法进行比较,结果表明改进的网络效果更优。
Face recognition is a practical technology. It also has high requirements in the process of use. The current research on face recognition is more diverse. However, it is difficult to exceed human vision in recognition rate and accuracy. Therefore, the goal of this paper is to achieve the result of human vision by increasing the rate and accuracy. Before imputing the network, the face image is preprocessed by Sobel operator. Then, the network structure and parameters are improved on the basis of convolution neural network, and SVM is used as the classifier. Accelerated by CUDA, it has greatly improved the speed and recognition rate of the net- work, being close to human visual effects. Compared with the face recognition methods of PCA, BP neural network and traditional CNN, this method is more effective. And it can greatly improve work efficiency in practical applications.
作者
黄剑
贺兴时
HUANG Jian;HE Xing-shi(College of Science,Xi'an Polytechnic University,Xi'an 710048,China)
出处
《渭南师范学院学报》
2018年第20期39-46,共8页
Journal of Weinan Normal University
基金
国际科技合作计划项目:基于深度学习的理想CPU流水线分支预测模型研究(2018KW021)