基于深度学习的图像修复方法研究  

Research on image restoration methods based on deep leaming

作　　者：徐慧铭 程海[1] 姜焕德 黄春光[1] XU Huiming;CHENG Hai;JIANG Huande;HUANG Chunguang(College of Electronic and Engineering,Heilongjiang University,Harbin 150080,China)

机构地区：[1]黑龙江大学电子工程学院,哈尔滨150080

出　　处：《黑龙江大学自然科学学报》2022年第1期114-120,共7页Journal of Natural Science of Heilongjiang University

基　　金：国家自然科学基金资助项目(61471158);黑龙江省省属高等学校基本科研业务费研究项目(KJCX201904)。

摘　　要：基于深度学习的图像修复方法已经成为图像处理中的热门研究问题,有着广泛的应用前景。基于编码器-解码器,提出了一种基于注意力机制的编码器-解码器修复图像网络模型,优化改进了损失函数。与CVPR2017的General Inpainting图像修复算法进行定性和定量比较结果表明,所提出的修复网络模型能修复出真实且具有丰富细节的图像,弥补了传统图像修复后马赛克效果和分辨率低的弊端,使纹理和结构相结合,具有良好的实用性。Image restoration methods based on deep learning have become a popular research problem in image processing, and have a wide range of application prospects in the field of deep learning. An attention mechanism-based restoration image network model based on encoder-decoder is proposed to optimize and improve the loss function. The experimental analysis of qualitative and quantitative comparison with the General Inpainting image restoration algorithm of CVPR2017 shows that the restoration network model, which can restore realistic and detail-rich images, compensates for the disadvantages of mosaic effect and resolution reduction after traditional image restoration, and enables the combination of texture and structure with good practicality.

关 键 词：图像修复 深度学习 损失函数 编码器-解码器 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]