首页|面向视频侦查应用的退化人像GFP深度复原技术

面向视频侦查应用的退化人像GFP深度复原技术

扫码查看
原文链接
  • 国家科技期刊平台
  • NETL
  • NSTL
  • 万方数据
在视频侦查工作中时常遇到视频人像质量过低难以辨识的情况,而GFP等人像深度复原方法只适用于单张人像的复原.为此,本文提出一种基于GFP的监控视频人像复原技术,将GFP方法扩展至视频侦查应用,便于及时锁定犯罪嫌疑人,提高破案效率.首先对监控视频进行视频分帧、人像裁剪对齐、倾斜透视校正等预处理操作,然后使用GFP方法对预处理后得到的人像进行深度复原,最后经过逆处理将复原人像整合成高质量人像视频.在大量经过预处理后的模拟退化人像和无参考退化人像测试集上进行对比实验,结果表明,GFP方法在主观视觉效果和FID、PSFR、SSIM、NIQE等客观量化指标上均优于其他人像深度复原方法,相较于其他人像深度复原方法,能够更有效地复原复杂应用场景下的低质量退化人像,更适用于视频侦查应用;通过使用YTF视频人像数据集进行对比测试实验,结果显示,本文所提出的添加预处理与逆处理过程的基于GFP的监控视频人像复原技术,对于低质量视频人像有更加优秀的复原效果.
Deep Restoration Technology of Degraded Portraits Based on GFP for Video Investigation Application
In video investigation work,high-quality surveillance video portraits can provide a lot of effective information about criminal suspects,help investigators determine the direction of investigation,narrow the scope of investigation,and lock the criminal suspect.However,it is often encountered that the quality of video face images are too low to be recognized and facial image depth restoration methods such as GFP are only suitable for the restoration of a single facial image.To this end,this paper proposes a video facial image restoration technology based on the GFP,which extends the GFP method to the application of video investigation,facilitates the timely locking of criminal suspects in the video investigation work,and improves the efficiency of solving cases.Firstly,pre-processing operations such as video framing,facial image crop alignment,and oblique perspective correction are performed on the surveillance video.Then the GFP method is used to restore the depth of the pre-processed facial image.Finally,the restored face images through the reverse processing such as facial image registration,image restoration and video synthesis to generate high-quality facial image video.This paper verifies the effectiveness of portrait restoration technology based on GFP through simulation experiment,practical application and comparison experiment.Among them,simulation experiment and practical application were carried out on a large number of pre-processed real degraded face images and simulated degraded face images.The results show that the GFP is superior to other facial image depth restoration methods in terms of subjective visual effects and objective quantitative index such as FID,PSFR,SSIM,and NIQE.And by using YTF facial image video data set for comparison experiment,the experimental results show that the method of adding pre-processing and reverse processing is superior to the traditional GFP method.The video portrait restoration technology based on GFP with pre-processing and inverse processing proposed in this paper has better restoration effect for low-quality video portraits.

video investigationsurveillance videofacial image restorationvideo framingoblique perspective correctionGFP

倪新龙、孙鹏、郎宇博、赵理夫、田天泽、周纯冰

展开 >

中国刑事警察学院,沈阳 110854

司法部司法鉴定重点实验室,上海 200063

锦州市公安局古塔分局,辽宁锦州 121000

视频侦查 监控视频 人像复原 视频分帧 倾斜透视校正 GFP

公安部科技计划司法部司法鉴定重点实验室开放基金中国刑事警察学院研究生创新能力提升项目国家级大学生创新创业训练计划

2021YY32KF2023172022YCZD06202110175015

2024

10.16467/j.1008-3650.2023.0049

刑事技术
公安部物证鉴定中心

刑事技术

CSTPCD
影响因子：0.315
ISSN：1008-3650
年,卷(期)：2024.49(2)
  • 15