远距离人脸识别中低质图像增强关键理论研究

Face recognition at a distance (FRAD) has recently become an important research topic. It aims at solving problems associated with the recognition of human face images under the challenging condition imposed by taking pictures at a distance (e.g., lower resolution, quality degradation, complex illuminance, and varying shoot angles). On the basis of the existing research results and considering the characteristics of face images taken in distance, we propose to pursue several in-depth studies as follows: (1) the degradation mechanism of recognizable features within distant face images and the establishment of a theoretical model for this degradation process; (2) matching of heterogeneous face images that are captured at varying distances and different illuminances; (3) face image enhancement methods for a better set of recognizable features and a higher recognition accuracy; and (4) cross-distance mapping relationship between probe face images taken at large distance and gallery face images taken at small distance. The proposed research project is expected to achieve a breakthrough in solving the problems associated with face recognition at distance so as to lay a theoretical and technical foundation for the future applications.

远距离人脸识别是目前人脸识别中的一个热门研究方向，其关键在于解决远距离条件下人脸图像的分辨率低、成像质量退化以及成像设备差异、视角多变等复杂因素下形成的异质性人脸图像识别问题。本项目拟在相关研究的基础上，结合远距离人脸低质图像的特点，研究人脸图像在随距离增加的过程中其可辨识特征的退化机理并建立能描述该过程的理论模型；解决因距离变化、不同光照环境引起的异质人脸图像匹配问题；研究面向识别的人脸图像增强方法，建立可识别特征的恢复和增强模型以提高人脸识别的准确率；研究远距离人脸测试图像和样本图像之间的异质化匹配问题，建立通过测试样本和训练样本进行远距离异质人脸图像高质量重构的理论模型，解决远距离人脸图像的增强匹配问题。本项目预期在解决该系列问题的理论上有所突破，方法上有所创新，为远距离人脸识别的应用研究奠定一定的理论和技术基础。

远距离人脸识别是目前人脸识别中的一个热门研究方向，其关键在于解决远距离条件下人脸图像的分辨率低、成像质量退化以及成像设备差异、视角多变等复杂因素下形成的异质性人脸图像识别问题。本项目在前期研究的基础上，开展了远距离人脸小图像降质模型、特征提取、暗图增强等方面的研究。提出了人脸图像在距离增加过程中特征增强和提取的新模型，形成了距离变化、光照变化情况下异质人脸图像的匹配方法；开展了面向识别的人脸图像增强方法，建立的增强模型有效恢复了人脸识别特征并提升了人脸识别的准确率；开展了远距离人脸图像的异质化匹配问题，在图像增强的同时实现了异质人脸图像高质量增强匹配。.本项目在远距离人脸图像增强、可识别特征恢复与提取、异质特征匹配等方面取得了相关的理论突破，实现了方法的创新。共发表代表性论文7篇，相关发明专利1项，培养硕士研究生6名，研究成果为远距离人脸识别的应用研究奠定一定的理论和技术基础。