参数估计的人脸表情识别算法研究

发布时间：2018-05-27 16:11

本文选题：人脸表情识别 + 协作表示　；参考：《山东大学》2017年硕士论文

【摘要】：随着人机交互、情感分析及情感计算等技术的深入研究,人脸表情识别技术得到了飞速的发展。面部表情是非语言交流的一种重要表现形式,是人们理解情感的重要途径。人脸表情识别表现出重要的理论研究价值和实际应用价值,已逐渐成为人工智能与计算机视觉领域热门的研究方向。近年来,基于压缩感知的稀疏表示开始广泛地应用于目标跟踪、人脸识别等图像处理领域。然而,稀疏表示分类算法忽略了类别间的协作表示对分类的影响,协作表示分类算法既充分发挥了稀疏性对分类的优势,又兼顾了类别间的协作表示对分类的提升。但是,目前并没有有效的协作表示分类算法应用于人脸表情识别系统。针对以上问题,基于协作表示分类模型,本论文提出了基于协作表示参数估计的人脸表情识别算法,主要完成了以下三方面的研究:(1)基于l2范数的协作表示分类模型,本论文对协作表示保真项进行范数近似估计,并引入正则化因子对其改进。(2)在协作表示分类模型的基础上,本论文提出了一种基于最大似然估计的正则化加权协作表示分类算法及其模型。该算法通过对协作表示保真项作加权迭代分析,实现了人脸表情像素的自适应权值分配,降低了边界像素的识别干扰;通过对协作表示保真项作最大似然估计,使协作残差最小化,有效提高了人脸表情识别系统的有效性。(3)从贝叶斯估计的角度,本论文提出了基于最大后验估计的正则化加权协作表示分类算法及其模型。该算法通过对协作表示保真项作最大后验估计,引入先验因子,实现了对人脸表情识别系统的多角度、多层次的评估;通过对先验因子的分布参数作牛顿迭代估计,有效简化了人脸表情识别系统的算法复杂度。基于最大似然估计的正则化加权协作表示分类算法及其模型提高人脸表情识别的准确性和自适应性,而基于最大后验估计的正则化加权协作表示分类算法及其模型实现对人脸表情识别系统的多角度、多层次的评估。以上研究,为人脸表情识别的协作模型提供了一种行之有效的机制,同时提供了一种高精度、高鲁棒性的算法,为人脸表情识别系统的实际应用打下了坚实的基础。
[Abstract]:With the in-depth study of human-computer interaction, emotional analysis and emotional computing, facial expression recognition technology has developed rapidly. Facial expression is an important form of non verbal communication and an important way for people to understand emotion. Facial expression recognition shows important theoretical research value and practical application value. In recent years, the sparse representation based on compressed sensing has been widely used in target tracking, face recognition and other image processing fields. However, the sparse representation classification algorithm ignores the effect of the cooperative representation among categories, and the cooperative representation classification algorithm is not only sufficient. The superiority of the sparsity to the classification is brought into play, and the collaboration among categories is taken into account to improve the classification. However, there is no effective cooperative representation classification algorithm applied to facial expression recognition system at present. The following three aspects are completed mainly: (1) a cooperative representation model based on L2 norm. This paper estimates the norm of the fidelity item by cooperative representation and introduces the regularization factor to improve it. (2) on the basis of the cooperative representation classification model, a regularized weighting based on maximum likelihood estimation is proposed in this paper. A cooperative representation of the classification algorithm and its model. The algorithm realizes the adaptive weight distribution of the facial expression pixels and reduces the recognition interference of the boundary pixels by the weighted iterative analysis of the cooperative representation of the fidelity term. By using the cooperative representation of the fidelity term as the maximum likelihood estimation, the combined residual error is minimized and the facial expression recognition is effectively improved. The effectiveness of the system. (3) from the perspective of Bayesian estimation, this paper proposes a regularized weighted cooperative representation classification algorithm and its model based on the maximum a posteriori estimation. By introducing a maximum a posteriori estimation and a priori factor, the algorithm realizes the multi angle and multi-level evaluation of the facial expression recognition system. A Newton iterative estimation of the distribution parameters of a priori factor is used to effectively simplify the algorithm complexity of the facial expression recognition system. The regularized weighted cooperative representation based on maximum likelihood estimation and its model improve the accuracy and adaptability of facial expression recognition, and the regularized weighted cooperative table based on the maximum posterior estimation The classification algorithm and its model realize multi angle and multi-level evaluation of facial expression recognition system. The above research provides an effective mechanism for the cooperation model of facial expression recognition, and provides a high precision and robust algorithm, which lays a solid foundation for the practical application of face surface recognition system.
【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

【相似文献】