基于深度学习的人脸表情识别研究与实现

发布时间：2018-03-01 06:38

本文关键词： 表情识别深度学习特征提取人脸检测人脸关键点检测　出处：《西南科技大学》2017年硕士论文　论文类型：学位论文

【摘要】：人脸表情识别是机器视觉和模式识别等领域的一大富有挑战性的研究课题。人脸表情识别一直作为领域内热门的研究方向,具有广泛的应用场景,例如人机交互、安全监控,谎言检测等等。随着相关领域技术的发展,特别是自2006年以来深度学习(Deep Learning)技术的飞速发展,省去了人工设计和提取特征的步骤,换之以大数据样本训练和自动学习有效特征,从而大幅度提高算法模型准确度和适应性。因此,结合深度学习进行人脸表情识别也成为当下热点之一。本文针对二维人脸图像进行表情分类识别研究。提出一种基于降噪自编码器的人脸表情识别算法,在提高识别准确率的同时能够有效降低表情之间的干扰程度;设计一种轻量级的卷积神经网络,能够快速和较为有效地识别人脸表情;提出结合卷积神经网络和循环神经网络的表情识别方法,针对人脸图像序列来识别表情。同时,提出一种数据筛选框架和两种较为简单的数据筛选方法,来对数据数据进行预处理。
[Abstract]:Facial expression recognition is a challenging research topic in the fields of machine vision and pattern recognition. As a hot research direction in the field, facial expression recognition has a wide range of applications, such as human-computer interaction, security monitoring, etc. With the development of technology in related fields, especially the rapid development of deep learning technology since 2006, the steps of artificial design and feature extraction have been eliminated, and big data sample training and automatic learning effective features have been replaced. Thus greatly improving the accuracy and adaptability of the algorithm model. Facial expression recognition based on depth learning has become one of the hotspots. In this paper, facial expression recognition algorithm based on de-noising self-encoder is proposed. At the same time, it can effectively reduce the degree of interference between expressions, and design a lightweight convolution neural network, which can recognize facial expressions quickly and effectively. An expression recognition method based on convolutional neural network and cyclic neural network is proposed to recognize facial expression for face image sequence. At the same time, a data screening framework and two simple data selection methods are proposed. To preprocess the data.
【学位授予单位】：西南科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】