双运算局部方向模式的人脸识别算法

Face Recognition Based on Double-Operation Local Directional Pattern

针对双空间局部方向模式(DSLDP)人脸识别算法只是单一采用作差运算提取特征的问题,提出一种双运算局部方向模式(DOLDP)的人脸识别方法。首先,将图像3pixel×3pixel邻域像素灰度值与8个Kirsch模板算子卷积,得到8个方向的边缘响应值;然后,将近邻边缘响应值按照逆时针方向分别作差和作和,得到两组8个方向的边缘响应差值和和值,将两组边缘响应值取绝对值,取各自最大值的方向编码成一个二位八进制数,构成DOLDP码。在YALE、ORL、AR和CAS-PEAL人脸库上的实验结果表明:该方法将和值空间和差值空间人脸特征信息结合,取得了更好的识别效果;和值空间人脸特征信息较强度空间起到了平滑作用,对光照、表情、遮挡等情况表现出更强的稳健性。

Aiming at the problem that the face recognition algorithm based on double space local directional pattern （DSLDP） only uses differential operation to extract features, we present a novel approach based on the double operation local directional pattern （DOLDP） for face recognition. Firstly, 3 pixelX 3 pixel neighborhood of facial image are convolved with eight Kitsch template operators to obtain eight directions of edge response values. Then, the neighboring edge response values are countered and summed in counterclockwise directions to obtain two sets of eight-direction edge response differences and sums, and the two sets of values are taken as absolute values. Finally, the directions of the maximum values of the two sets of edge response values are encoded into a two-digit octal number to form a DOLDP code. The experimental results on YALE, ORL, AR and CAS-PEAL face databases show that the proposed method combines the sum space and the difference space face feature information, and achieves a better recognition effect. Compared with the intensity space, the sum space face feature information plays a smooth role and shows stronger robustness to light, expression, and occlusion.