基于双密度双树复小波变换多字典的人脸特征稀疏分类方法

Sparse representation of face feature recognition based on multiple dictionaries of double-density dual-tree complex wavelet transform

基于超完备字典的人脸稀疏表示方法的难点是其字典构成。针对此问题,首先采用双密度双树复小波变换(DD-DT CWT)提取人脸图像不同尺度的高频子带,然后根据能量平均分布最大原则选择能量较大的部分子带构成对应尺度的超完备字典。同时,将测试样本相应的人脸DD-DT CWT子带特征看成超完备字典中原子的线性组合,并组合多字典上的稀疏表示进行识别。在AR人脸图像库上进行了实验,结果表明该方法是一种有效的人脸特征表示及分类方法。

The difficulty in sparse representation of facial images based on over-complete dictionary is the dictionary generation.This paper first introduced the Double-Density Dual-Tree Complex Wavelet Transform（DD-DT CWT） for filtering the high-frequency sub-bands and the principle of energy distribution for selecting some sub-bands as the feature of a facial image to form multi-scale dictionaries,then viewed the similar feature of a test sample as the linear combination of some atoms in the overcomplete dictionary,finally got the recognition results via ensembling sparse representations on these dictionaries.The experimental results on AR face database demonstrate the efficiency of the proposed algorithm.