摘要:
本文提出了一种基于离散余弦变换与Wallis的人脸光照处理算法，该算法首先将人脸图像变换到对数域，在对数域中计算离散余弦变换(DCT)，舍弃部分低频DCT系数，再计算其离散余弦反变换。然后用Wallis算法对人脸图像的高频部分进行增强。在人脸识别阶段，采用主成分分析法(PCA)提取人脸特征，运用基于余弦距离的最近邻分类器进行分类判别。在Yale B正面人脸库中的实验结果表明，本文提出的方法可以削弱人脸光照的影响，合理选择相关参数，人脸识别率能达到好的效果。
Abstract:
In this paper, a novel approach based on discrete cosine transform (DCT) and Wallis for face illumination is discussed. Firstly, the DCT is calculated in logarithm domain for face image. Some low-frequency coefficients are discarded in zigzag pattern. Secondly, after inverse discrete cosine transform (IDCT), the Wallis algorithm is used to enhance the high-frequency detail of face image. Thirdly, the principal component analysis (PCA) and the nearest neighborhood classifier using cosine distance are adopted for face recognition. The experiment results on Yale B frontal face database demonstrate that the presented algorithm can decrease the influence of face illumination. The face recognition rate has a good effect when some parameters are chosen properly.