【作者】 刘晓山； 曾春艳； 杜明辉；

【Author】 LIU Xiaoshan~(1)2)),ZENG Chunyan~(1)),Du Minghui~(1)) 1)(School of Electronic and Information Engineering,South China University of Technology,Guangzhou 510640,china) 2)(School of Physics and Communication Electronics,Jiangxi Normal University,Nanchang 330022,China)

【机构】 华南理工大学电子与信息学院； 江西师范大学物理与通信电子学院；

【摘要】 研究了基于双树复小波变换(Dual-Tree Complex Wavelet Transform,DTCWT)的人脸特征提取方法,并且针对不同系数组合形成的分类特征在多个公用的人脸库上进行了对比实验。实验结果表明将DTCWT的各层高频复系数的振幅作为人脸的分类特征识别性能最优,且不需要进行下采样操作。更多还原

【Abstract】 Methods of facial feature extraction based on DTCWT are discussed and comparison experiments on several common face databases are conducted to investigate the classification performance of combination of different coefficients.Experiment results show that the scheme that select amplitude of all high frequency coefficients as facial classification features has the best performance and this scheme does not require down-sampling operation.更多还原