年龄自适应人脸识别算法研究

项目名称： 年龄自适应人脸识别算法研究

项目编号： No.61305009

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 自动化技术、计算机技术

项目作者： 苏亚

作者单位： 北京科技大学

项目金额： 25万元

中文摘要： 年龄变化对面部产生的影响表现为非线性流形，使得现有的人脸识别方法不再适用，严重影响了人脸识别技术在日常生活中的广泛普及。克服年龄变化的人脸识别算法主要有两类，即年龄重建类方法和判别类方法。前者由于过度依赖年龄流形而引入了年龄估计误差，后者由于忽略年龄流形使得年龄变化的影响无法消除。为了更好的利用年龄流形，本课题基于贝叶斯理论对年龄变化下的人脸识别问题进行三个方面的探讨。首先，建立基于贝叶斯推理的人脸识别框架，将年龄的估计结果作为先验知识对识别问题进行指导，实现年龄自适应的人脸识别算法。其次，根据年龄对表观的影响呈现分段线性的现象，引入概率年龄段估计算法，并实现年龄段自适应的人脸识别。最后，基于年龄段自适应算法选择合适的特征提取算法。上述研究拟达到的目标是，将年龄变化下人脸识别的识别率提高5%，为多流形下的人脸识别提供新的思路。

中文关键词： 稀疏表示；人脸识别；年龄变化；贝叶斯推理；

英文摘要： Aging effect on the face appears as a nonlinear manifold. It makes current face recognition algorithms inapplicable, and hinders the technique from widely spread in everyday life. There are two kinds of methods attempting to conquer aging effect, i.e., age-reconstruction methods and discriminative methods. However, the former introduces age estimation error for over-reliance on age estimation. The latter fails to eliminate the aging effect because it neglects the nonlinear age manifold. In order to make better use of age manifold, the project discusses three aspects of face recognition under aging variation with the help of Bayesian theory. First, build Bayesian inference based face recognition, which guides the recognition task using age estimation prior. It results in age adaptive face recognition algorithm. Second, since aging effect appears to be piecewise linear, we introduce probabilistic age group estimation algorithm, which leads to age-group adaptive face recognition algorithm. Finally, select appropriate feature extraction algorithms based on the age-group adaptive face recognition algorithm. Above researches aim to improve face recognition under aging effect by 5%. It also provides a new viewpoint for face recognition under multiple manifolds.

英文关键词： sparse representation；face recognition；age variation；bayesian inference；