Eureka AIR delivers breakthrough ideas for toughest innovation challenges, trusted by R&D personnel around the world.

Attention-fused single image rain removal method

A single image, attention technology, applied in the field of image processing, can solve the problems of inconvenient processing, consider the scene as a rain line, long running time, etc., to avoid the disappearance of the gradient and solve the problem of high-dimensional channel feature fusion.

Pending Publication Date: 2022-08-09
SOUTH CHINA AGRI UNIV
View PDF0 Cites 2 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

Luo et al. proposed a single image rain removal algorithm based on dictionary learning. By learning strong mutual exclusion learning dictionary, using highly discriminable sparse coding to accurately separate the background layer and rain layer, this method can effectively separate the background and rain lines , but when the scene in the image is similar to the structure of the rain line, this method will regard the scene as the rain line, resulting in excessive smearing of the background
Li et al. used a Gaussian mixture model to separate the background layer and rain layer to adapt to multi-directional and different-scale rain patterns. This method requires the user to manually select a suitable area in the image to provide prior information to construct a Gaussian mixture model, which is inconvenient to handle. and long running time

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-fused single image rain removal method
  • Attention-fused single image rain removal method
  • Attention-fused single image rain removal method

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0127] The specific embodiments of the present invention will be further described below in conjunction with the accompanying drawings:

[0128] like figure 1 As shown, this embodiment provides a method for removing rain from a single image based on fusion attention, including the following steps:

[0129] S1), build a neural network architecture including 3D attention, Transformer and encoder-decoder architecture, such as figure 2 As shown, the neural network architecture includes an encoder, an intermediate layer and a decoder;

[0130] Among them, the encoder and decoder are composed of convolution layer and global residual dense block, use convolution for downsampling operation, use deconvolution as upsampling operation; use corresponding feature map matrix to add fusion encoder and Decoder path features.

[0131] Each said global residual block consists of two 3D attention local residual dense blocks and residual connections, each 3D attention local residual dense blo...

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 relates to an attention-fused single image rain removal method, which comprises the following steps of: firstly, carrying out batch normalization processing on an input image, and then inputting the input image into an encoder network in which a three-dimensional attention mechanism and a residual dense block structure are combined to obtain high-dimensional features; the global feature relevance is calculated by using a Transform mechanism; reducing the feature matrix to the size of an original input image step by step by using a decoder constructed by a three-dimensional attention residual dense block to obtain an output image; in order to solve the problem that image structure information and detail high-frequency information are erased due to an algorithm in the rain removal process, multi-scale structural loss and a common rain removal loss function are combined to participate in rain removal network training. According to the invention, a rain-removed image with higher quality can be obtained by carrying out rain removal on a single rain image. According to the method, the three-dimensional attention mechanism, the Transform and the coder-decoder architecture are combined, so that the rain removal performance of the network can be better improved.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a method for removing rain from a single image fused with attention. Background technique [0002] In life and social environment application scenarios, there are mainly two types of image data: image data and video data. Video can be regarded as a sequence of images stacked continuously before and after one frame, and there is continuous data between adjacent frames of the original video. The early rain removal algorithm mainly processed video data at the initial stage, and the video rain removal algorithm can use the rich inter-frame information to help generate the rain removal image. The processing object of single image deraining algorithm is single image data. Unlike video data, single image loses a lot of inter-frame information between consecutive frames, so the research of single image deraining algorithm is more complicated. The research on the single image r...

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): G06T5/00G06T7/40G06N3/04G06N3/08G06V10/80
CPCG06T7/40G06N3/08G06T2207/20081G06T2207/20084G06N3/048G06N3/045G06F18/253G06T5/73Y02A90/10
Inventor 王美华柯凡晖廖磊
Owner SOUTH CHINA AGRI 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
Eureka Blog
Learn More
PatSnap group products