OD-LBP与加权HOG特征融合表情识别方法研究

Research on expression recognition based on feature fusion of OD-LBP and weighted HOG

面部表情识别是人机交互领域的重要核心,丰富的面部表情特征是提高面部表情识别率的关键之一。正交差分局部二值模式(Orthogonal difference-local binary pattern,OD-LBP)和方向梯度直方图(Histogram of Oriented Gradient,HOG)的融合特征可以很好地表达出面部表情的局部和全局特征信息,但是没有考虑到面部不同部分对表情识别贡献程度不同。因此提出了一种先对人脸图像中的面部表情敏感区域提取OD-LBP特征,再将人脸图像均匀分块并提取HOG特征,计算每子块的改进空间频率值对HOG特征加权,然后与OD-LBP特征融合形成新的特征,并利用主成分分析(Principal Component Analysis,PCA)降维,最后利用分类器中的支持向量机(Support Vector Machine,SVM)完成特征分类的面部表情识别方法。基于Pycharm平台,在表情数据集JAFFE和CK上验证该算法的有效性。仿真实验结果表明,该算法的表情识别率分别为95.4%和96.9%,较未考虑区域重要性的融合特征的识别率提高了2.2%和2.1%,且在不同姿态、光照条件下具有良好的鲁棒性。

Facial expression recognition is an important core in the field of human-computer interaction,enriching the facial expression features is the key to improve the recognition rate of facial expressions.Fusion features of Orthogonal Difference-local Binary Pattern(OD-LBP)and Histogram of Oriented Gradient(HOG)can well express the local and global feature information of facial expressions,but not consider that different parts of the face image contribute differently to facial expression recognition.Therefore,the paper proposes a method of facial expression recognition.Firstly,the research extracts the OD-LBP feature from the facial expression sensitive area in the face image,divides the face image into blocks to extract the HOG feature,and calculates the improved spatial frequency value of each sub-block to weight the HOG feature.Then the paper fusions with OD-LBP feature to form a new feature,and uses PCA algorithm for dimensionality reduction.Finally,the paper adopts support vector machine(SVM)for feature classification.Based on the Pycharm platform,the effectiveness of the algorithm is verified in expression datasets such as JAFFE and CK.The experimental results show that the expression recognition rate of the algorithm reaches 95.4%and 96.9%respectively.Compared with the fusion feature that does not consider the importance of the area,the expression recognition rate is improved by 2.2%and 2.1%,and it has good robustness under different postures and lighting conditions.