Face Detection with Expression Recognition using Artificial Neural Networks
Face Detection with Expression Recognition using Artificial Neural Networks
复制标题
使用人工神经网络进行面部检测和表情识别
DOI：
10.1109/mecbme.2016.7745421
复制
发表时间：
2016
期刊：
影响因子：
--
通讯作者：
Moussa Iskandar
中科院分区：
文献类型：
--
作者：
M. Owayjan;Roger Achkar;Moussa Iskandar
本文提出了一种使用人工神经网络进行表情识别的人脸检测系统。它是一个使用MATLAB设计和实现的自动化视觉系统。面部表情识别系统通过两个阶段完成面部表情识别。首先对捕获的图像进行处理以检测人脸，然后识别面部表情。这两个阶段分五个阶段完成。系统的前两个阶段使用图像处理（特别是 Viola-Jones 对象检测框架）来检测和裁剪面部。第三阶段涉及将裁剪图像的颜色从 RGB 转换为灰度并应用适当的平滑滤波器。第四阶段包括使用人工神经网络进行特征提取，以便将提取的特征与训练样本进行比较。最后阶段对给定的输出进行分类并显示面部表情识别结果。然后它确定主体是快乐、愤怒还是处于中立状态。人工神经网络使用多层感知器（MLP）和反向传播算法进行特征提取和分类。它有 4097 个输入节点、1 个包含 50 个神经元的隐藏层和 1 个输出层。测试结果表明，该系统可用于解读三种面部表情：快乐、愤怒和中性。它提取准确的输出，可用于其他研究领域，例如心理评估。最后，结果的高精度允许未来开发实时响应自发面部表情的不同应用程序。
This paper presents a Face Detection System with Expression Recognition using Artificial Neural Networks. It is an automated vision system designed and implemented using MATLAB. The Face Detection with Expression Recognition system accomplishes facial expression recognition through two phases. The captured image is processed first to detect the face, and then the facial expression is recognized. These two phases are completed in five stages. The first two stages of the system deal with detecting and cropping the face using image processing, in particular the Viola-Jones object detection framework. The third stage deals with converting the colors of the cropped image from RGB into gray scale and applying the appropriate smoothing filter. The fourth stage consists of feature extraction using Artificial Neural Networks, so as the extracted features are compared with training samples. The final stage classifies the given outputs and shows facial expression recognition results. It then determines whether the subject is happy, angry or in neutral state. The Artificial Neural Network uses Multi-Layer-Perceptron (MLP) with back propagation algorithm for features extraction and classification. It has 4097 input nodes, one hidden layer with 50 neurons, and one output layer. Testing results show that this system can be used for interpreting three facial expressions: happiness, anger and neutral. It extracts accurate outputs that can be employed in other fields of studies such as psychological assessment. Finally, the high precision of the results allow future development of different applications which respond to spontaneous facial expressions in real time.