基于局部特征和集成学习的鲁棒彩色人脸识别算法
Robust color face recognition algorithm based on local features and ensemble learning

为了充分利用人脸图像的局部信息、改善现有基于整体特征的彩色人脸识别算法的识别率,提出了一种基于局部特征和集成学习分类器的鲁棒彩色人脸识别算法.在特征提取阶段,使用自适应四元数pseudo-Zernike矩(AQPZMs)来描述图像子块的特征.对于具有较大熵的图像子块使用较高阶次的四元数pseudo-Zernike矩(QPZMs)提取特征,反之则使用较低阶次的QPZMs.在匹配识别阶段,使用集成学习分类器进行判别.针对不同彩色人脸图像库的测试结果表明,当人脸图像受到光照、表情等因素影响时,与采用QPZMs或者四元数二维主成分分析(Q2DPCA)进行整体特征提取的识别算法相比,所提算法的识别率更高.
To make full use of local information of face images and improve the recognition rate of the existing color face recognition algorithm based on global features, a robust color face recognition algorithm based on local features and ensemble learning classifier is proposed. In the feature extraction stage, the adaptive quaternion pseudo-Zernike moments(AQPZMs)are used to describe the features of image blocks. The features of image blocks with larger entropy are described by quaternion pseudo-Zernike moments(QPZMs)with higher order. On the contrary, the QPZMs with lower order are used to describe the features of image blocks with smaller entropy. In the classification stage, the ensemble learning classifier is used for identification. The experimental results of different color face datasets show that compared with the recognition algorithms exploiting QPZMs or quaternion two-dimensional principal component analysis(Q2DPCA)to extract global features, the proposed algorithm can achieve higher accuracy when the face images are affected by the factors such as illumination, facial expression and so on