为了使仿人机器人的视觉系统能够识别出人脸,利用BP神经网络的理论知识,研究人脸识别技术,对图像中人脸朝向判别进行实验研究。实验采用Matlab工具箱进行BP网络设计,实现对人脸角度方向的判别。讨论了输入和目标向量设计BP神经网络结构的设计,以及网络参数和训练参数的设定等问题。实验结果表明,BP神经网络可以根据输入图像的二值化信息,以一定的准确率判别该图像的人脸朝向。
In order to enable humanoid robot's vision system to recognize out people's faces,this paper use BP neural network to form a method of face's orientation recognition,which is an experimental study of the discrimination of human face's orientation in images.This experiment uses MATLAB toolbox for the design of BP network to achieve the discrimination on the direction of face.The paper discusses the design of input and target vectors,the design BP neural network architecture and the settings of network parameters and training parameters and so on.The experimental results show that BP neural network can tell the directions which faces are toward to in a certain degree of accuracy.