Attention mechanism-based image blind deblurring method and system

A technology of blind deblurring and attention, applied in the field of image processing, to achieve the effect of improving feature extraction and expression ability, edge improvement, and optimizing training process

Active Publication Date: 2020-09-25
INNOVATION ACAD FOR MICROSATELLITES OF CAS +1
View PDF4 Cites 28 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0010] The purpose of the present invention is to provide a blind image deblurring method and system based on the attention mechanism, so as to solve the p

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
  • Attention mechanism-based image blind deblurring method and system
  • Attention mechanism-based image blind deblurring method and system
  • Attention mechanism-based image blind deblurring method and system

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0080] This embodiment provides an attention mechanism-based blind image deblurring method based on an asymmetric encoding-decoding network structure. The structure is simple and compact, and has been widely used in various computer vision tasks. Recently, in the field of image restoration, many learning-based end-to-end methods have been developed around this. In the video deblurring task, a codec network combined with skip connections is used. In the field of single image dynamic scene deblurring, the method based on residual blocks is proposed. The codec network uses a multi-scale cyclic structure to gradually estimate clear images from coarse to fine. It is currently a network with a small number of parameters and better performance in the deblurring method based on deep learning.

[0081] In the encoding and decoding network, the tasks of the encoding side and the decoding side are different, and the encoding side starts from the blurred image I Blur Downsampling extract...

Embodiment 2

[0139] This embodiment provides a blind image deblurring system based on an attention mechanism. The blind image deblurring system based on an attention mechanism includes: a multi-scale attention network that directly restores a clear image in an end-to-end manner; The multi-scale attention network adopts an asymmetric encoding and decoding structure, and the encoding side of the encoding and decoding structure adopts a residual dense network block to complete the feature extraction and expression of the input image by the multi-scale attention network; the encoding and decoding The decoding side of the structure is provided with a plurality of attention modules, and the attention module outputs a preliminary restored image, and the preliminary restored image forms an image pyramid multi-scale structure; the attention module also outputs an attention feature map, and the attention The force feature map models the relationship between distant regions from a global perspective 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 provides an attention mechanism-based image blind deblurring method and system, and the method comprises the steps that a multi-scale attention network directly recovers a clear image inan end-to-end mode, wherein the multi-scale attention network adopts an asymmetric encoding and decoding structure, and the encoding side of the encoding and decoding structure adopts a residual dense network block to complete feature extraction and expression of an input image by the multi-scale attention network; a plurality of attention modules are arranged on the decoding side of the encodingand decoding structure, the attention modules output preliminary restored images, and the preliminary restored images form a pyramid-type multi-scale structure of the image; the attention module further outputs an attention feature map, and the attention feature map models the relationship between the long-distance areas from the global perspective to process the blurred image; dark channel priorloss and multi-scale content loss form a loss function, and the loss function is used for reversely optimizing the network and is not self-optimized.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to an attention mechanism-based blind image deblurring method and system. Background technique [0002] Dynamic scene blur is a common phenomenon caused by factors such as the shaking of the shooting device, the movement of the target object at different depths of field, and the camera being out of focus during the image acquisition process. Blurred images not only affect the visual effect, but also are not conducive to subsequent computer vision tasks such as target detection and semantic segmentation. Therefore, deblurring has always been a basic but very important problem in the field of image processing. [0003] Usually blurred images are mathematically modeled as: [0004] [0005] in, Represents a two-dimensional spatial convolution operation. Blurred Image I Blur is defined by the clear image I Sharp Convolved with the blur kernel K, and then superimposed n...

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): G06T5/00G06T5/50G06N3/04G06N3/08
CPCG06T5/003G06T5/50G06T2207/10016G06T2207/20081G06T2207/20084G06N3/084G06N3/045Y02T10/40
Inventor 林晨王子健尹增山
Owner INNOVATION ACAD FOR MICROSATELLITES OF CAS
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