【摘要】 现有的多数人脸识别系统都要依赖于面部特征(比如眼睛中心位置)的严格配准来归一化人脸以便提取人脸描述特征,但面部特征配准的准确度如何影响人脸识别算法的性能却没有得到足够的重视.该文作者首次针对这一问题进行了系统的研究,并提出了一种基于误配准学习的解决方案.为了揭示现有典型识别算法的识别性能对特征配准准确度的敏感程度,通过对眼睛位置人为加扰,作者对Fisherface算法的识别性能随平移、旋转和尺度改变而变化的情况进行了实验评估.结果表明Fisherface的识别性能随着误配准的增大而急剧下降——称这一现象为“误配准灾难”问题.针对此问题,作者提出了一种基于扰动学习的“误配准灾难”解决方案,该方法通过在模型训练阶段加入扰动配准偏差来提高判别分析方法对误配准的鲁棒性.在FERET人脸图像数据库和CAS PEAL R1人脸库上的实验表明该方法可以有效地提高识别算法对误配准的鲁棒性.更多还原

【Abstract】 In this paper, authors investigate the rarely concerned curse of mis-alignment problem in face recognition, and propose a novel perturbation learning solution. Mis-alignment problem is firstly empirically investigated through evaluating systematically the Fisherface’s sensitivity to mis-alignment on the FERET face database by perturbing the eye coordinates, which reveals that the imprecise localization of the facial landmarks abruptly degrades the Fisherface system. Authors explicitly define this problem as curse of mis-alignment for highlighting its significance. Aiming at this problem, authors propose a set of measurement combining the recognition rate with the alignment error distribution to evaluate the overall performance of specific face recognition approach with its robustness against the mis-alignment considered. Finally, a perturbation learning method,named E-Fisherface, is proposed to reinforce the recognizer to model the mis-alignment variations in the training stage.Experimental results on FERET and CAS-PEAL-R1 have impressively indicated the effectiveness of the proposed E-Fisherface to tackle the curse of mis-alignment problem.更多还原