摘要
针对传统表情识别系统不能充分提取关键子区域及有效特征的缺陷,设计了基于关键子区域及特征提取的表情识别系统。首先使用面部关键点检测技术及面部编码系统筛选出关键子区域;然后对其进行特征提取。提出一种改进的局部梯度编码算子(LGC)、局部均值梯度编码算子(LMGC-HD);改进的算子具有更低的维度,能够充分地描述局部形变;且受随机噪声及边缘变化影响小。最后使用支持向量机(SVM)进行分类识别。采用CK+数据集进行实验,结果证明该系统能够有效地提高人脸表情的识别率。
An expression recognition system based on key sub-region and feature extraction is designed for the traditional expression recognition system can not fully extract the key sub-regional and effective features. First,the key point detection technology and facial coding system were used to filter the key sub-area. Then,an improved LGC operator is proposed： local mean gradient coding operator（ LMGC-HD）,the improved operator has a lower dimension,can fully describe the local deformation,and by random noise and the influence of edge change is small.Finally,SVM is used for classification and recognition. Using CK ＋ database,it is proved that the system can effectively improve the recognition rate of facial expression.
出处
《科学技术与工程》
北大核心
2017年第34期257-262,共6页
Science Technology and Engineering
