双差值局部方向模式的人脸识别

针对差值局部方向模式（DLDP）特征提取不够充分和对光照、噪声等比较敏感的问题，提出一种双差值局部方向模式（DDLDP）人脸识别方法。首先，分别将半径为1的3×3领域像素灰度值和半径为2的5×5领域像素灰度值与8个Kirsch模板算子卷积，得到两组对应8个灰度响应值。然后，将半径为1的灰度响应值，按照相邻前后作差的方式，得到8个灰度响应差值，再将半径为1和2得到的灰度响应值上下作差，也得到8个灰度响应差值。最后，将两组灰度响应差值取绝对值，其最大绝对值所对应下标位置构成DDLDP码。仿真实验结果表明，相比同类基于局部方向模式的单一人脸识别算法，该方法具有更好识别效果。DDLDP更加完整地提取了人脸特征，且表现出对光照和噪声更好的鲁棒性。
To solve the problem of insufficient feature extraction and sensitivity to the noise and illumination encountered using difference local directional pattern (DLDP) method, this article proposes a double difference local direction pattern (DDLDP) face feature extraction method. First, a 3×3 domain pixel gray value with a radius of 1 and a 5×5 domain pixel gray value with a radius of 2 were convolved with eight Kirsch template operators to obtain two groups of eight gray response values. Then, the gray-scale response value with a radius of 1 was obtained as a difference between the values of the neighboring pixels at both sides to obtain eight gray-scale response differences. Meanwhile, the edge response difference values of different radius were also calculated. Finally, the two sets of gray response differences were taken as absolute values, and their maximum absolute values correspond to the subscripts form DDLDP code. Simulation experiment results show that the proposed algorithm has better recognition effect than other single face recognition algorithms based on local directional pattern (LDP). The DDLDP algorithm can fully extract the facial features, and have better robustness to illumination and noise