基于兴趣点的人脸识别流形算法

MANIFOLD ALGORITHM OF FACE RECOGNITION BASED ON INTEREST POINTS

在人脸识别应用领域中,已有的流形算法都是直接使用像素间的内部结构进行识别,而没有考虑像素点在人脸图像中所起的作用大小。为了克服上述缺点,基于概率密度的兴趣点检测,首先对图像进行加权计算,经过加权后的图像的目标特征更加明显,不易受视角和形状变化的影响;然后使用流行算法进行识别。在标准人脸库上的实验结果表明,该方法达到了较高的识别率,能提高流形算法的识别效果。

In face recognition applications,existing manifold algorithms are all use directly the internal structure between pixels to recognise but not consider the roles played by the pixels themselves in face image.To get over this deficiency,first,we carry out the weighted computation on the image based on interested points of probability density,the weighted image has more distinct target characteristics and is less sensible to the variations of visual angle and shape;then we use the manifold method to recognise the face.Experiments on standard face databases show that the proposed algorithm achieves quite high recognition rate and is able to improve the recognition effect of the manifold method.