Image deblurring method based on ADMM neural network

A neural network and deblurring technology, applied in biological neural network models, neural learning methods, neural architectures, etc., can solve problems such as large effect gaps, small effect gaps, image deblurring, etc., and achieve the effect of reducing workload

Active Publication Date: 2020-07-03
大连厚仁科技有限公司
View PDF7 Cites 2 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

The advantage of the end-to-end defuzzification algorithm based on the neural network is that it does not need to establish a more complex mathematical solution model, but only needs to follow the principles of deep learning. The adjustment of relevant parameters in the training process mainly relies on the back propagation mechanism and the optimizer. It does not require human intervention. The disadvantage is that many neural networks with better effects are obtained through a large number of experiments, and their essence cannot be explained by relevant theories. Secondly, the training of neural networks is mainly to learn the potential laws of the data set. If the data set If it does not contain a certain necessary underlying law, then the effect will vary greatly in the application of different scenarios, that is, there is an over-fitting problem
The advantage of the image deblurring algorithm based on optimization theory is that it can be used without training, which can avoid the over-fitting problem of the neural network, so the effect is not much different in the application of different scenarios. The disadvantage is that it needs to establish a corresponding solution model. Not only does it require relevant personnel to have a lot of prior knowledge, but the key parameters in the algorithm usually need to be obtained through a lot of human experiments, otherwise it will greatly affect the actual effect
So far, there is no relevant report on the fusion of the advantages of neural network-based algorithms and optimization theory-based algorithms for image deblurring

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
  • Image deblurring method based on ADMM neural network
  • Image deblurring method based on ADMM neural network
  • Image deblurring method based on ADMM neural network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0040] The image deblurring method based on ADMM neural network of the present invention, carry out according to the following steps:

[0041] 010 preprocessing stage

[0042] Step C011: Configure the software environment of the PC, including Python3.6, Tensorflow2.0, CUDA10.0, etc.;

[0043] Step C012: using GOPRO training set;

[0044] Step C013: Initialize the camera;

[0045] Step C014: configure the local area network where the PC and camera are located;

[0046] 020 training stage

[0047] Step C021: Construct a mathematical model and use ADMM to solve the split term as follows:

[0048]

[0049] where x (i) Refactor layer Restore for stage i (i) output of x (i-1) Refactor layer Restore for stage i-1 (i -1) The output of , y is the original blurred image, z (i) Denoise layer Denoise for the i-th stage (i) the output of z (i-1) Denoise layer Denoise for stage i-1 (i-1) output of β (i) Update the layer Multipler for the i-th stage multiplier (i) , β (i-1...

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 an image deblurring method based on an ADMM neural network. Modeling is carried out for the problem of image deblurring. Split item solving is carried out on the problem by utilizing an ADMM algorithm; the related neural network is constructed according to the three solved sub-problems, the ADMM algorithm parameters and the related regular terms which better accord with thereality are obtained through training of the constructed neural network, the workload of participation in regular term selection can be reduced, and the actual effect and the operation efficiency canbe further improved. According to the method, the excellent theoretical basis of a traditional optimization algorithm is reserved, the trainability of the neural network is utilized, and deblurring processing can be well carried out on the moving image.

Description

technical field [0001] The method of the invention relates to an image deblurring method aimed at image features, in particular to an image deblurring method based on ADMM neural network. Background technique [0002] In recent years, due to the vigorous development of image processing and deep learning, a large number of image deblurring algorithms have emerged. The existing popular image deblurring algorithms are mainly divided into end-to-end image deblurring algorithms based on neural networks and image deblurring algorithms based on optimization theory. The advantage of the end-to-end defuzzification algorithm based on the neural network is that it does not need to establish a more complex mathematical solution model, but only needs to follow the principles of deep learning. The adjustment of relevant parameters in the training process mainly relies on the back propagation mechanism and the optimizer. It does not require human intervention. The disadvantage is that man...

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): G06T5/00G06N3/04G06N3/08
CPCG06N3/08G06T2207/10004G06N3/045G06T5/73Y02T10/40
Inventor 傅博傅世林吴越楚董宇涵
Owner 大连厚仁科技有限公司
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products