Abstract: To eliminate the limit that the recent subspace learning methods for facial feature extraction are sensitive to the variations of orientation, position and illumination in capturing facial images, a novel facial feature extraction approach is proposed. Facial images are decomposed into a sequence of binary images using a Modified Pulse-Coupled Neural Network（M-PCNN）, and then the information entropies of each binary image are calculated and regarded as features. A support vector machine-based classifier is employed to implement recognition and classification. Simultaneously, it overcome the disadvantage of standard PCNN model with high number of parameters. Theoretical and experimental results show that the proposed approach is robust to the variations of orientation, position and illumination conditions in comparison with the recent sub-space based methods.

Key words: Pulse-Coupled Neural Network（PCNN）, Modified Pulse-Coupled Neural Network（M-PCNN）, face recognition, feature extraction, information entropy, Support Vector Machine（SVM）

摘要： 针对基于子空间分解的人脸特征提取方法对人脸图像在采集过程中的光照、旋转、位置等变化较为敏感的问题，提出了一种改进的脉冲耦合神经网络人脸特征提取方法。该方法模拟生物视觉的感知过程，将人脸图像分解成由若干二值图像组成的认知序列，计算序列中的每幅二值图像的熵作为人脸特征，基于支持向量机实现分类与识别；同时克服了标准的脉冲耦合神经网络模型参数过多的缺点，识别率也有所改善。理论与实验结果表明，该方法与现有的基于子空间分解的人脸特征提取方法相比，对人脸图像在采集过程中的光照、旋转、位置等变化有较强的鲁棒性，而且具有较低的维数。

关键词: 脉冲耦合神经网络（PCNN）, 改进的脉冲耦合神经网络（M-PCNN）, 人脸识别, 特征提取, 信息熵, 支持向量机（SVM）