摘要
针对韦伯局部特征(WLD)仅计算中心像素与周围像素差异提取特征的不足,提出了一种韦伯梯度编码(WGC)特征描述的人脸表情识别算法。首先计算当前像素点周围水平、垂直和对角位置上的数值差与当前像素点的差异构成WGC特征的差动激励;然后进一步提出基于水平和对角线优先原则的WGC_HD特征;最后利用最佳分块方式得到行分块WGC_HD特征,采用自动优化参数的SVM分类器完成人脸表情识别。在公共人脸表情库JAFFE和CK库上进行交叉实验,平均识别率及平均特征提取时间分别为95.49%、12.30 ms和97.63%、31.54 ms。行分块WGC_HD特征考虑了不同梯度方向的像素差异,较好描述了表情图像的局部结构信息且具有较低的时间复杂度,与目前典型的表情识别算法结果对比也验证了算法具有较高的识别精度。
The traditional Weber local descriptor （WLD） algorithm has limitation in analyzing the center and neighboring pixels of the gray relationship. To identify facial expression accurately, a facial expression method based on Weber Gradient Coding （WGC） is proposed in this paper. First, the Weber Gradient encoding to the horizontal, vertical and diagonal gradient is respectively calculated to produce the differential excitation. Then, an optimized WGC operator based on horizontal and diagonal prior principle （WGCHD） is proposed. Finally, the SVM classifier is used to implement the facial expression recognition based on row block WGC_HD feature. The experiments on the proposed method are performed using JAFFE and Cohn-Kanade （CK） , the average recognition rate is 95.49%, 97. 63% and the average duration of the extraction is 12.30 ms and 31.54 ms, respectively. The row block WGC_HD characteristics considering the difference of pixels of different gradient direction well expressed in local structure information of facial images and has lower time complexity. The recognition results of proposed method are better tban those of tbe typical facial expression recognition method.
出处
《电子测量与仪器学报》
CSCD
北大核心
2017年第4期566-572,共7页
Journal of Electronic Measurement and Instrumentation
关键词
韦伯局部特征
韦伯梯度编码
水平和对角优先原则
人脸表情识别
Weber local descriptor （WLD）
Weber gradient coding （WGC）
horizontal and diagonal prior principle
facialexpression recognition
