Image enhancement method based on all 1*1 convolution neural network

A convolutional neural network and image enhancement technology, applied in the field of image processing, can solve problems such as high computational cost and limited wide application, achieve good accuracy, light weight CNN structure, and prevent overfitting.

Active Publication Date: 2019-01-22
HANGZHOU DIANZI UNIV
View PDF4 Cites 6 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

This results in high computational cost, limiting its wide application

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 enhancement method based on all 1*1 convolution neural network
  • Image enhancement method based on all 1*1 convolution neural network
  • Image enhancement method based on all 1*1 convolution neural network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0032] The present invention is further analyzed below in conjunction with specific examples.

[0033] The present invention can be used for various image enhancement tasks such as color constancy (color constancy, also called image color shift correction, white balance, etc.), image defogging, low-light image enhancement, and image noise level estimation. In the following, specific implementations of the present invention are mainly introduced for two embodiments of color constancy and image defogging.

[0034] 1 Image preprocessing

[0035] Randomly rearrange the local image blocks or the pixels in the entire image in the low-quality imaging image (PixelShuffle); the random rearrangement of the pixels is local rearrangement or global rearrangement. Pixels in the image, rearranging all their pixel position order randomly. Re-arranging (Pixel Shuffle) image blocks or pixels in the entire image will not change their statistical properties, but will destroy the spatial structu...

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 enhancement method based on an all 1*1 convolution neural network. A 1*1 convolution neural network is constructed, and pixels in a local image block in a low-qualityimage or an entire image are rearranged randomly (Pixel shuffle), and that rearranged image blocks or images are used as inputs. Then the latent variables estimated by the network are processed. Basedon the imaging model corresponding to the latent variable, the mathematical expression of the clear image estimated from the low-quality image and the latent variable is obtained, and the enhanced result is calculated. Compared with the convolution kernel method which is widely used in traditional convolution neural networks, it achieves the purpose of maintaining the model representation abilitywith fewer parameters and less computational load, so as to estimate the potential variables in image enhancement quickly and accurately.

Description

technical field [0001] The invention belongs to the field of image processing, and designs an image enhancement method based on a full 1*1 convolutional neural network. Background technique [0002] The image enhancement problem in image processing is mostly an image inverse problem (Inverse Problem) or an ill-posed problem (Ill-posed Problem). Researchers have proposed many methods, including: heuristic methods based on image priors, regular constraint modeling based Optimization methods, machine learning based methods. These methods can be interpreted as modeling certain statistical characteristics of natural images, which can effectively solve latent variables in ill-conditioned problems, and then enhance images. However, modeling the statistical properties in natural images for efficient image enhancement remains a challenging task due to the complexity, diversity, and high-dimensional distribution of pixels. [0003] A feasible solution is to assume statistical prior ...

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): G06T5/00G06T5/20
CPCG06T5/20G06T2207/20021G06T2207/20084G06T2207/20081G06T5/94G06T5/70
Inventor 张敬曹洋王洋查正军文成林
Owner HANGZHOU DIANZI UNIV
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

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

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap