Unlock instant, AI-driven research and patent intelligence for your innovation.

Super-resolution image reconstruction method based on convolutional neural network

A convolutional neural network and super-resolution technology, applied in neural learning methods, biological neural network models, neural architectures, etc., which can solve the problems of slow running speed of convolutional neural networks, easy disappearance of training network gradients, and unsatisfactory image quality. problem, to improve the effect of image reconstruction, the network structure is clear and easy to understand, and the effect of improving gradient disappearance

Inactive Publication Date: 2021-05-25
HARBIN UNIV OF SCI & TECH
View PDF4 Cites 5 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0005] In view of the above problems, the present invention proposes a super-resolution image reconstruction method based on a convolutional neural network, in order to solve the problems of the existing super-resolution convolutional neural network with slow running speed, unsatisfactory image quality and low image resolution. Low, training network gradients are easy to disappear, etc.

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
  • Super-resolution image reconstruction method based on convolutional neural network
  • Super-resolution image reconstruction method based on convolutional neural network
  • Super-resolution image reconstruction method based on convolutional neural network

Examples

Experimental program
Comparison scheme
Effect test

specific Embodiment 1

[0079] The training process of the convolutional neural network requires a large number of matrix operations, and this operation implements super-resolution training on the CPU. The embodiment uses the training set as the DIV2K data set in https: / / github.com / xinntao / BasicSR, and the data set A training set containing 800 HR (high resolution HR) images and a test set of 100 HR images. The images in the dataset in DIV2K are all color images, which is a high-quality (2K resolution) image dataset for image restoration tasks.

[0080] The evaluation standard of this embodiment is evaluated by calculating the peak signal-to-noise ratio PSNR (Peak Signal-to-Noise Ratio, PSNR) index of the original image and the generated image. Peak Signal-to-Noise Ratio (PSNR) is usually used to measure reconstructed images of lossy transformations (such as image compression, image restoration), and is a quantitative quality method for evaluating and comparing models, indicating how close the recons...

specific Embodiment 2

[0095] In this embodiment, on the basis of the network model in the first embodiment, the part of the discriminant network in the generative confrontation network is added. The training is also performed on the CPU, and the open source data set of DIV2K is used for model training and testing. Use the PIL library in Python for image processing. The PIL library allows the use of different convolution kernels for filtering, color space conversion, image size conversion, image rotation and various affine transformations. First, image processing is performed on the trained high-resolution image to make it a low-resolution image to be processed, and then the processed low-resolution image is obtained through the improved super-resolution convolutional neural network method of the present invention to obtain a reconstructed high-resolution image. rate image. The training set images are randomly trained to improve the generalization ability of the network.

[0096](1) Experimental t...

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 discloses a super-resolution image reconstruction method based on a convolutional neural network, belongs to the technical field of image processing, and is used for solving the problems that an existing super-resolution convolutional neural network is low in operation speed, the quality of an obtained image is not ideal, the image resolution is relatively low, and the gradient of a training network is easy to disappear. According to the technical key points, the method comprises: improving an existing super-resolution convolutional neural network model: enabling image zooming to occur in the rear section of the model, and carrying out post-up-sampling operation; post up-sampling is an up-sampling-sub-pixel method based on learning; deepening the number of network layers, and adding a residual network into the network; and further taking the improved super-resolution convolutional neural network model as a generative network in a generative adversarial network, and integrating the generative network with the adversarial network, so that the image reconstruction efficiency is further improved. The method can be widely applied to the field of super-resolution image reconstruction research.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a super-resolution image reconstruction method based on a convolutional neural network. Background technique [0002] Neural network has become an important technical means in the field of image recognition and image classification, and it has a growing trend. Therefore, the application of super-resolution convolutional neural network technology to reconstruct blurred pictures into high-definition image technology research is of great importance to the computer. Both visual development and the development of artificial intelligence have important value and significance. When using computer technology to operate on objects such as pictures, videos, and voices, due to the influence of computer CPU, GPU and other hardware conditions, when we want to obtain high-quality target content more efficiently and quickly, we can only retreat Next, modify and integrate the network s...

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
IPC IPC(8): G06T3/40G06K9/46G06K9/62G06N3/04G06N3/08
CPCG06T3/4053G06N3/08G06V10/44G06N3/048G06N3/045G06F18/2132G06F18/241
Inventor 李鹏飞李丽丽
Owner HARBIN UNIV OF SCI & TECH