Open Access

Research on Image Super-resolution Reconstruction Based on Deep Learning

Jingyu Jiang
Jingyu Jiang
, 
Li Zhao
Li Zhao
 and 
Yan Jiao
Yan Jiao
   | May 28, 2023
International Journal of Advanced Network, Monitoring and Controls's Cover Image
About this article
Previous Article
Next Article
Cite
Published Online: May 28, 2023
Page range: 1 - 21
DOI: https://doi.org/10.2478/ijanmc-2022-0001
Keywords
© 2022 Jingyu Jiang et al., published by Sciendo
This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Figure 1.

Super-resolution method based on deep learning
Super-resolution method based on deep learning

Figure 2.

SRCNN network model
SRCNN network model

Figure 3.

VDSR network structure
VDSR network structure

Figure 4.

Schematic diagram of the basic structure of SRGAN
Schematic diagram of the basic structure of SRGAN

图 1

基于深度学习的超分辨率方法
基于深度学习的超分辨率方法

图 2

SRCNN 网络模型
SRCNN 网络模型

图 3

VDSR 网络结构
VDSR 网络结构

图 4

SRGAN 基本结构示意图
SRGAN 基本结构示意图

Introduction to commonly used data sets

Data set name Number of pictures Image Format Brief description of the data set
Set5[24] 5 PNG The pictures included are from babies, birds, butterflies, children's heads, and a lady.
Set14[25] 14 PNG The included pictures come from characters, animals, insects, flowers, vegetables, comedians, etc.
Berkeley segmentation[26] 500 JPG Referred to as BSD500. The pictures included are from animals, buildings, food, people and plants, etc. One of the 100 or 300 pictures is often used, which is called the BSD100 BSD300 data set.
Urban100[27] 100 PNG The pictures included are mainly different types of urban buildings.
Manga109[28] 109 PNG The pictures included are all from Japanese manga.
T91[29] 91 PNG The included pictures come from vehicles, flowers, fruits and human faces. Often used as a training set.
General-100[30] 100 BMP The pictures included are from animals, daily necessities, food, plants, people, etc. It is also often used as a training set.
DIV2K[23] 1000 PNG A dataset of high-definition pictures, with pictures from natural environments, landscapes, handicrafts and people, etc. Among them, 800 pictures are often used as training sets.
Flickr2K[15] 2650 PNG The included pictures come from people, animals, landscapes, etc., and are often used as larger training sets.

常用数据集介绍

数据集名称 图片数量 图片格式 数据集简单描述
Set5[24] 5 PNG 包含的图片分别来自婴儿、鸟、蝴蝶、小孩的头部和一个女士。
Set14[25] 14 PNG 包含的图片来自人物、动物、昆虫、花、蔬菜和喜剧演员等。
Berkeley segmentation[26] 500 JPG 简称 BSD500。包含的图片来自动物、建筑、食物、人和 植物等。经常使用其中的 100 张或 300 张图片，称为 BSD100 BSD300 数据集。
Urban100[27] 100 PNG 包含的图片主要是不同类型的城市建筑物。
Manga109[28] 109 PNG 包含的图片均来自日本漫画。
T91[29] 91 PNG 包含的图片来自车辆、花、水果和人脸等。常被用作训练集。
General-100[30] 100 BMP 包含的图片来自动物、日常用品、食物、植物和人物等。 也常用作训练集。
DIV2K[23] 1000 PNG 高清图片数据集，图片来自自然环境、风景、手工艺品和 人物等。其中的 800 张图片经常作为训练集使用。
Flickr2K[15] 2650 PNG 包含的图片来自人物、动物和风景等，常作为较大规模训 练集使用。
eISSN:
2470-8038
Language:
English
Publication timeframe:
4 times per year
Journal Subjects:
Computer Sciences, other
Journal RSS Feed