静态图像中正面人脸表情识别算法研究
发布时间：2018-12-19 21:29
【摘要】：近年来人工智能在学术界和工业界都得到了很大的发展,尤其是AlphaGo在围棋比赛中4:1战胜李世石之后,人工智能技术受到了全民的重视。人脸表情识别技术作为人机交互的重要方式,也受到了更多的重视,一些企业更是提供了判定人脸微笑与否的API给用户使用。本课题针对目前六种基本表情识别技术并未商用的现状,提出了一种简单的在静态图片中识别人脸基本表情的算法框架,并在MATLAB环境下编程实现了文中提出的算法。本课题紧跟当前人工智能迅速发展的步伐,对静态图片中六种基本人脸表情的自动识别技术进行了深入地研究。为了验证本文算法的正确性,本文提出的算法在CK+与JAFFE数据库上进行了应用。同时,本文提出算法获得的识别率与一些文献中获得的识别率进行了比较,进一步说明了该算法的正确性和有效性。本文的主要研究内容和创新研究如下:(1)本文研究了静态图片中六种基本表情与其中性脸的相关性,将卡尔·皮尔森相关系数应用到了定义人脸表情活跃区域的研究工作之上。本文首先对比了各个表情与其中性脸的静态图像的相关性,然后根据各种表情与其中性脸的相关性数值的大小,确定了人脸表情的活跃区域。(2)本文在人脸表情识别研究上首次提出并应用了活跃区域归一化方法。本文首先归一化了人脸活跃区域,并在人脸的活跃区域提取了 LBP与HOG特征。其中,为了更精确地定位人脸的活跃区域,一种在人脸上定位多个精确关键点的人脸对齐算法被应用到了本文中。(3)LBP与HOG特征在归一化之后的活跃区域中被提取出来,并且在该文章的实验中这两种特征被较好地融合在了一起。实验证明,两种特征融合之后的识别率较之单个特征效果更好。本文首次将伽玛校正方法应用到了 LBP特征之上,该方法在很大程度上提高了人脸表情的识别率。(4)本文定义了σ参数,研究者可以通过找到该参数的最大值来找到合适的伽玛校正值。本文设计的实验证明了该参数很大程度上缩减了找到合适伽玛校正值的工作量。(5)在本文中,一种基于手工提取特征识别六种基本人脸表情的算法框架被提出,并且该算法框架被应用到了 CK+和JAFFE数据库之上。本文提出的算法在CK+数据库上取得了目前已知最好的识别率,同时其在JAFFE上也取得了很有竞争力的识别结果。
[Abstract]:In recent years, artificial intelligence has been greatly developed in both academia and industry, especially after AlphaGo defeated Li Shishi at 4:1 in the go game, artificial intelligence technology has been attached great importance to by all people. As an important way of human-computer interaction, facial expression recognition technology has been paid more attention to, and some enterprises provide users with API to judge whether people smile or not. In view of the fact that six basic facial expression recognition techniques are not commercially available at present, this paper proposes a simple algorithm framework for facial expression recognition in static images, and implements the proposed algorithm under the MATLAB environment. Following the rapid development of artificial intelligence, the automatic recognition technology of six basic facial expressions in static images is studied in this paper. In order to verify the correctness of the proposed algorithm, the proposed algorithm is applied to CK and JAFFE databases. At the same time, the recognition rate obtained by the proposed algorithm is compared with that obtained in some literatures, which further demonstrates the correctness and effectiveness of the algorithm. The main contents and innovations of this paper are as follows: (1) this paper studies the correlation between six basic expressions and their neutral faces in static pictures. Karl Pearson correlation coefficient is applied to the research work of defining the active region of facial expression. This paper first compares the correlation between each expression and the static image of its neutral face, and then according to the magnitude of the correlation between each expression and its neutral face, The active region of facial expression is determined. (2) in the research of facial expression recognition, the method of active region normalization is proposed and applied for the first time. In this paper, the active region of human face is normalized, and the LBP and HOG features are extracted from the active region of the face. In order to locate the active region of human face more accurately, a human face alignment algorithm is applied in this paper. (3) LBP and HOG features are extracted from the normalized active region. And in the experiment of this paper, the two features are well fused together. Experiments show that the recognition rate of the two features is better than that of a single feature. In this paper, the gamma-ray correction method is applied to LBP features for the first time. This method improves the recognition rate of facial expressions to a great extent. (4) the 蟽 parameters are defined in this paper. Researchers can find the appropriate gamma correction value by finding the maximum value of the parameter. The experiments designed in this paper show that this parameter greatly reduces the workload of finding the appropriate gamma correction value. (5) in this paper, an algorithm framework based on manual feature extraction for recognition of six basic facial expressions is proposed. And the algorithm framework is applied to CK and JAFFE database. The algorithm proposed in this paper has obtained the best recognition rate in the CK database at present, and it has also obtained the competitive recognition result on the JAFFE at the same time.
【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP18
，
本文编号：2387492
[Abstract]:In recent years, artificial intelligence has been greatly developed in both academia and industry, especially after AlphaGo defeated Li Shishi at 4:1 in the go game, artificial intelligence technology has been attached great importance to by all people. As an important way of human-computer interaction, facial expression recognition technology has been paid more attention to, and some enterprises provide users with API to judge whether people smile or not. In view of the fact that six basic facial expression recognition techniques are not commercially available at present, this paper proposes a simple algorithm framework for facial expression recognition in static images, and implements the proposed algorithm under the MATLAB environment. Following the rapid development of artificial intelligence, the automatic recognition technology of six basic facial expressions in static images is studied in this paper. In order to verify the correctness of the proposed algorithm, the proposed algorithm is applied to CK and JAFFE databases. At the same time, the recognition rate obtained by the proposed algorithm is compared with that obtained in some literatures, which further demonstrates the correctness and effectiveness of the algorithm. The main contents and innovations of this paper are as follows: (1) this paper studies the correlation between six basic expressions and their neutral faces in static pictures. Karl Pearson correlation coefficient is applied to the research work of defining the active region of facial expression. This paper first compares the correlation between each expression and the static image of its neutral face, and then according to the magnitude of the correlation between each expression and its neutral face, The active region of facial expression is determined. (2) in the research of facial expression recognition, the method of active region normalization is proposed and applied for the first time. In this paper, the active region of human face is normalized, and the LBP and HOG features are extracted from the active region of the face. In order to locate the active region of human face more accurately, a human face alignment algorithm is applied in this paper. (3) LBP and HOG features are extracted from the normalized active region. And in the experiment of this paper, the two features are well fused together. Experiments show that the recognition rate of the two features is better than that of a single feature. In this paper, the gamma-ray correction method is applied to LBP features for the first time. This method improves the recognition rate of facial expressions to a great extent. (4) the 蟽 parameters are defined in this paper. Researchers can find the appropriate gamma correction value by finding the maximum value of the parameter. The experiments designed in this paper show that this parameter greatly reduces the workload of finding the appropriate gamma correction value. (5) in this paper, an algorithm framework based on manual feature extraction for recognition of six basic facial expressions is proposed. And the algorithm framework is applied to CK and JAFFE database. The algorithm proposed in this paper has obtained the best recognition rate in the CK database at present, and it has also obtained the competitive recognition result on the JAFFE at the same time.
【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP18
，
本文编号：2387492
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2387492.html
