高斯核方向导数和RILPQ融合的人脸表情识别

Facial Expression Recognition Based on Gaussian Kernel Direction Derivative and RILPQ

针对人脸表情识别中特征提取出纹理信息粗糙、边缘轮廓不清的问题,论文提出了一种基于高斯核方向导数与RILPQ相结合图像特征提取方法。在RILPQ算法中引入高斯核多方向导数形成滤波器,在支持向量机中进行表情分类,将算法应用于JAFFE数据集表情数据集。实验结果为在滤波窗口半径为11像素,论文算法识别率最优,并高于LPQ算法、RLPQ算法识别率。同时也证明,高斯窗窗口半径和滤波方向对算法的表情识别率有影响。

In view of the problem that the feature of facial expression recognition is not clear,this paper proposes a method of image feature extraction based on Gauss kernel direction derivative and RILPQ. In the RILPQ algorithm,the Gauss kernel multi direction derivative is introduced to form a filter,and the algorithm is applied to the expression data set of JAFFE data set. Experimental results for the filter window radius of 11 pixels,the algorithm recognition rate is optimal,and higher than the LPQ algorithm,RLPQ algorithm recognition rate. At the same time,it is proved that the Gauss window radius and the direction of filtering have effect on the recognition rate of the algorithm.