基于重引力搜索和深度学习的图像表情识别研究  

Image expression recognition algorithm based on gravity search and deep learning

作　　者：杨芳[1] 郭宏刚[2,3,4] YANG Fang;GUO Honggang(Police scientific research department,Hebei Vocational College of Public Security Police,Shijiazhuang 050091,China;College of Computer and Cyber Security,Hebei Normal University,Shijiazhuang 050024,China;Key Laboratory of Network&Information Security,Hebei Normal University,Shijiazhuang 050024,China;Hebei Provincial Engineering Research Center for Supply Chain Big Data Analytics&Data Security,Hebei Normal University,Shijiazhuang 050024,China)

机构地区：[1]河北公安警察职业学院警务科研处,河北石家庄050091 [2]河北师范大学计算机与网络空间安全学院,河北石家庄050024 [3]河北师范大学河北省网络与信息安全重点实验室,河北石家庄050024 [4]河北师范大学河北省供应链大数据分析与数据安全工程研究中心,河北石家庄050024

出　　处：《光学技术》2020年第5期626-633,共8页Optical Technique

基　　金：国家自然科学基金资助项目(61572170);河北省自然科学基金项目(F2019205163)。

摘　　要：传统表情识别技术采用单一类型的特征表示方法,由于每个特征类型对不同数据集的表示效果存在差异,导致传统技术对不同数据集的表情识别效果也存在较大的差异。设计一种多类型混合特征的选择方案,用以提高不同数据集的表情识别准确率。将面部不同区域、不同类型的特征集作为基础特征集,利用重引力搜索算法从基础特征集中选择优化的特征子集。将优化的特征子集输入深度信念网络进行训练和半监督学习,采用训练的网络模型对表情进行识别。实验结果表明,在不同数据集条件下,采用该方法均能够保持较高的识别准确率。Traditional expression recognition techniques adopt a unique representation type of features,because each type of features has different effects to various datasets,so the techniques show different expression recognition effects to various datasets.A selection approach for multiple types mixed features is designed,and is to improve the expression recognition accuracy of different datasets.The algorithm treats the features of different types and different areas as base feature set,it also takes advantage of gravity search algorithm to select the optimized feature subset from the base feature set.On the other hand,the algorithm inputs optimized feature subset to deep belief nets for training and semi-supervised learning,finally the trained network is used to recognize expressions.Experimental results show that in different datasets conditions,the proposed algorithm keeps high recognition accuracy all the time.

关 键 词：深度学习 深度神经网络 重引力搜索算法 表情识别 特征选择 表达式目录树 

分 类 号：TP394.1[自动化与计算机技术—计算机应用技术] TH691.9[自动化与计算机技术—计算机科学与技术]