基于感知去模糊的高分辨率破损图像修复方法  

High Resolution Damaged Inpainting Method Based on Perceptual Deblurring

作　　者：濮毅 PU Yi(Department of Electronic Information,Maanshan Technical College,Maanshan 243031,China)

机构地区：[1]马鞍山职业技术学院电子信息系,安徽马鞍山243031

出　　处：《常州工学院学报》2024年第2期8-13,共6页Journal of Changzhou Institute of Technology

基　　金：安徽省教育厅高等学校省级质量工程项目(2022jyxm1577,2022cjrh046,2021jxtd286)。

摘　　要：为提高修复图像的视觉效果,提出基于感知去模糊的高分辨率破损图像修复方法。将高分辨率图像样本输入卷积自编码生成对抗网络中,利用编码器降维处理并输出其低维特征矩阵后,由解码器对其升维并解码,最终采用生成器完成映射学习。通过不断搜寻,获得与输入高分辨率图像L_(1)距离差异最小的生成图像,由判别网络对其作真假判断,实现高分辨率破损图像的粗修复后,再将其输入感知去模糊网络模型中,增强图像细节信息后实现高分辨率图像修复。实验结果表明:该方法修复后的高分辨率图像细节丰富、颜色自然、视觉效果突出。To improve the visual effect of repaired images,a high-resolution damaged image restoration method based on perceptual deblurring is proposed.Using high-resolution image samples as the input for convolutional self encoding to generate adversarial networks,the encoder is used to reduce the dimensionality and output its low dimensional feature matrix.After that,the decoder increases the dimensionality and decodes it.Finally,the generator is used to complete mapping learning.By continuously searching for the generated image with the smallest distance difference from the input high-resolution image L_(1),the discriminant network determines whether it is true or false.After the coarse repair of the high-resolution damaged image,it is used as input to the perceptual deblurring network model to enhance image detail information and achieve high-resolution image restoration.The experimental results show that the high-resolution images repaired by this method have rich details,natural colors,and outstanding visual effects.

关 键 词：感知去模糊 高分辨率破损图像 特征矩阵空间 粗修复 L_(1)距离差异 真假判断 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]