基于自回归模型的动态表情识别
发布时间：2018-10-13 19:19
【摘要】：采用几何信息和纹理信息融合的混合特征,基于自回归(AR)模型,提出一种基于线段的相似度判决方法实现动态表情识别.首先在6种基本表情的图像序列训练集上进行训练得到6种AR模型,然后给定测试表情序列,对每一个测试序列通过6种AR模型生成6种预测序列,接着比较每种预测序列与实际给定序列的相似性,最终根据相似性判断所给序列的表情类别.为了更好地比较预测序列与给定序列的相似性,提出了一种基于线段的相似度判决方法.基于Cohn-Kanade+人脸表情库进行实验结果表明,该方法在动态表情识别上取得了良好的效果.
[Abstract]:Based on the autoregressive (AR) model, a similarity decision method based on line segments is proposed to realize dynamic facial expression recognition based on the hybrid features of geometric information and texture information fusion. At first, six kinds of AR models are trained on the image sequence training set of six basic expressions. Then, given the test expression sequence, six prediction sequences are generated for each test sequence through six AR models. Then, the similarity between each predicted sequence and the actual given sequence is compared, and the facial expression category of the given sequence is determined according to the similarity. In order to better compare the similarity between predicted sequences and given sequences, a similarity decision method based on line segments is proposed. The experimental results based on Cohn-Kanade facial expression database show that the proposed method is effective in dynamic facial expression recognition.
【作者单位】： 华中师范大学国家数字化学习工程技术研究中心;华中师范大学教育信息技术协同创新中心;
【基金】：国家社会科学基金(16BSH107)
【分类号】：TP391.41
[Abstract]:Based on the autoregressive (AR) model, a similarity decision method based on line segments is proposed to realize dynamic facial expression recognition based on the hybrid features of geometric information and texture information fusion. At first, six kinds of AR models are trained on the image sequence training set of six basic expressions. Then, given the test expression sequence, six prediction sequences are generated for each test sequence through six AR models. Then, the similarity between each predicted sequence and the actual given sequence is compared, and the facial expression category of the given sequence is determined according to the similarity. In order to better compare the similarity between predicted sequences and given sequences, a similarity decision method based on line segments is proposed. The experimental results based on Cohn-Kanade facial expression database show that the proposed method is effective in dynamic facial expression recognition.
【作者单位】： 华中师范大学国家数字化学习工程技术研究中心;华中师范大学教育信息技术协同创新中心;
【基金】：国家社会科学基金(16BSH107)
【分类号】：TP391.41
【参考文献】
相关期刊论文 前8条
1 王大伟;周军;梅红岩;张素娥;;人脸表情识别综述[J];计算机工程与应用;2014年20期
2 罗元;崔叶;王艳;张毅;;基于融合DCT和LBP特征提取的面部表情识别[J];半导体光电;2014年02期
3 易积政;毛峡;Ishizuka Mitsuru;薛雨丽;;基于特征点矢量与纹理形变能量参数融合的人脸表情识别[J];电子与信息学报;2013年10期
4 蒋斌;贾克斌;杨国胜;;人脸表情识别的研究进展[J];计算机科学;2011年04期
5 周书仁;梁昔明;朱灿;杨秋芬;;基于ICA与HMM的表情识别[J];中国图象图形学报;2008年12期
6 詹永照;李婷;周庚涛;;基于混合特征和多HMM融合的图像序列表情识别[J];计算机辅助设计与图形学学报;2008年07期
7 唐京海;张有为;;基于动态序列特征的人脸表情识别方法[J];计算机工程与应用;2008年08期
8 刘晓e，
本文编号：2269650
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2269650.html
