CN109410149B - CNN denoising method based on parallel feature extraction - Google Patents
CNN denoising method based on parallel feature extraction Download PDFInfo
- Publication number
- CN109410149B CN109410149B CN201811323811.XA CN201811323811A CN109410149B CN 109410149 B CN109410149 B CN 109410149B CN 201811323811 A CN201811323811 A CN 201811323811A CN 109410149 B CN109410149 B CN 109410149B
- Authority
- CN
- China
- Prior art keywords
- feature extraction
- cnn
- network model
- image
- cnn denoising
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 50
- 238000000605 extraction Methods 0.000 title claims abstract description 47
- 238000012549 training Methods 0.000 claims abstract description 26
- 230000006870 function Effects 0.000 claims description 14
- 238000013178 mathematical model Methods 0.000 claims description 9
- 230000004913 activation Effects 0.000 claims description 6
- 230000004927 fusion Effects 0.000 claims description 5
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 3
- 230000003247 decreasing effect Effects 0.000 claims description 3
- 238000013527 convolutional neural network Methods 0.000 description 47
- 230000000694 effects Effects 0.000 description 17
- 230000009467 reduction Effects 0.000 description 14
- 238000001914 filtration Methods 0.000 description 12
- 238000012360 testing method Methods 0.000 description 9
- 238000002474 experimental method Methods 0.000 description 5
- 238000013507 mapping Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 230000003044 adaptive effect Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 210000000744 eyelid Anatomy 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
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
- 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]
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (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 Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a CNN denoising method based on parallel feature extraction, which comprises six steps: step one, a CNN denoising network model for parallel feature extraction is built; initializing training parameters of a CNN denoising network model; step three, constructing a training set; designing a loss function, and training a CNN denoising network model by taking the minimum loss function as a target to obtain a CNN denoising model; taking the noise image as the input of the CNN denoising model, wherein the output is the noise information learned by the network model; and step six, subtracting the noise information learned in the step five from the noise image to obtain a denoised clean image. The invention can thoroughly remove noise, well retain the texture information of the image and obviously improve objective indexes PSNR and SSIM.
Description
Technical Field
The invention relates to the field of image denoising, in particular to a CNN denoising method based on parallel feature extraction.
Background
In reality, digital images are often affected by interference of imaging equipment and external environment noise during digitization and transmission, and are called noisy images or noisy images. The final purpose of image denoising is to improve a given image and solve the problem that the quality of the actual image is reduced due to noise interference. The image quality can be effectively improved through the denoising technology, the signal-to-noise ratio is increased, the information carried by the original image is better reflected, and people widely study the image denoising algorithm as an important preprocessing means.
At present, the classical methods for image denoising are many, but can be roughly divided into two types, one is based on spatial domain filtering, such as mean filtering, median filtering and the like; the other is based on transform domain filtering, such as bayesian least squares in gaussian scale mixture models. Some of the existing denoising algorithms obtain better effect in low-dimensional signal image processing, but are not suitable for high-dimensional signal image processing; or the denoising effect is good, but the edge information of the partial image is lost; or the research is carried out to detect the edge information of the image, the image details are kept, the filtering is not carried out in the global range, and the relation between the natural image blocks and the blocks is not considered. Therefore, the denoising effect obtained by the existing method is not satisfactory.
In order to solve the problems of the conventional denoising method, a neural network is used for image denoising. The patent "intelligent filtering method and system for CNN-LMS image noise" (patent number: 201810128238.0) discloses a method for obtaining a filtered image by embedding a CNN intelligent control model in an LMS adaptive filtering system, adjusting parameters of the LMS adaptive filtering system, and filtering or suppressing image noise to remove image noise. The patent "an image denoising method based on a compressed convolutional neural network" (patent No. 201710286383.7) discloses a method for performing image denoising by replacing a convolutional layer of an original denoising convolutional neural network with a convolutional layer decompressed by low-rank matrix decomposition. The patent "an image denoising method based on a ReLU convolutional neural network" (patent number: 201610482594.3) discloses a real-time denoising method based on a ReLU convolutional neural network model. They differ from the present design in that:
(1) two parallel MPFE feature extraction modules are designed in the invention.
(2) The present invention uses dense connections to forward extracted features from the bottom layer to higher layers.
(3) The method fuses the features extracted at the bottom layer and the features of different scales advanced by the MPFE feature extraction module, so that the extracted features represent the image information to the maximum extent.
Compared with the intelligent filtering method and the intelligent filtering system for CNN-LMS image noise, the image denoising method based on the ReLU convolutional neural network and the image denoising method based on the ReLU convolutional neural network, the method has the advantages that:
(1) the invention carries out zero filling operation before each convolution layer, ensures that the size of the image is not changed, and can reserve the edge information of the image as much as possible.
(2) The invention uses 5 feature extraction modules MPFE which are two-path parallel networks, and different convolution kernels are respectively used for extracting different features, thereby being convenient for realizing high-quality denoising.
(3) The invention uses dense connection to connect the image characteristics provided by different depths together, and ensures that the network can fully utilize various characteristics of the image to learn the mapping relation between input and output after characteristic fusion.
The invention aims to provide a high-quality image denoising method, which can retain the edge information and detail information of an image as much as possible while denoising.
Disclosure of Invention
In order to overcome the defects in the prior art, the invention aims to provide an image denoising method based on multi-scale parallel CNN, so as to improve the denoising effect of an image.
The invention relates to a CNN denoising method based on parallel feature extraction, which is characterized by comprising the following steps:
step one, a CNN denoising network model for parallel feature extraction is built;
initializing training parameters of a CNN denoising network model;
step three, constructing a training set;
designing a loss function, and training a CNN denoising network model by taking the minimum loss function as a target to obtain a CNN denoising model;
taking the noise image as the input of the CNN denoising model, wherein the output is the noise information learned by the network model;
and step six, subtracting the noise information learned in the step five from the noise image to obtain a denoised clean image.
Further, the CNN denoising method based on parallel feature extraction is characterized in that in the step one, the established CNN denoising network model based on parallel feature extraction comprises 5 feature extraction modules MPFE which is a two-path parallel network, one path is a convolution kernel with the length of 3 multiplied by 3 in series and the length of 5 multiplied by 5 in series, the other path is a convolution kernel with the length of 5 multiplied by 5 in series and the length of 3 multiplied by 3, finally, the two paths are subjected to feature fusion, the mathematical model of the MPFE is as follows,
where n 1, 2., 5, ω and b represent weights and offsets, respectively, the superscript represents the number of layers in which it is located, the subscript represents the convolution kernel size, d represents the input channel, MPnIAnd MPnORepresents the input and output of the n-th MPFE, [ MP ]nI,A2,B2]Indicating the serial operation of the features.
Further, the CNN denoising method based on parallel feature extraction is characterized in that in 5 feature extraction modules MPFE included in the built CNN denoising network model based on parallel feature extraction, the input mathematical models of the first MPFE and the second MPFE are,
wherein the meanings of the upper and lower indices are the same as in formula (1).
Further, the CNN denoising method based on parallel feature extraction is characterized in that in the step one, the built CNN denoising network model based on parallel feature extraction comprises 22 convolution layers, the size of the convolution core is 3 x 3 or 1 x 1, wherein an activation layer with an activation function of ReLU is arranged behind the 3 x 3 convolution core, the mathematical model of the CNN denoising network model is as follows,
wherein the meanings of the upper and lower indices are the same as in formula (1).
Further, the CNN denoising method based on parallel feature extraction is characterized in that, in the second step, training parameters of the CNN denoising network model are specifically set as: training 75 generations together, using Adam as the optimizer, the initial value of learning efficiency was set to 0.001, decreasing by half every 10 generations, with batch _ size set to 64 in each generation and step _ per _ epoch set to 2000.
Further, the CNN denoising method based on parallel feature extraction is characterized in that in the third step, the training set is constructed by adding Gaussian noise with determined concentration randomly into 400 standard grayscale images of 180 × 180; cutting the standard image into a plurality of 40 x 40 image blocks according to the step length of 10; and then turning over each image block up and down, rotating at any angle and the like to finally obtain 23.84 ten thousand image blocks to form a training set.
Further, the CNN denoising method based on parallel feature extraction is characterized in that in the fourth step, the designed loss function is,
wherein Y isiAndwhich respectively represent the ideal clean image and the estimated clean image corresponding to the ith noise image, and ω and b represent the weight and the offset, respectively.
Compared with the prior art, the invention has the following technical effects:
(1) the invention carries out zero filling operation before each convolution layer, ensures that the size of the image is not changed, and can reserve the edge information of the image as much as possible.
(2) The invention uses 5 feature extraction modules MPFE which are two-path parallel networks, and different convolution kernels are respectively used for extracting different features, thereby being convenient for realizing high-quality denoising.
(3) The invention uses dense connection to connect the image characteristics provided by different depths together, and ensures that the network can fully utilize various characteristics of the image to learn the mapping relation between input and output after characteristic fusion.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
FIG. 1 is a flow chart of denoising according to the present invention;
FIG. 2 is a block diagram of the feature extraction module MPFE;
FIG. 3 is a CNN denoising network model diagram based on parallel feature extraction;
FIG. 4 is a 6 widely used test images;
FIG. 5 is a diagram of the denoising result of the present invention and the prior denoising method;
wherein (a) standard image, (b) noise image/14.14 dB, (c) result of BM 3D/29.85 dB, (d) result of WNNM/30.28 dB, (e) result of EPLL/29.08 dB, (f) result of TNRD/29.53 dB, (g) result of MLP/29.94 dB, (h) result of DnCNN-S/30.36 dB, (i) result of the present invention/30.59 dB.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
As shown in fig. 2, the present invention discloses a CNN denoising method based on parallel feature extraction, which includes six steps. Step S1, constructing a CNN denoising network model for parallel feature extraction; step S2, initializing the training parameters of the CNN denoising network model; step S3, constructing a training set; step S4, designing a loss function, and training a CNN denoising network model by taking the minimum loss function as a target to obtain a CNN denoising model; step S5, taking the noise image as the input of the CNN denoising model, wherein the output is the noise information learned by the network model; and step S6, subtracting the noise information learned in the step five from the noise image to obtain a denoised clean image.
Further, the CNN denoising method based on parallel feature extraction is characterized in that, in the step one, the established parallel feature extraction CNN denoising network model comprises 5 feature extraction modules MPFE, as shown in fig. 2, which is a two-path parallel network, one path is a 3 × 3 series 5 × 5 convolution kernel, the other path is a 5 × 5 series 3 × 3 convolution kernel, and finally, the two paths are subjected to feature fusion, the mathematical model of MPFE is,
where n 1, 2., 5, ω and b represent weights and offsets, respectively, the superscript represents the number of layers in which it is located, the subscript represents the convolution kernel size, d represents the input channel, MPnIAnd MPnORepresents the input and output of the n-th MPFE, [ MP ]nI,A2,B2]Indicating the serial operation of the features.
Further, the CNN denoising method based on parallel feature extraction is characterized in that in 5 feature extraction modules MPFE included in the built CNN denoising network model based on parallel feature extraction, the input mathematical models of the first MPFE and the second MPFE are,
wherein the meanings of the upper and lower indices are the same as in formula (1).
Further, the CNN denoising method based on parallel feature extraction is characterized in that, in the first step, as shown in fig. 3, the set up CNN denoising network model based on parallel feature extraction includes 22 convolutional layers, the size of the convolutional core is 3 × 3 or 1 × 1, wherein an activation layer with an activation function of ReLU is arranged behind the 3 × 3 convolutional core, the mathematical model of the CNN denoising network model is,
wherein the meanings of the upper and lower indices are the same as in formula (1).
Further, the CNN denoising method based on parallel feature extraction is characterized in that, in the second step, training parameters of the CNN denoising network model are specifically set as: training 75 generations together, using Adam as the optimizer, the initial value of learning efficiency was set to 0.001, decreasing by half every 10 generations, with batch _ size set to 64 in each generation and step _ per _ epoch set to 2000.
Further, the CNN denoising method based on parallel feature extraction is characterized in that in the third step, the training set is constructed by adding Gaussian noise with determined concentration randomly into 400 standard grayscale images of 180 × 180; cutting the standard image into a plurality of 40 x 40 image blocks according to the step length of 10; and then turning over each image block up and down, rotating at any angle and the like to finally obtain 23.84 ten thousand image blocks to form a training set.
Further, the CNN denoising method based on parallel feature extraction is characterized in that in the fourth step, the designed loss function is,
wherein Y isiAndrespectively representing an ideal clean image and an estimated clean image corresponding to the ith noise imageLike, ω and b represent weight and bias, respectively.
The CNN denoising model is trained by using a training set by taking a formula (4) as an objective function, so that a mapping function between an input noise image and an output can be obtained(which is the noise information learned by the CNN denoising model), subtracting the mapping function from the input noise imageAnd obtaining the denoised clean image.
To verify the effectiveness of the present invention, simulation experiments were performed.
The experiments were run in a Keras environment on PCs with Intel (R) core (TM) i5-8300H CPU2.30GHz and Nvidia1050Ti GPUs.
The training parameters are specifically set to be 75 generations, the learning efficiency is set to be 0.001, the learning efficiency is reduced by half every 10 generations, the base _ size in each generation is set to be 64, the step _ per _ epoch is set to be 2000, 23.84 thousands of 40 × 40 Image blocks are used to form a training set, the denoising model of the invention is trained by using the training set, two experiments are carried out according to different test sets, and the comparison method is respectively compared with several advanced denoising methods, wherein the larger the Noise reduction effect is given by BM3D (K.Dabov, et al, Image denoising by space 3-D transform-Noise source filtering, IEEE transport processing, 2007,16 (8): 2080, 2095), the better the Noise reduction effect is given by BM3, Gu, Weighted Noise reduction effect is given by using the Noise reduction algorithm (R) and the Noise reduction effect is given by IEEE 2. reflection, reflection Noise reduction algorithm, Noise reduction effect is given by using the Noise reduction algorithm, Noise reduction effect is given by the following algorithm, Noise reduction effect is given by the general evaluation of the Noise reduction algorithm, Noise reduction effect, Noise reduction effect, Noise reduction.
Experiment one, using the image in fig. 4 as the test image, table 1 is the test result, with black bold representing the highest index. In the experiment, the noise level sigma is respectively set to be 15, 25 and 50, and the test result shows that the objective indexes of the method are higher than those of other methods, and the denoising effect is better than that of other denoising methods.
TABLE 1 comparison of the results of the process of the invention with several advanced processes
Experiment two, in order to further illustrate the denoising effect of the present invention, BSD68 was selected as a test set and compared with several methods of the most advanced, and the results are shown in table 2:
TABLE 2 test (PSNR) results on BSD68 test set
As can be seen from the test results in Table 2, the denoising method of the present invention can obtain better PSNR and SSIM.
Fig. 5 is a result image of an image with 50% gaussian noise added thereto, which is denoised by different methods and by the method of the present invention, and obviously, the present invention well protects the edge information of the image, retains the feature information of the image, such as the features of the double eyelid, and obtains the best denoising effect.
The above description is only an example of the present invention, and is not intended to limit the scope of the present invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention shall fall within the protection scope of the present invention.
Claims (4)
1. A CNN denoising method based on parallel feature extraction is characterized by comprising the following steps:
step one, a CNN denoising network model for parallel feature extraction is built;
initializing training parameters of a CNN denoising network model;
step three, constructing a training set;
designing a loss function, and training a CNN denoising network model by taking the minimum loss function as a target to obtain a CNN denoising model;
taking the noise image as the input of the CNN denoising model, wherein the output is the noise information learned by the network model;
subtracting the noise information learned in the fifth step from the noise image to obtain a denoised clean image;
in the first step, the built parallel feature extraction CNN denoising network model comprises 5 feature extraction modules MPFE which is a two-path parallel network, one path is a convolution kernel with 3 multiplied by 3 connected in series by 5 multiplied by 5, the other path is a convolution kernel with 5 multiplied by 5 connected in series by 3 multiplied by 3, finally, the two paths are subjected to feature fusion, the mathematical model of the MPFE is,
where n-1, 2, …,5, ω and b represent weights and offsets, respectively, the superscript indicates the number of layers in which it is located, the subscript indicates the convolution kernel size, d indicates the input channel, MPnIAnd MPnORepresents the input and output of the n-th MPFE, [ MP ]nI,A2,B2]Indicating the serial operation of the features.
2. The CNN denoising method based on parallel feature extraction as claimed in claim 1, wherein, in the 5 feature extraction modules MPFE included in the built CNN denoising network model based on parallel feature extraction, the input mathematical models of the first and second MPFE are,
wherein the meanings of the upper and lower indices are the same as in formula (1).
3. The CNN denoising method based on parallel feature extraction as claimed in claim 1, wherein in the step one, the built CNN denoising network model of parallel feature extraction comprises 22 convolutional layers, the size of the convolutional core is 3 x 3 or 1 x 1, wherein an activation layer with an activation function of ReLU is arranged behind the 3 x 3 convolutional core, the mathematical model of the CNN denoising network model is,
wherein the meanings of the upper and lower indices are the same as in formula (1).
4. The CNN denoising method based on parallel feature extraction as claimed in claim 1, wherein in the second step, the training parameters of the CNN denoising network model are specifically set as: training 75 generations together, using Adam as the optimizer, the initial value of learning efficiency was set to 0.001, decreasing by half every 10 generations, with batch _ size set to 64 in each generation and step _ per _ epoch set to 2000.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811323811.XA CN109410149B (en) | 2018-11-08 | 2018-11-08 | CNN denoising method based on parallel feature extraction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811323811.XA CN109410149B (en) | 2018-11-08 | 2018-11-08 | CNN denoising method based on parallel feature extraction |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109410149A CN109410149A (en) | 2019-03-01 |
| CN109410149B true CN109410149B (en) | 2019-12-31 |
Family
ID=65472193
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811323811.XA Active CN109410149B (en) | 2018-11-08 | 2018-11-08 | CNN denoising method based on parallel feature extraction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109410149B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110189260B (en) * | 2019-04-15 | 2021-01-26 | 浙江大学 | Image noise reduction method based on multi-scale parallel gated neural network |
| CN110782414A (en) * | 2019-10-30 | 2020-02-11 | 福州大学 | Dark light image denoising method based on dense connection convolution |
| CN110852966B (en) * | 2019-11-04 | 2022-04-22 | 西北工业大学 | Image noise estimation method based on deep convolutional neural network |
| CN111275643B (en) * | 2020-01-20 | 2022-09-02 | 西南科技大学 | Real noise blind denoising network system and method based on channel and space attention |
| CN111667424B (en) * | 2020-05-28 | 2022-04-01 | 武汉大学 | Unsupervised real image denoising method |
| CN111860273B (en) * | 2020-07-14 | 2022-07-05 | 吉林大学 | Magnetic resonance underground water detection noise suppression method based on convolutional neural network |
| CN112488934B (en) * | 2020-11-26 | 2024-02-09 | 杭州电子科技大学 | CS-TCGAN-based finger vein image denoising method |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107862705A (en) * | 2017-11-21 | 2018-03-30 | 重庆邮电大学 | A kind of unmanned plane small target detecting method based on motion feature and deep learning feature |
| CN108280811A (en) * | 2018-01-23 | 2018-07-13 | 哈尔滨工业大学深圳研究生院 | A kind of image de-noising method and system based on neural network |
| CN108446661A (en) * | 2018-04-01 | 2018-08-24 | 桂林电子科技大学 | A kind of deep learning parallelization face identification method |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10032256B1 (en) * | 2016-11-18 | 2018-07-24 | The Florida State University Research Foundation, Inc. | System and method for image processing using automatically estimated tuning parameters |
| CN106875361B (en) * | 2017-02-17 | 2019-06-04 | 深圳市唯特视科技有限公司 | A method of poisson noise is removed based on depth convolutional neural networks |
| US10475165B2 (en) * | 2017-04-06 | 2019-11-12 | Disney Enterprises, Inc. | Kernel-predicting convolutional neural networks for denoising |
| CN107590518A (en) * | 2017-08-14 | 2018-01-16 | 华南理工大学 | A kind of confrontation network training method of multiple features study |
| CN107689034B (en) * | 2017-08-16 | 2020-12-01 | 清华-伯克利深圳学院筹备办公室 | Denoising method and denoising device |
| CN107516304A (en) * | 2017-09-07 | 2017-12-26 | 广东工业大学 | A kind of image de-noising method and device |
| CN108564555B (en) * | 2018-05-11 | 2021-09-21 | 中北大学 | NSST and CNN-based digital image noise reduction method |
-
2018
- 2018-11-08 CN CN201811323811.XA patent/CN109410149B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107862705A (en) * | 2017-11-21 | 2018-03-30 | 重庆邮电大学 | A kind of unmanned plane small target detecting method based on motion feature and deep learning feature |
| CN108280811A (en) * | 2018-01-23 | 2018-07-13 | 哈尔滨工业大学深圳研究生院 | A kind of image de-noising method and system based on neural network |
| CN108446661A (en) * | 2018-04-01 | 2018-08-24 | 桂林电子科技大学 | A kind of deep learning parallelization face identification method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109410149A (en) | 2019-03-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109410149B (en) | CNN denoising method based on parallel feature extraction | |
| CN109003240B (en) | Image denoising method based on multi-scale parallel CNN | |
| CN110599409B (en) | Convolutional neural network image denoising method based on multi-scale convolutional groups and parallel | |
| CN106204467B (en) | Image denoising method based on cascade residual error neural network | |
| CN110378844B (en) | Image blind motion blur removing method based on cyclic multi-scale generation countermeasure network | |
| CN106204468B (en) | A kind of image de-noising method based on ReLU convolutional neural networks | |
| CN110163813B (en) | Image rain removing method and device, readable storage medium and terminal equipment | |
| CN108230278B (en) | Image raindrop removing method based on generation countermeasure network | |
| CN113450288B (en) | Single image rain removing method and system based on deep convolutional neural network and storage medium | |
| CN111080567A (en) | Remote sensing image fusion method and system based on multi-scale dynamic convolution neural network | |
| CN107274379B (en) | Image quality evaluation method and system | |
| CN111260591B (en) | Image self-adaptive denoising method based on attention mechanism | |
| CN110517195B (en) | Unsupervised SAR image denoising method | |
| CN107085835B (en) | Color image filtering method based on quaternary number Weighted Kernel Norm minimum | |
| CN110070539A (en) | Image quality evaluating method based on comentropy | |
| CN111179196B (en) | Multi-resolution depth network image highlight removing method based on divide-and-conquer | |
| CN113256510A (en) | CNN-based low-illumination image enhancement method with color restoration and edge sharpening effects | |
| CN113284061A (en) | Underwater image enhancement method based on gradient network | |
| CN109949334B (en) | Contour detection method based on deep reinforced network residual error connection | |
| CN114511521A (en) | Tire defect detection method based on multiple representations and multiple sub-field self-adaption | |
| Zeng et al. | Progressive feature fusion attention dense network for speckle noise removal in OCT images | |
| CN113191962A (en) | Underwater image color recovery method and device based on environment background light and storage medium | |
| CN109859141B (en) | Deep vertical shaft well wall image denoising method | |
| CN111145103A (en) | Monte Carlo denoising method based on detail retention neural network model | |
| CN111047537A (en) | System for recovering details in image denoising |
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 |