面向档案修复的低质图像修复与重建算法设计  

Algorithm Design of Low-quality Image Restoration and Reconstructionfor File Restoration

作　　者：董碧娜[1] DONG Bina(School of Marxism,Shaanxi Institute of Technology,Xi’an 710300,China)

机构地区：[1]陕西国防工业职业技术学院,马克思主义学院,陕西西安710300

出　　处：《微型电脑应用》2024年第8期59-62,共4页Microcomputer Applications

基　　金：陕西省教育厅科研计划项目(21JK0044);陕西省教育科学“十四五”规划2021年度一般课题(SGH21Y0515)。

摘　　要：传统的低质图像修复算法难以对图像的细节进行学习,由此导致修复后的图像边缘较为模糊,且细节信息也存在一定的不足。针对这一问题,对低质图像进行分析,并在改进深度卷积生成对抗网络的基础上提出一种图像修复与色彩重建算法。该算法使用粗细尺度网络生成器结构代替原模型结构,其中,粗尺度网络可学习全局特征,而细尺度网络则能学习图像边缘细节信息。使用多尺度注意力机制自动填充颜色,进而完成图像的重建。实验测试表明,所提算法处理后,图像的PSNR指标、SSIM指标以及算法运行时间均优于对比算法,且重建后的图像细节丰富,故所提算法具有一定的工程实用价值。The traditional low-quality image restoration algorithm is difficult to learn the details of the image,which leads to the blurred edge of the repaired image,and there are some deficiencies in the detail information.To solve this problem,this paper analyzes the low-quality image,and proposes an image restoration and color reconstruction algorithm based on the improved deep convolutional generation adversarial network.The algorithm uses the coarse and fine scale network generator structure to replace the original model structure,in which the coarse-scale network can learn the global features,while the fine-scale network can learn the image edge details.The multi-scale attention mechanism is used to automatically fill the color,and then complete the image reconstruction.In the experimental test,the PSNR index,SSIM index and the running time of the proposed algorithm are better than the comparison algorithm,and the reconstructed image is rich in detail,so the algorithm has certain engineering practical value.

关 键 词：低质图像修复 深度卷积生成对抗网络 多尺度网络 注意力机制 图像重建 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]