基于离散隐马尔可夫模型和三维人脸表情的情感判决
- 来源：新浪科技 作者：橘子 时间：2008-05-05 20:45:16
-
--计算机科学
贺虎 高三
中国人民大学附属中学
摘要
传统的人机交互，缺乏对人情感的分析和理解。而配合着人脸表情识别，会使计算机更充分的了解人的情感和意图。因此，人脸面部表情识别技术是当今人机工程技术中的重要内容。
目前绝大部分表情识别工作，都基于二维静态图像数据，这使得表情识别的结果很受限制，因为表情是一个动态的过程。所不同的是，本系统选用了动态的时序三维表情数据作为情感判决的依据。为了解决三维表情识别在实际应用中的困难，本文还提出了一种将三维数据投影到二维平面的方法，兼顾了三维数据的高精确性和二维表情识别的便利性。
在具体实现过程中，本文采用Motion Capture获取的三维数据，并推倒了相应的离散隐马尔科夫模型，成功地对人脸的表情作出判决，并得到了较高的正确率。
DHMM Based Emotion Classification of 3D Face Expression
Computer Science
He Hu , Senior 3
The High School Affiliated to Renmin University of China
Abstract
In traditional human-machine interaction, we usually merely use mouses and keyboards without any judgment about human’s emotions. As a result, we can’t communicate with computers naturally as we communicate with other human beings.
Facial Expressions, the embodiment of intelligence, acting as the principle conveyable approach of emotion, are a significant method of people understanding emotions. For long, Expression Recognition has been a direction of importance in research of not only artificial intelligence but visual sphere of computer. What’s more, Expression Recognition is the essence of intelligence, as well as the natural communication between human beings and computers.
Therefore, techniques of Facial Expression Recognition are attached with increasingly greater importance. However, most of the present Facial Expression Recognition systems are based on static data, such as pictures. But these recognition results are limited, for expressing emotion is an active process. In this recognition system, emotion classification is based on active 3D data, which is about time-state. In all, this is the most significant difference from other Facial Expression Recognition Systems.
In this article, we describe how to make judgment about face expression based on 3D human face coordinates taken from Motion Capture. First, we perform training on existing sample data，and generate HMMs of four different kinds of emotion. When using, we input a set of expression data that needs to be judged into the system, compare it with the four HMMs, and find the most matching HMM. Finally, we output the emotion it represents, thus make the final emotion judgment. Through this method, we can get high correctness of emotion classification.
-
1
- [收藏] [推荐] [评论] [打印] [关闭] 点击：