摘要
为了实现短波红外-可见光人脸图像的跨模态识别,提出了基于内容特征提取的短波红外-可见光人脸识别框架。首先建立了短波红外-可见光人脸图像数据集,对图像翻译框架DRIT进行改进,更为准确地获取图像的内容特征并得到更好的翻译结果;接着,采用改进的图像翻译框架中的内容特征提取器进行内容特征提取,以克服模态差异对识别的干扰,然后设计识别网络,基于内容特征完成跨模态的短波红外-可见光人脸识别任务。在自建短波红外-可见光人脸图像数据集上对改进的图像翻译框架和跨模态人脸识别框架进行测试,实验结果表明,改进的DRIT图像翻译框架中的内容特征提取器可以更准确地进行内容特征提取,应用于识别任务时识别准确率提升了12.89%,整体识别框架对短波红外人脸识别准确率达到88.86%。本文提出的基于内容特征提取的识别方案有效克服了模态差异,获得了较好的短波红外-可见光人脸识别结果。
To recognize shortwave-infrared(SWIR)face images according to enrolled visible-light(VIS)face images,a SWIR-VIS face recognition framework based on content feature extraction is proposed.Initially,a SWIR-VIS face image dataset was established.DRIT–an image translation frame–is modified to extract content features more accurately,and consequently obtains better translation results.Then,the content feature extractors in the improved DRIT framework overcome the interference of the modal differ⁃ence on the recognition.The network used to recognize SWIR faces based on content features was adopted to complete the cross-modal SWIR-VIS face recognition task.The proposed network is evaluated on a self-built SWIR-VIS face image dataset,and compared with the existing widely used methods.Experi⁃mental results indicate that the improved DRIT could extract content features more accurately,and conse⁃quently the recognition accuracy with content extractors from the improved DRIT model is 12.89%higher than that with the original DRIT content extractors.The recognition accuracy of the proposed framework in the task of SWIR-VIS recognition was 88.86%.The proposed framework can effectively overcome the modality gap and improves the recognition accuracy.
作者
胡麟苗
张湧
楼晨风
HU Lin-miao;ZHANG Yong;LOU Chen-feng(Shanghai Institute of Technical Physics,Chinese Academy of Science,Shanghai 200083,China;Key Laboratory of Infrared System Detection and Imaging Technology,Chinese Academy of Science,Shanghai 200083,China;University of Chinese Academy of Science,Beijing 100049,China)
出处
《光学精密工程》
EI
CAS
CSCD
北大核心
2021年第1期160-171,共12页
Optics and Precision Engineering
基金
国家十三五国防预研资助项目(No.JZX2016-0404/Y72-2)
上海市现场物证重点实验室基金资助项目(No.2017XCWZK08)。
关键词
图像翻译
短波红外图像
人脸识别
内容特征
image translation
short-wave infrared images
face recognition
content feature
