摘要： 针对短时傅里叶变换频率分辨率较差的缺点，提出一种基于双树复小波变换(DTCWT)和局部二进制模式(LBP)直方图的低分辨率人脸识别方法。使用DTCWT获得人脸图像的多尺度多方向的频率幅度响应，采用LBP获取频率幅度响应的统计直方图，通过基于统计的一致性模式得到更加紧凑的统计分布特征。实验结果表明，该方法在低分辨率人脸上可以达到较高的识别准确率。

关键词: 人脸识别, 低分辨率, 双树复小波变换, 局部二进制模式, 特征提取, 一致性模式

Abstract: To address the frequency resolution insufficiency of Short-term Fourier Transform(STFT), a novel low-resolution face recognition method is proposed in this paper by combining Dual-tree Complex Wavelet Transform(DTCWT) and compact Local Binary Pattern(LBP) histogram. Face images are processed using DTCWT to obtain multi-scale and multi-direction frequency response, and then face features are conducted by applying LBP histogram. In addition, a statistical uniform pattern is introduced to improve the efficiency of the proposed algorithm. Experimental results show that this method has excellent recognition accuracy on low-resolution face recognition.

Key words: face recognition, low-resolution, Dual-tree Complex Wavelet Transform(DTCWT), Local Binary Pattern(LBP), feature extraction, consistency pattern

中图分类号: 

• TP391.4