基于LTP和局部PCA的低分辨率人脸识别算法

Low-resolution face recognition based on local ternary pattern and local principal component analysis

针对单特征提取人脸识别算法识别率较低的问题,提出一种基于多特征融合的低分辨率人脸识别算法.首先,利用局部三值模式(local ternary pattern,LTP)和局部主成分分析(principal component analysis,PCA)提取低分辨率人脸特征,将其分割成若干块并统计各子块的特征直方图;其次,融合各子块的局部主成分分析和局部三值模式的直方图并级联各个分块,作为新的人脸特征;最后,通过卡方距离度量训练集和测试集直方图的相似度,采用最近邻算法识别相似度.实验结果表明,所提算法对环境和光照变化更具鲁棒性,识别率得到有效提升.

On account of the problem of low recognition rate caused by single feature extraction algorithm,this paper comes up with a novel method which is based on multi-feature fusion.Firstly,the features of low-resolution face image are extracted by LTP(local ternary pattern)operator and local PCA(principal component analysis)operator,then two facial features are divided into several blocks and histograms of each block are calculated;secondly,the histograms of each block on LTP and local PCA are linearly concatenated as well as each block is cascaded as a final facial feature;lastly,Chi-square distance is used to measure the histogram similarities between training sets and testing sets.The nearest neighbor algorithm is adopted for recognition.The experimental results show that the proposed method is more robust to changes of illumination and noise,moreover it further improves the accuracy of face recognition.