人脸识别中光照梯度补偿算法的改进

光照变化是影响人脸识别的重要因素。光照梯度补偿是解决这一问题的方法之一。该方法通过最小二乘法计算出一个光照平面,然后将原图像减去光照平面,用所得的差图像来识别人脸。由于人脸的两半并非共面,它们与入射光线形成的角度是不同的,因此光照平面应该是不同的。如果分别计算两个光照平面又将导致差图像不连续。本文提出了一种算法,将两个半脸的光照平面系数合成一个系数向量,一同求解这些系数,然后调整其中的常系数项,使两边连续。从而实现,两边的光照平面不同,又保持了连续性。实验结果表明效果良好。

A modification for illumination gradient compensation algorithm in facial recognition

Variant lighting is a important factor which effect human face recognition.Illumination gradient compensation is a method for overcoming this problem.First,a illumination plane is calculated by means of least squares solution.Then subtract illumination plane from original image,the difference image is used for face recognition.Due to two half of human face is non-coplanar.The angle of two half face with respect to lighting is different.So the illumination plane is not the same for either half face.But if il...