摘要
图像修复在研究中仍然具有挑战性,现阶段的图像修复算法往往存在修复边缘模糊,修复区域与其周边区域存在明显不一致的问题。本文提出一种有效的基于MSE损失函数和感知损失函数相结合的图像修复算法。该网络模型由修复网络、全局判别网络和局部判别网络组成,经过对网络的多次迭代和参数更新后,最终能够完成人脸修复。经在celeb A数据集上进行测试,通过主观和客观的评价标准证明可以获得良好的修复结果。
Image restoration is still challenging in today’s research.At present,image restoration algorithms often have the problem of repairing edge blur,and the repaired area is obviously inconsistent with its surrounding area.This paper proposes an effective MSE loss function and content loss.A combination of functions for image restoration algorithms.The network model consists of a completion network,a global discriminator network,and a local discriminator network.After multiple iterations of the network and parameter updates,the face repair can be completed.The experiment was tested on the celeb A dataset,and good completion results were obtained by subjective and objective evaluation criteria.
作者
王可新
王力
WANG Kexin;WANG Li(Department of Big Data and Information Engineering,Gui Zhou University,Gui Yang 550000,China)
出处
《智能计算机与应用》
2020年第4期9-12,共4页
Intelligent Computer and Applications
关键词
人脸修复
感知损失函数
局部判别网络
the face of a repair
content loss
a local discriminator network
