CNN深度学习模型用于表情特征提取方法探究

摘要

Convolutional Neural Network (CNN), as a deep learning framework, shows great performance in extracting feature accurately and reduc-ing model complexity. In consideration of the CNN's advantage in image processing, proposes a feature extraction method based on CNN. Extracts face expression images' features using a 8-layered AlexNet model, and Support Vector Machine(SVM) is used for prediction. Pre-diction results indicate that CNN outperforms other classical methods in extracting image essential features.%卷积神经网络（CNN）作为一种深度学习架构，在精确提取图像特征的同时降低模型复杂度。针对CNN在图像识别方面的优势，提出一种基于CNN的人脸表情特征提取方法。使用具有8层网络结构的AlexNet模型对融合的人脸表情图像进行特征提取，再使用支持向量机（SVM）进行分类预测。将预测结果与一些经典方法如SVM、PCA等做比较，可以发现在样本图片拍摄条件变化较大的情况下，CNN在提取图像本质特征方面有其他方法不可比拟的效果。