基于脉冲耦合神经网络的人脸朝向分析

The Study of Face’s Orientation Recognition Based on the PCNN

人脸朝向特征提取是人脸朝向识别的关键。本文采用基于脉冲耦合神经网络(Pulse Coupled Neural Network,简称PCNN)的特征提取方法,分别基于其熵序列、对数序列、时间序列、标准方差序列,完成了多维信息到一维信息的转化,并针对椒盐噪声影响下对数序列的分类效果进行分析。通过欧氏距离法进行人脸朝向的分类识别,结果表明,基于PCNN对数序列对人脸朝向特征提取的正确率可达96%,并且具有一定的抗噪能力。

The feature extraction of face orientations is the key of the face’s orientation recognition. In this paper, four methods including entropy sequence, logarithmic series, time series, and standard deviation series based on the PCNN have been introduced to extract the features of face orientations. It is able to achieve the transform from multi-dimensional information to one-dimensional information. The classified results of loga-rithmic series which are impacted by salt &pepper have been analyzed. The method of Euclidean distance has been used to classify and recognize the face orientations. The results prove that the method based on logarithmic series whose accurate identification rate was 96% has certain antinoise ability.