摘要
提出了一种基于图像融合和DCT域增强的人脸识别方法.首先对人脸图像分别进行直方图均衡化和对指数变换,将二者变换结果融合,减少光照对人脸识别的影响;然后进行DCT变换,在滤除高频分量的同时进行图像增强,抑制块效应,在此基础上进行IDCT图像重建,利用二维主成分分析提取人脸特征,降低空间维数;最后通过最邻近分类法实现人脸识别.Yale人脸库仿真实验表明,该方法在光照变化较大和人脸样本较少的情况下具有较高的识别率.
A method for face recognition using image preprocessing fusion and DCT enhancement is pro- posed to address the problem of variable illumination in face image and block effect in DCT. Firstly, histo- gram equalization and logarithm transformation are respectively used to preprocess the face image, and then two processed images are fused to eliminate the illumination effects. DCT domain enhancement is used to filter the high-frequency components and reduce the block effect. IDCT is employed to image reconstruction and two-dimensional principal component analysis （2DPCA） is used to extract face feature,which may re- duce the feature dimensions and low the computational cost. Finally, the nearest neighbor classification is used to complete the face recognition. The simulations on the Yale face database show that this method has better recognition rates, especially for the face image with large variable illumination and the smaller num- ber of face samples.
出处
《兰州交通大学学报》
CAS
2013年第6期1-5,共5页
Journal of Lanzhou Jiaotong University
基金
国家自然科学基金项目(61261029)
高等学校基本科研业务费项目(212090)
金川公司预研基金资助(JCYY201309)
关键词
人脸识别
离散余弦变换
图像增强
二维主成分分析
face recognition
discrete cosine transform（DCT）
image enhancement
two-dimensional prin- cipal component analysis
