基于GAN网络的去除人脸运动模糊算法

Facial Motion Blur Removal Algorithm Based on GAN Network

随着社会科技水平的日新月异,拍照设备以及视频监控已经融入到了人们生活中的方方面面,日常拍照中的相机抖动和人物运动会导致图像产生运动模糊,影响图像的拍摄质量;视频监控中有时需要对目标人物进行定位、识别、跟踪,但是很多情况下,由于相机抖动或者人物运动,导致图像产生运动模糊,增加了人脸识别的难度,有时甚至根本无法分辨。为了解决人脸运动模糊的问题,本文提出一种基于GAN网络的网络模型算法,在GAN网络的基础上,通过上采样卷积网络和残差网络加深网络的深度,使用跳跃链接对网络进行优化,加强网络前后特征间的联系,最后对损失函数权重进行训练调整,得到了最终的端到端网络。实验结果表明,这种网络模型能够很好地去除人脸运动模糊,恢复人脸轮廓和细节。

With the rapid development of social science and technology,camera equipment and video monitoring have been integrated into all aspectsof people’s lives.Camera jitter and character movement in daily photographing will cause motion blur of images,which will affect the quali-ty of images.Sometimes,it is necessary to locate,identify and track the target characters in video monitoring,but in many cases,due tocamera shaking Motion or human motion leads to motion blur of image,which increases the difficulty of face recognition,and sometimeseven cannot be distinguished at all.In order to solve the problem of facial motion blur,a network model algorithm based on GAN network isproposed in this paper.On the basis of GAN network,the depth of the network is deepened by up sampling convolution network and residu-al network,and the network is optimized by jump link to strengthen the connection between the features before and after the network.Final-ly,the weight of loss function is trained and adjusted to obtain the final end-to-end End network.The experimental results show that thisnetwork model can remove facial motion blur and restore face contour and detail.