摘要
为解决人脸性别识别这一传统模式识别领域中富有挑战性的难题,提出一种联合人脸高层特征学习和低层特征学习的深层网络模型。利用卷积和下采样交替提取人脸局部的、抽象的特征,重建原始的输入人脸图像,以学习人脸全局的特征作为补充。联合两类特征时,加入一个可训练的权重进行调节,利用这种多特征结构进行最终的性别分类。实验结果表明,该方法在学习能力和泛化能力上具有超越现有人脸性别识别方法的优秀性能。
To address the challenging problems of face gender recognition based on facial images in the traditional pattern recognition filed,a deep neural network model that learnt the joint features of human face was proposed.The convolution together with sub-sampling was adopted to obtain the human face’s abstract and local features.The human face’s global feature was extracted through reconstructing the facial image as further information.When combining these two kinds of features,a trainable weight was added in the network to improve the final gender recognition performance.Experimental results verify that the proposed method has better learning ability and generalization ability than current face gender recognition methods.
出处
《计算机工程与设计》
北大核心
2016年第1期226-231,共6页
Computer Engineering and Design
基金
国家自然科学基金项目(61303249)
海南省重大科技基金项目(JDJS2013006)
关键词
人脸性别识别
深度学习
多特征学习
学习能力
泛化能力
gender recognition
deep learning
multi-feature learning
learning ability
generalization ability
