摘要
考虑中心像素值对梯度幅值计算的影响,提出绝对尺度不变特征变换(ASIFT)描述子,结合主动形状模型(ASM)定位人脸特征点,提出ASM-ASIFT特征提取方法。将该方法所提特征与通过划分人脸表情区域并赋予权值的局部二值模式(LBP)特征进行结构化融合。实验结果表明,结构化融合后的正脸和侧脸表情识别的准确率分别为83.44%和71.19%,较ASM-ASIFT以及区域LBP方法,分别提高了4.77%、4.78%和6.98%、8.45%,表明融合后的特征能更加完整、精确地描述面部表情的细节信息,具有更强的表征能力。
Considering the influence of the central pixel value on the gradient magnitude calculation,an absolute scale invariant feature transform(ASIFT)descriptor was proposed.By combining active shape model(ASM)to locate facial feature points,ASM-ASIFT feature extraction method was proposed.The ASM-ASIFT feature and the local binary patterns(LBP)feature obtained by dividing the facial expression area and assigning weights were structured fused.Extensive experiments demonstrate that the accuracy of frontal and profile expression recognition after structured fusion is 83.44%and 71.19%,which is 4.77%,4.78%and 6.98%,8.45%higher than that of ASM-ASIFT method and regional LBP method respectively.The results show that the proposed method can extract the detailed information of facial expressions more completely and accurately,and the fusion features have stronger representation ability.
作者
黄倩露
王强
HUANG Qian-lu;WANG Qiang(School of Electronics and Information,Nantong University,Nantong 226019,China)
出处
《计算机工程与设计》
北大核心
2019年第11期3325-3330,共6页
Computer Engineering and Design
基金
江苏省科技成果转化专项基金项目(BA2015045)
南通市科技计划基金项目(GY2015010)
江苏省前瞻性联合研究基金项目(BY2016053-10)
南通大学研究生科研创新计划基金项目(YKC16013)
关键词
表情识别
LBP
尺度不变特征变换
主动外观模型
局部特征描述子
facial expression recognition
local binary pattern
scale invariant feature transform
active shape model
local feature descriptor
