Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

Efficient super-resolution reconstruction method based on deep learning

A super-resolution reconstruction and deep learning technology, applied in neural learning methods, image data processing, instruments, etc., can solve the problems of blurred high-frequency details, difficult super-resolution reconstruction, lack of high-frequency details, etc., to avoid The effect of vanishing gradients, easy training, and high reconstruction quality

Active Publication Date: 2020-06-19
GUANGXI NORMAL UNIV
View PDF9 Cites 3 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

Although this processing effect is fast, it will cause blurred high-frequency details in the picture, causing severe distortion
[0003] With the development of convolutional neural network (CNN), a large number of convolutional neural networks are used in the field of super-resolution reconstruction. CNN with complex structure can complete the reconstruction process through feature extraction and mapping of multiple modules, although it will form real detailed information. , but due to the complexity of the network, it is difficult to achieve real-time and fast super-resolution reconstruction
The CNN with a simple structure can complete the feature extraction and reconstruction process through a relatively rough module structure. Although it has high real-time performance, it will lack relatively complete high-frequency details.

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Efficient super-resolution reconstruction method based on deep learning
  • Efficient super-resolution reconstruction method based on deep learning
  • Efficient super-resolution reconstruction method based on deep learning

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0039] see Figure 1-6 , in a preferred embodiment of the present invention, an efficient super-resolution reconstruction method based on deep learning, comprising the following steps:

[0040] Step S1: Obtain the commonly used standard data sets for super-resolution. The standard data sets include "91-imags", "Set 5", "Set14", "BSDS100", "manga109", "DIV2K" and "Urban100". For the standard data The concentrated images are rotated, shifted, cut, and scaled to obtain an enhanced dataset.

[0041] Step S2: Read the image information, combine the low-resolution image and the high-resolution image into an information pair, and make a data set required by the network structure.

[0042] Step S3: In the network structure, the input image is subjected to two-branch processing; among them, the first branch uses the nearest neighbor interpolation method to enlarge the low-resolution image to the target size to form an interpolated and up-sampled image, and then proceed to step S8 for ...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention relates to the field of image processing and computer vision, relates to a convolutional neural network image super-resolution reconstruction related technology, in particular to an efficient super-resolution reconstruction method based on deep learning. The reconstruction method is based on a relatively simple and efficient FSRCNN network structure in a super-resolution reconstruction field structure. A nearest neighbor interpolation algorithm, residual connection, dense connection and transfer learning are added; a new network structure, namely a LuNet structure, is formed; theLuNet structure and an existing common super-resolution algorithm (Bicubic, SCN, SRCNN, FSRCNN) are subjected to an experimental test in performance, and the LuNet structure is obtained. The test structure shows that the LuNet structure formed by the method has higher reconstruction quality and more perfect performance under the condition that the model parameter quantity is kept unchanged.

Description

technical field [0001] The present invention relates to the fields of image processing and computer vision, and relates to technologies related to convolutional neural network image super-resolution reconstruction, specifically an efficient super-resolution reconstruction method based on deep learning. Background technique [0002] Image super-resolution reconstruction, also known as image enlargement, aims to enlarge the low-resolution image input by the user while improving the clarity of the image. After the image super-resolution reconstruction process, the low-resolution image becomes clearer. Traditional super-resolution image reconstruction methods use interpolation methods such as linear interpolation, bilinear interpolation, and nearest neighbor interpolation. This processing effect is fast, but it can blur high-frequency details in the picture and cause serious distortion. [0003] With the development of convolutional neural network (CNN), a large number of conv...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Applications(China)
IPC IPC(8): G06T3/40G06N3/08
CPCG06T3/4046G06T3/4076G06T3/4023G06N3/084
Inventor 黎海生鲁健恒黄华锋薛帆
Owner GUANGXI NORMAL UNIV
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products