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

Image denoising method based on deep convolutional network

A deep convolution, image technology, applied in image enhancement, image analysis, image data processing and other directions, can solve problems such as slow convergence speed, inability to flexibly handle noise, etc., to achieve improved image denoising performance, fast operation speed, and improved performance Effect

Active Publication Date: 2021-11-09
NAT UNIV OF DEFENSE TECH
View PDF5 Cites 4 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

Since DnCNN is a simple fully convolutional network, its training optimization process faces the problem of slow convergence caused by shallow gradients approaching zero.
Secondly, DnCNN needs to train a model weight for each noise level, and cannot flexibly handle different levels of noise

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 denoising method based on deep convolutional network
  • Image denoising method based on deep convolutional network
  • Image denoising method based on deep convolutional network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0052] 1. Formal modeling of the problem

[0053] Denote the column-first expansion vector of the Bayer image as b∈R n , the red, green, and blue channels of the corresponding linear RGB image are expanded in the same way and the three vectors are r, g, b∈R n . And the linear RGB image vector formed by the connection of three channel vectors is The image degradation model is

[0054] b=H(x)=Ax+η (1)

[0055] where A∈R n×3n is the color filter matrix, which simulates the process of color filter formation mosaic through matrix multiplication, and samples the linear RGB image x into a Bayer image b; η∈R n is the noise vector.

[0056] When solving the problem of joint demosaicing and denoising, the present invention directly learns from the data set {(x i ,b i )|i=1,...,m} Construct a mapping F(b; θ) composed of parameter models, such that

[0057]

[0058] where θ is the set of model parameters. When we have texture-rich and noise-free image x i , data set {(x i ...

Embodiment 2

[0093] Embodiment 1 studies the method of jointly demosaicing and denoising Raw images by using a deep convolutional network. This embodiment extends this method to the denoising problem of RGB images. This method is originally suitable for non-blind denoising conditions, which require an input noise level. In view of this problem, this embodiment improves the model so that it can solve the problem of blind denoising, so that only the noise image needs to be input during denoising, and the noise level does not need to be input.

[0094] RGB images are divided into linear RGB images and sRGB images. The difference between them is that the color and brightness processing processes are different, resulting in different distribution of noise. The problem modeling and method description in this embodiment do not involve specific noise distribution, and the quantitative experiment of Gaussian noise is mainly used in the experiment, supplemented by real sRGB, so there is no need to ...

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 denoising method based on a deep convolutional network. The method comprises the following steps: establishing an image degradation model, and sampling an image into a noise image; forming a training image pair by using the noise map and the original image block; inputting training data into a deep convolutional network for learning, enabling the deep convolutional network to use a convolutional layer and a non-linear activation layer to expand channels and then pass through a plurality of residual blocks, and reducing the number of the channels to the number of output channels through the convolutional layer and the non-linear activation layer; training by using a training algorithm, and learning a mapping function from a noise image to a noise-free image from the training image pair through an optimization method; and optimizing by using a loss function until the parameters of the deep convolutional network converge. The image priori of the data set is mined through training, the recognition capability of the network on the noise intensity is enhanced by receiving the noise level information for multiple times, and the performance of joint demosaicing and denoising is greatly improved.

Description

technical field [0001] The invention belongs to the technical field of digital image processing, in particular to an image denoising method based on a deep convolutional network. Background technique [0002] The limitation of non-learning denoising methods is that they can only mine priors such as image similarity in the current image, while learning methods can automatically mine image priors from a large amount of image data, so that better denoising performance can be achieved. The existing deep learning denoising methods do not consider the input noise level in the model, and multiple denoising models need to be trained for different noise levels; or the noise level information may not be fully recognized in the model if the noise level is only input once. [0003] Image denoising is a fundamental problem in low-level vision. Given a noisy image y, the goal of denoising is to recover a clear image x from an imaging model y = x + n, where n is the noise. In this imagin...

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): G06T5/00G06T3/40G06N3/04G06N3/08
CPCG06T3/4015G06N3/08G06T2207/20081G06T2207/20084G06N3/045G06T5/70Y02T10/40
Inventor 谭瀚霖肖华欣彭杨刘煜张茂军
Owner NAT UNIV OF DEFENSE TECH
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

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic, Popular Technical Reports.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com