Abstract: A face recognition algorithm based on coefficient and statistical features of block DCT is proposed. The image is partitioned into a set of blocks, choosing the DCT coefficient of the low frequency part as its features. At the same time the DCT transform is performed on each block in order to decompose it into a low-pass filtered image and a reversed L-shape blocks containing the high frequency coefficients of the DCT; the statistical measures such as mean, variance, and entropy rate are then computed on the low-pass filtered image and a reversed L-shape blocks; SVM and the nearest neighbor classifier are selected to perform face classification. The experimental results on ORL and Yale face databases show that the algorithm based on statistical features of block DCT achieves high recognition rate.

Key words: Discrete Cosine Transform（DCT）, Support Vector Machine（SVM）, nearest neighbor classifier, statistical features, face recognition

摘要： 提出了一种基于分块DCT系数及其统计特征的人脸识别算法。对图像进行分块，对每一块进行DCT变换，选择低频部分的系数作为识别的特征，将每一块分解为一幅低通滤波图和一个包含DCT高频系数的反L型块；分别对这两块求其均值、方差和熵这三个统计特征；利用支持向量机（SVM）和最近邻分类器对这些特征进行分类识别。在ORL、Yale人脸数据库上的仿真实验表明，使用基于分块DCT系数及其统计特征可达到较高的识别率。

关键词: 离散余弦变换（DCT）, 支持向量机, 最近邻分类器, 统计特征, 人脸识别