项目名称: 面向识别的低质量人脸特征超分辨率重建技术研究
项目编号: No.61471013
项目类型: 面上项目
立项/批准年度: 2015
项目学科: 无线电电子学、电信技术
项目作者: 李晓光
作者单位: 北京工业大学
项目金额: 82万元
中文摘要: 受采集距离、环境光照、压缩失真等各种因素的影响,视频监控等应用中的人脸图像往往是模糊不清、低分辨率的低质量图像,这类非约束条件下的人脸识别难以获得理想的识别性能。本项目将重点开展低质量人脸识别技术的研究。首先从机理上对压缩失真、低分辨率、噪声等因素所导致的特征降质过程进行深入的探讨和研究,建立人脸特征的统一降质模型。在此基础上,利用机器学习的方法,建立不同分辨率人脸图像的低质特征到高质特征的映射模型,实现低质人脸特征到高质人脸特征的超分辨率重建,有效解决不同分辨率人脸特征维度不匹配的问题。通过将人脸可鉴别先验信息引入到超分辨率重建中,提高特征的区分能力,从而提高低质量人脸图像的机器识别率。本项目的研究成果可以用于视频监控、刑侦、身份识别等社会安全和信息安全领域,为人脸图像的识别提供新的技术手段和方法,进一步扩大人脸识别的应用范围。
中文关键词: 智能信息处理;人脸识别;超分辨率重建;可鉴别特征
英文摘要: Due to the influence of the capturing distance,illumination and compression,facial images in surveillance video usually suffer from compressed distortion, low resolution and blur. In such unconstraint situation, the perfermance of face recognition is unsatisfactory.In this proposal, we will investigate the techniques of face recognition for low quality images. Therefore,we will investigate the principle of facial features degradation due to low-bit compression,low resolution and noises.Then, we will build a degradation model for discriminative features. Finally, a framework of learning based high quality reconstruction for discriminative features will be explored. The facial discriminative priors will be introduced to enhance the discriminative information of the reconstructed feature and improve the recognition rate. The research findings of this proposal can be applied in the fields of social and information security, such as video surveillance, criminal investigation and identity management. They will provide new technical means and methods for the face recognition and further expand the scope of application of face recognition.
英文关键词: Intelligent Information Processing;Face Recognition;Super Resolution;Discriminative Feature