摘要
传统局部二值模式(LBP)算法应用在人脸表情识别中,不能准确描述眼睛、嘴巴、额头等表情特征区域在不同方向上的灰度变化趋势,识别效果不理想。本文改进传统局部二值模式的灰度比较关系,分别从水平、垂直以及对角3个方向对邻域像素的灰度变化进行二值编码,融合3个方向的特征,得到一种基于方向性的局部二值模式(DLBP)。在JAFFE数据库和Cohn-Kanade数据库上的实验结果均表明,DLBP算子相比LBP算子、Gabor算子能更准确描述人脸基本表情,识别率平均分别提高了5%和1%;相比LBP算子对椒盐噪声和高斯白噪声具有更强的鲁棒性;且与LDP算子相比,识别率基本不变,但特征提取时间缩减近50%。由此可见,DLBP算子是一种快速有效的人脸表情描述子。
The traditional local binary pattern( LBP) algorithm for facial expression recognition could not describe the gray value change in different directions of somel expression regions,such as eyes,mouth,forehead,etc. The recognition result is not satisfied. This paper presents a simple and robust method,namely local binary pattern based on the directions( DLBP),which improves the coding pattern of LBP and encoded the difference from the horizontal,vertical and diagonal directions. Experimental results on JAFFE and Cohn-Kanade databases show that DLBP algorithm has achieved 5% and 1% higher recognition rates than other existing algorithms,such as LBP and Gabor. It has a strong robustness to Gaussian noise and salt and pepper noise compared with LBP,and Its feature extraction time is reduced by 50% compared to LDP. Therefore,the DLBP algorithm is a fast and effective feature descriptor.
出处
《智能系统学报》
CSCD
北大核心
2015年第3期422-428,共7页
CAAI Transactions on Intelligent Systems
基金
江苏省自然科学基金资助项目(BK20131342)
关键词
人脸表情识别
局部二值模式
中心最近邻分类
方向性局部二值模式
Gabor:LDP
facial expression recognition
local binary pattern(LBP)
central nearest neighbor classification
directional local binary pattern(DLBP)
Gabor
local directional pattern(LDP)