【作者】 黄华； 何惠婷；

【Author】 Hua Huang, Huiting He School of Electronics and Information Engineering, Xi’an Jiaotong University, Xi’an 710049, China

【机构】 西安交通大学电子与信息工程学院；

【摘要】 针对低分辨率人脸图像识别率较低的问题,提出了一种利用识别特征的超分辨率重建得到低分辨率人脸图像在高分辨率空间对应的识别特征的方法。基于流形学习的理论,认为高低分辨率人脸图像的识别特征是由共同的内在结构生成的,利用典型相关分析增强高低分辨率人脸图像识别特征邻域关系的一致性,以更好地满足邻域重构思想的假设;在典型相关分析变换得到的相关子空间内利用邻域重构获得测试低分辨率人脸图像对应的高分辨率人脸图像识别特征,最后利用此特征识别人脸。实验表明,本文方法所得识别率受图像分辨率大小影响较小,所得识别率较高。更多还原

【Abstract】 Since the problem of low resolution facial images resulting in low recognition rate, an algorithm is proposed utilizing features in the high resolution space which are obtained from feature-domain super-resolution reconstruction. The theory of manifold learning assumes that recognition features of high and low resolution facial images are generated from a similar internal structure. Applying canonical correlation analysis to enforce the consistency of neighborhood relations between the recognition features of high and low resolution facial images will more satisfy the assumption of neighborhood reconstruction. Then neighborhood reconstruction is applied in the correlation subspace after canonical correlation analysis transformation to obtain the high resolution facial images recognition features corresponding to the testing low resolution facial images. Finally, we utilize the features for face recognition. Experiment results show that the proposed algorithm is less influenced by the resolutions of face images, and has higher recognition rate.更多还原