CN113034381A - Single image denoising method and device based on cavitated kernel prediction network - Google Patents
Single image denoising method and device based on cavitated kernel prediction network Download PDFInfo
- Publication number
- CN113034381A CN113034381A CN202110184461.9A CN202110184461A CN113034381A CN 113034381 A CN113034381 A CN 113034381A CN 202110184461 A CN202110184461 A CN 202110184461A CN 113034381 A CN113034381 A CN 113034381A
- Authority
- CN
- China
- Prior art keywords
- prediction
- kernel
- image
- convolution
- noise
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 42
- 230000006835 compression Effects 0.000 claims abstract description 16
- 238000007906 compression Methods 0.000 claims abstract description 16
- 238000000605 extraction Methods 0.000 claims abstract description 16
- 230000004927 fusion Effects 0.000 claims abstract description 12
- 238000005457 optimization Methods 0.000 claims abstract description 8
- 238000012549 training Methods 0.000 claims description 19
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims description 13
- 230000006870 function Effects 0.000 claims description 10
- 239000011800 void material Substances 0.000 claims description 8
- 238000004590 computer program Methods 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 5
- 238000010276 construction Methods 0.000 claims description 2
- 230000000694 effects Effects 0.000 abstract description 3
- 238000007792 addition Methods 0.000 description 6
- 238000013135 deep learning Methods 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 238000013461 design Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 235000019580 granularity Nutrition 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/50—Image enhancement or restoration using two or more images, e.g. averaging or subtraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20212—Image combination
- G06T2207/20221—Image fusion; Image merging
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a single image denoising method and a single image denoising device based on a cavitated kernel prediction network, which comprise the following steps: constructing a nuclear prediction network based on voiding, which comprises a feature extraction module, a feature compression module, a nuclear prediction module and an image reconstruction module; performing parameter optimization on the cavitated core prediction network for later use; when the method is applied, a noise image is input into a kernel prediction network based on cavitation after parameter optimization, a feature extraction unit is used for extracting an advanced feature map from the noise image, a feature compression module is used for compressing the advanced feature map into a compressed feature map, a kernel prediction module is used for extracting a prediction convolution kernel according to the advanced feature map, an image reconstruction module is used for obtaining the prediction map based on the prediction convolution kernel and the compressed feature map and then fusing the prediction map with the noise image, and the denoised image is obtained. A kernel prediction network is introduced for the single image denoising problem, and the single image denoising task is realized by using techniques such as cavity convolution, multi-kernel channel fusion, feature map compression and the like, so that the denoising effect is greatly improved.
Description
Technical Field
The invention relates to the field of computer science image processing, in particular to a single image denoising method and device based on a cavitated kernel prediction network.
Background
Image denoising is a big basic problem in the field of image processing, and in recent years, the rapid development of a deep learning network provides an efficient solution for a denoising algorithm. However, the conventional convolution network uses the same convolution kernel on the whole image pixel, and the method of joint noise estimation and image denoising based on deep learning, as disclosed in application publication No. CN109658348A, does not fully utilize the characteristic difference existing between pixel regions.
Although some recent researches have attempted to introduce an attention mechanism to emphasize this local difference, such as an attention mechanism-based image adaptive denoising method disclosed in application publication No. CN111260591A, the existing deep learning network still cannot well maintain fine image details.
A Kernel Prediction Network (Kernel Prediction Network) is used as a deep learning Network capable of predicting a special convolution Kernel of each pixel, and the special neighborhood characteristics of each pixel are fully evaluated. However, the method is not well applied to the single image denoising problem at present, and the main difficulty is that enough information is extracted from only one reference image for the pixel-by-pixel kernel prediction.
Disclosure of Invention
In view of the foregoing, an object of the present invention is to provide a single image denoising method and apparatus based on a cavitated kernel prediction network. The method and the device introduce a kernel prediction network for the single image denoising problem, and the kernel prediction network is enabled to be better competent for a single image denoising task by using techniques such as cavity convolution, multi-kernel channel fusion, feature map compression and the like, so that the denoising effect is greatly improved.
In order to achieve the purpose, the invention provides the following technical scheme:
in a first aspect, a single image denoising method based on a cavitated kernel prediction network includes the following steps:
constructing a kernel prediction network based on voiding, which comprises a feature extraction module, a feature compression module, a kernel prediction module and an image reconstruction module, wherein the kernel prediction module comprises at least 3 void branches with different void step lengths and used for generating branch prediction convolution kernels, and a kernel fusion operation for fusing all the branch prediction convolution kernels to obtain prediction convolution kernels;
performing parameter optimization on the cavitated core prediction network for later use;
when the method is applied, a noise image is input into a kernel prediction network based on cavitation after parameter optimization, a feature extraction unit is used for extracting an advanced feature map from the noise image, a feature compression module is used for compressing the advanced feature map into a compressed feature map, a kernel prediction module is used for extracting a prediction convolution kernel according to the advanced feature map, an image reconstruction module is used for obtaining the prediction map based on the prediction convolution kernel and the compressed feature map and then fusing the prediction map with the noise image, and the denoised image is obtained.
Preferably, the feature extraction module includes several convolutional layers and several residual blocks (residual blocks) for extracting a high-level feature map (high-level feature map) including a plurality of channels from the noise image, wherein the residual blocks are designed according to a residual learning strategy, and stack main units of network depth for the feature extraction module, including convolutional layers, a linear rectification function (ReLU), and a dot addition operation.
Preferably, the feature compression module includes at least one convolution layer for compressing the advanced feature map to obtain a compressed feature map with the same number of channels as the noise image, for example, the compressed feature map is a single channel for a grayscale image and a three channel for a color image.
Preferably, the kernel prediction module includes 3 hole branches of different hole step sizes (e.g. 1, 2, 3), each hole branch being used to generate a respective branch prediction convolution kernel, and a kernel fusion operation for fusing all the branch prediction convolution kernels to obtain the prediction convolution kernel.
Preferably, each hole branch comprises a number of hole residual blocks (scaled residual blocks) comprising hole convolution layers (scaled convolution), linear rectification functions and point addition operations, and at least one convolution layer for extracting a branch prediction convolution kernel from the high-level feature map.
Preferably, the kernel fusion operation includes splicing all the branch prediction convolution kernels in the channel dimension, and then compressing the channel number by using the convolution layer to obtain the final prediction convolution kernel.
Preferably, the image reconstruction module includes a pixel-by-pixel convolution operation, at least one convolution layer, and a point addition operation, the pixel-by-pixel convolution operation is configured to apply an input prediction convolution kernel to the compressed feature map to obtain a prediction map, the convolution layer is configured to convert the prediction map into a noise residual image, and the point addition operation is configured to fuse the noise residual image and the noise image to obtain a denoised image.
Preferably, when the parameters of the voided kernel prediction network are optimized, a training sample pair including a clean image with a fixed noise level and a noise image is input into the voided kernel prediction network, and an end-to-end network training is performed by using a mean square error loss function as a loss function.
Preferably, the construction process of the training sample pair is as follows: selecting a clean image, generating noise with a fixed noise level, adding the noise to the clean image to obtain a noise image, wherein the noise image and the clean image form a training sample pair; a single training involves only one noise level.
In a second aspect, a single image denoising device based on a cavitated nuclear prediction network comprises a computer memory, a computer processor and a computer program stored in the computer memory and executable on the computer processor, wherein the computer processor implements the single image denoising method based on the cavitated nuclear prediction network when executing the computer program.
Compared with the prior art, the single image denoising method and device based on the cavitation kernel prediction network provided by the embodiment of the invention have the beneficial effects that at least:
(1) the kernel prediction module is constructed to evaluate neighborhood information of each pixel and generate a specific prediction convolution kernel for each pixel, and the characteristic improves the sensitivity of the network to the difference of each pixel, so that finer image details are reserved in the denoising process;
(2) the hole convolution is introduced into the kernel prediction module, and the reception field (reliable field) is greatly increased, so that the network can view a larger pixel neighborhood, and more information is provided for the prediction of the convolution kernel. The design simultaneously enables a predicted convolution kernel to keep smaller size due to the fact that enough neighborhood information is obtained, and therefore the correlation calculation amount is reduced;
(3) three-way parallel cavity branch structure and core fusion strategy are designed. The branches with different cavity step lengths provide neighborhood characteristic perception with different granularities, and the fusion of the three branches ensures that the final prediction convolution kernel is more accurate;
(4) the compressed feature map is designed to replace the original image, and the generated prediction convolution kernel acts on the compressed feature map instead of the original image. Due to the multiplexing of the advanced feature extraction module with the core prediction module, the design adds only a very small amount of additional parameters to the network. And since the compressed feature map can be adjusted in a learning way in the network training process, the compressed feature map is more matched with the final prediction convolution kernel.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without creative efforts.
FIG. 1 is a flowchart of a method for denoising a single image by using a kernel prediction network based on cavitation according to an embodiment of the present invention;
fig. 2(a) and fig. 2(b) are structural diagrams of a residual block and a hole residual block according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be further described in detail with reference to the accompanying drawings and examples. It should be understood that the detailed description and specific examples, while indicating the scope of the invention, are intended for purposes of illustration only and are not intended to limit the scope of the invention.
Fig. 1 is a flowchart of a method for denoising a single image by using a kernel prediction network based on cavitation according to an embodiment of the present invention. Referring to fig. 1, the single image denoising method includes the following processes:
a training data set is prepared. The original training image is taken from the BSD 300. Selecting a clean original image, generating Gaussian noise with a selected noise level, generating the Gaussian noise with a fixed noise level only by single training, adding the noise to the original image, and obtaining a pair of training samples by the obtained noise image and the original image. The number of training samples is greatly increased by a method of cutting image blocks, and data enhancement methods such as multi-size transformation, random rotation, random mirror image and the like are assisted.
And building a core prediction network based on cavitation. As shown in fig. 1, the nuclear prediction network based on voiding includes a feature extraction module, a feature compression module, a nuclear prediction module, and an image reconstruction module, wherein the feature extraction module includes a plurality of convolution layers and a plurality of residual error modules, the feature compression module includes a plurality of convolution layers, the nuclear prediction module includes 3 void branches with different void step lengths and a nuclear fusion operation, each void branch includes a plurality of void residual error blocks and convolution layers, and the image reconstruction module includes a pixel-by-pixel convolution operation, a convolution layer, and a point addition operation.
The input noise image is subjected to feature extraction by a feature extraction module to obtain an advanced feature map, and the advanced feature map is simultaneously input into a feature compression module and a kernel prediction module; the feature compression module performs channel compression processing on the high-level feature map by using the convolution layer to obtain a single-channel compressed feature map corresponding to the input single-channel gray image; and after the kernel prediction module channel generates respective branch prediction convolution kernels according to the high-level characteristic diagram by using the three cavity branches, fusing the three branch prediction convolution kernels through kernel fusion operation to obtain a final prediction convolution kernel. In the image reconstruction module, after a prediction convolution kernel acts on a compressed characteristic image through pixel-by-pixel convolution to obtain a prediction image, the prediction image passes through a convolution layer to obtain a noise residual image, and the noise residual image is combined with an original input noise image to obtain a denoised image. Due to the characteristicsThe structure of the residual block in the extraction module and the hole residual block in the kernel prediction module are relatively complex, and details are shown in fig. 2, in this embodiment, the number of the residual block and the number of the hole residual block are respectively 9 and 2, that is, the number corresponds to n in fig. 21And n2。
The voided core prediction network of the invention can be trained end-to-end. The loss function selected during training is the mean square error loss function. To improve training efficiency and prevent over-fitting, the optimizer used was an Adam optimizer, the initial learning rate was set to 0.0001, and decayed to one tenth of the original value every 30 epochs. The mini-batch size used for training was 32. And storing the trained model parameters for subsequent denoising.
When the method is applied, a noise image is input into a kernel prediction network based on cavitation after parameter optimization, a feature extraction unit is used for extracting an advanced feature map from the noise image, a feature compression module is used for compressing the advanced feature map into a compressed feature map, a kernel prediction module is used for extracting a prediction convolution kernel according to the advanced feature map, an image reconstruction module is used for obtaining the prediction map based on the prediction convolution kernel and the compressed feature map and then fusing the prediction map with the noise image, and the denoised image is obtained.
The embodiment also provides a single image denoising device based on the cavitated nuclear prediction network, which comprises a computer memory, a computer processor and a computer program stored in the computer memory and executable on the computer processor, wherein the computer processor implements the single image denoising method based on the cavitated nuclear prediction network when executing the computer program.
In practical applications, the computer memory may be volatile memory at the near end, such as RAM, or may be non-volatile memory, such as ROM, FLASH, floppy disk, mechanical hard disk, etc., or may be a remote storage cloud. The computer processor can be a Central Processing Unit (CPU), a Microprocessor (MPU), a Digital Signal Processor (DSP), or a Field Programmable Gate Array (FPGA), i.e., the single image denoising method steps based on the cavitated kernel prediction network can be implemented by these processors.
Compared with the existing method, the result obtained by the single image denoising method provided by the invention has better PSNR index result and better subjective denoising effect.
The above-mentioned embodiments are intended to illustrate the technical solutions and advantages of the present invention, and it should be understood that the above-mentioned embodiments are only the most preferred embodiments of the present invention, and are not intended to limit the present invention, and any modifications, additions, equivalents, etc. made within the scope of the principles of the present invention should be included in the scope of the present invention.
Claims (10)
1. A single image denoising method based on a cavitated kernel prediction network is characterized by comprising the following steps:
constructing a kernel prediction network based on voiding, which comprises a feature extraction module, a feature compression module, a kernel prediction module and an image reconstruction module, wherein the kernel prediction module comprises at least 3 void branches with different void step lengths and used for generating branch prediction convolution kernels, and a kernel fusion operation for fusing all the branch prediction convolution kernels to obtain prediction convolution kernels;
performing parameter optimization on the cavitated core prediction network for later use;
when the method is applied, a noise image is input into a kernel prediction network based on cavitation after parameter optimization, a feature extraction unit is used for extracting an advanced feature map from the noise image, a feature compression module is used for compressing the advanced feature map into a compressed feature map, a kernel prediction module is used for extracting a prediction convolution kernel according to the advanced feature map, an image reconstruction module is used for obtaining the prediction map based on the prediction convolution kernel and the compressed feature map and then fusing the prediction map with the noise image, and the denoised image is obtained.
2. The method of claim 1, wherein the feature extraction module comprises a plurality of convolutional layers and a plurality of residual modules, and is configured to extract an advanced feature map comprising a plurality of channels from the noisy image, and the residual modules comprise convolutional layers, linear rectification functions and point addition operations.
3. The single-image denoising method based on the cavitation kernel prediction network as claimed in claim 1, wherein the feature compression module comprises at least one convolution layer for compressing the high-level feature map to obtain the same number of compressed feature maps as the number of noise image channels.
4. The method of claim 1, wherein the kernel prediction module comprises 3 hole branches of different hole step sizes, each hole branch being configured to generate a respective branch prediction convolution kernel, and further comprising a kernel fusion operation configured to fuse all the branch prediction convolution kernels to obtain the prediction convolution kernel.
5. The method of claim 1 or 4, wherein each hole branch comprises a number of hole residual blocks and at least one convolution layer for extracting a branch prediction convolution kernel from the high-level feature map, and the hole residual blocks comprise hole convolution layers, linear rectification functions and point addition operations.
6. The single-image denoising method based on the cavitation kernel prediction network as claimed in claim 1 or 4, wherein the kernel fusion operation includes splicing all branch prediction convolution kernels in channel dimension, and then compressing the number of channels by using convolution layers to obtain a final prediction convolution kernel.
7. The method as claimed in claim 1, wherein the image reconstruction module comprises a pixel-by-pixel convolution operation, at least one convolution layer and a point addition operation, the pixel-by-pixel convolution operation is used for applying an input prediction convolution kernel to the compressed feature map to obtain a prediction map, the convolution layer is used for converting the prediction map into a noise residual image, and the point addition operation is used for fusing the noise residual image and the noise image to obtain a denoised image.
8. The single-image denoising method based on the voided kernel prediction network of claim 1, wherein when performing parameter optimization on the voided kernel prediction network, a training sample pair comprising a clean image with a fixed noise level and a noise image is input into the voided kernel prediction network, and an end-to-end network training is performed by using a mean square error loss function as a loss function.
9. The single-image denoising method based on the cavitation-based kernel prediction network of claim 8, wherein the construction process of the training sample pair is as follows: selecting a clean image, generating noise with a fixed noise level, adding the noise to the clean image to obtain a noise image, wherein the noise image and the clean image form a training sample pair; a single training involves only one noise level.
10. A single image denoising device based on a cavitated nuclear prediction network, comprising a computer memory, a computer processor and a computer program stored in the computer memory and executable on the computer processor, wherein the computer processor implements the single image denoising method based on the cavitated nuclear prediction network according to any one of claims 1 to 9 when executing the computer program.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110184461.9A CN113034381B (en) | 2021-02-08 | 2021-02-08 | Single image denoising method and device based on cavitated kernel prediction network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110184461.9A CN113034381B (en) | 2021-02-08 | 2021-02-08 | Single image denoising method and device based on cavitated kernel prediction network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113034381A true CN113034381A (en) | 2021-06-25 |
CN113034381B CN113034381B (en) | 2022-06-21 |
Family
ID=76460908
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110184461.9A Active CN113034381B (en) | 2021-02-08 | 2021-02-08 | Single image denoising method and device based on cavitated kernel prediction network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113034381B (en) |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109685072A (en) * | 2018-12-22 | 2019-04-26 | 北京工业大学 | A kind of compound degraded image high quality method for reconstructing based on generation confrontation network |
CN110113593A (en) * | 2019-06-11 | 2019-08-09 | 南开大学 | Wide baseline multi-view point video synthetic method based on convolutional neural networks |
CN110913218A (en) * | 2019-11-29 | 2020-03-24 | 合肥图鸭信息科技有限公司 | Video frame prediction method and device and terminal equipment |
CN111007566A (en) * | 2019-12-27 | 2020-04-14 | 西南石油大学 | Curvature-driven diffusion full-convolution network seismic data bad channel reconstruction and denoising method |
CN111402128A (en) * | 2020-02-21 | 2020-07-10 | 华南理工大学 | Image super-resolution reconstruction method based on multi-scale pyramid network |
CN111667445A (en) * | 2020-05-29 | 2020-09-15 | 湖北工业大学 | Image compressed sensing reconstruction method based on Attention multi-feature fusion |
-
2021
- 2021-02-08 CN CN202110184461.9A patent/CN113034381B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109685072A (en) * | 2018-12-22 | 2019-04-26 | 北京工业大学 | A kind of compound degraded image high quality method for reconstructing based on generation confrontation network |
CN110113593A (en) * | 2019-06-11 | 2019-08-09 | 南开大学 | Wide baseline multi-view point video synthetic method based on convolutional neural networks |
CN110913218A (en) * | 2019-11-29 | 2020-03-24 | 合肥图鸭信息科技有限公司 | Video frame prediction method and device and terminal equipment |
CN111007566A (en) * | 2019-12-27 | 2020-04-14 | 西南石油大学 | Curvature-driven diffusion full-convolution network seismic data bad channel reconstruction and denoising method |
CN111402128A (en) * | 2020-02-21 | 2020-07-10 | 华南理工大学 | Image super-resolution reconstruction method based on multi-scale pyramid network |
CN111667445A (en) * | 2020-05-29 | 2020-09-15 | 湖北工业大学 | Image compressed sensing reconstruction method based on Attention multi-feature fusion |
Non-Patent Citations (3)
Title |
---|
EL HELOU, M.,ET.AL: "Blind universal Bayesian image denoising with Gaussian noise level learning", 《IEEE TRANSACTIONS ON IMAGE PROCESSING》 * |
MARINČ, T.,ET.AL: "Multi-kernel prediction networks for denoising of burst images", 《IN 2019 IEEE INTERNATIONAL CONFERENCE ON IMAGE PROCESSING (ICIP) 》 * |
孙忠凡等: "基于空间元学习的放大任意倍的超分辨率重建方法", 《计算机应用》 * |
Also Published As
Publication number | Publication date |
---|---|
CN113034381B (en) | 2022-06-21 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11250595B2 (en) | Tiled image compression using neural networks | |
CN109101975B (en) | Image semantic segmentation method based on full convolution neural network | |
CN107689034B (en) | Denoising method and denoising device | |
US20220261965A1 (en) | Training method of image processing model, image processing method, apparatus, and device | |
CN110580704A (en) | ET cell image automatic segmentation method and system based on convolutional neural network | |
CN113989129A (en) | Image restoration method based on gating and context attention mechanism | |
CN113014927A (en) | Image compression method and image compression device | |
CN110675339A (en) | Image restoration method and system based on edge restoration and content restoration | |
US11948281B2 (en) | Guided up-sampling for image inpainting | |
CN111192211A (en) | Multi-noise type blind denoising method based on single deep neural network | |
CN112330761B (en) | Deep ECT image enhancement method based on mixed precision training | |
CN113487618A (en) | Portrait segmentation method, portrait segmentation device, electronic equipment and storage medium | |
CN116681584A (en) | Multistage diffusion image super-resolution algorithm | |
CN114373104A (en) | Three-dimensional point cloud semantic segmentation method and system based on dynamic aggregation | |
CN116778165A (en) | Remote sensing image disaster detection method based on multi-scale self-adaptive semantic segmentation | |
Zhou et al. | MSAR‐DefogNet: Lightweight cloud removal network for high resolution remote sensing images based on multi scale convolution | |
CN111524060B (en) | System, method, storage medium and device for blurring portrait background in real time | |
CN113034381B (en) | Single image denoising method and device based on cavitated kernel prediction network | |
CN117651144A (en) | Deep learning-based building point cloud compression method and system | |
CN110472732B (en) | Image feature extraction system based on optimized feature extraction device | |
CN116823610A (en) | Deep learning-based underwater image super-resolution generation method and system | |
CN115941966A (en) | Video compression method and electronic equipment | |
CN116051850A (en) | Neural network target detection method, device, medium and embedded electronic equipment | |
Jin et al. | HyA-GAN: remote sensing image cloud removal based on hybrid attention generation adversarial network | |
CN112529064A (en) | Efficient real-time semantic segmentation method |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |