CN110634103A - Image demosaicing method based on generation of countermeasure network - Google Patents
Image demosaicing method based on generation of countermeasure network Download PDFInfo
- Publication number
- CN110634103A CN110634103A CN201910804744.1A CN201910804744A CN110634103A CN 110634103 A CN110634103 A CN 110634103A CN 201910804744 A CN201910804744 A CN 201910804744A CN 110634103 A CN110634103 A CN 110634103A
- Authority
- CN
- China
- Prior art keywords
- image
- network
- training
- generator
- mosaic
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 24
- 238000012549 training Methods 0.000 claims abstract description 56
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 10
- 238000012545 processing Methods 0.000 claims abstract description 9
- 238000012360 testing method Methods 0.000 claims description 16
- 238000010276 construction Methods 0.000 claims description 6
- 230000000694 effects Effects 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000000605 extraction Methods 0.000 claims description 4
- 230000003044 adaptive effect Effects 0.000 claims description 3
- 230000003042 antagnostic effect Effects 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 238000003672 processing method Methods 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 2
- 238000013135 deep learning Methods 0.000 abstract description 10
- 230000000007 visual effect Effects 0.000 abstract description 4
- 238000005516 engineering process Methods 0.000 description 4
- 238000001514 detection method Methods 0.000 description 3
- 230000003287 optical effect Effects 0.000 description 3
- 238000013461 design Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 239000000969 carrier Substances 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000006855 networking Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000003786 synthesis reaction Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
Images
Classifications
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4015—Image demosaicing, e.g. colour filter arrays [CFA] or Bayer patterns
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to the field of image processing, deep learning and convolutional neural networks, and aims to remove mosaics of digital images and improve the visual quality of the digital images by using a generated countermeasure network in a deep learning algorithm. To this end, the invention relates to an image demosaicing method based on a generation countermeasure network, comprising the following steps: (1) constructing a data set; (2) adding a mosaic; (3) constructing and generating a confrontation network model; (4) training to generate an confrontation network; (5) and processing the image by using the trained network. The invention is mainly applied to the image processing occasion.
Description
Technical Field
The invention belongs to the field of image processing, deep learning and convolutional neural networks, and relates to detection, identification and other related applications based on natural image recovery. And more particularly to image demosaicing methods based on generation of a competing network.
Background
Today, with the rapid development of information technology, digital images have become one of the most popular information carriers in social networking and technical applications. The optical sensor has the advantages of low price, excellent performance, small volume and low power consumption. These unique advantages make optical sensors convenient to place in a variety of environments, and the large number of digital images collected can support a wide variety of applications. However, due to the fact that the quality of the civil-grade optical sensor is uneven, mosaic in the digital image often appears due to the loss of details; on the other hand, the internet technology is rapidly developed and personal information is increasingly sensitive, and although a large number of digital images containing potential information exist in the internet, the images are low in quality because users often use mosaics to cover useful information in photos, and a series of image-based applications cannot be supported. Therefore, the mosaic of the digital image is removed, the visual quality of the image is improved, and clear and detailed images are obtained, which is particularly important for various applications.
The prior image demosaicing algorithm is usually based on a probability prior model of an image or based on a filtering method to process the image in various transformation domains so as to realize mosaic removal and image restoration. However, the modeling mode of the traditional algorithm is not perfect, the mosaic removal effect is not high, and a part of high-frequency information of the image is lost while the mosaic is removed, so that the image is blurred in a large area.
The deep learning technology has been developed rapidly in recent years, and significant achievements are achieved in image recognition, target detection and voice processing. The generation of a countermeasure network (GAN) is a new type of network in deep learning algorithm, and its design inspiration comes from binary zero and game process. Generation of countermeasure Network two parts are constructed using Convolutional Neural Networks (CNN): and the generator and the discriminator are used for carrying out antagonistic training so as to learn the distribution of the data and realize an image conversion or synthesis task.
[1]Isola P,Zhu J Y,Zhou T,et al.Image-to-Image Translation with Conditional Adversarial Networks[J].2016.
[2]Zhou R,Achanta R,Süsstrunk S.Deep Residual Network for Joint Demosaicing and Super-Resolution[J].arXiv preprint arXiv:1802.06573,2018.
[3]Goodfellow I,Pouget-Abadie J,Mirza M,et al.Generative adversarial nets[C]//Advances in neural information processing systems.2014:2672-2680。
Disclosure of Invention
In order to overcome the defects of the prior art, the invention aims to remove the mosaic of the digital image by using the generated countermeasure network in the deep learning algorithm and improve the visual quality of the digital image. Therefore, the technical scheme adopted by the invention is that the image demosaicing method based on the generation of the countermeasure network comprises the following steps:
(1) and (3) construction of a data set: selecting an ImageNet data set as training for generating a countermeasure network, randomly selecting m types of images in the ImageNet data set, randomly selecting n images in each type as training data, and selecting images except the n images in each type as test data to construct a data set;
(2) adding a mosaic: adding mosaics at random positions on 10000 training images by utilizing Matlab, and randomly selecting the sizes and the degrees of the mosaics;
(3) construction of a generation countermeasure network model: the generation of the countermeasure network consists of two parts: a generator and a discriminator; respectively forming a generator and a discriminator by using a convolutional neural network, constructing the generator on the basis of AlexNet, designing the first two layers of networks in the generator into convolutional layers of convolutional kernels with sizes of 4 multiplied by 4 and 2 multiplied by 2, and carrying out convolution by a step length consistent with the size of the convolutional kernels, thereby realizing down-sampling of a feature map, compressing the spatial resolution of the feature map and facilitating feature extraction, in addition, removing all connecting layers in the AlexNet, adding a deconvolution layer to the networks to form a full convolutional network, and keeping the sizes of an input image and an output image unchanged;
(4) training to generate an antagonistic network: inputting the constructed training data set into a generated countermeasure network, for training data, randomly selecting a batch of images for learning by the network each time, and alternately training a discriminator and a generator respectively, wherein the training process adopts a random gradient descent (SGD) method and an adaptive motion optimizer to optimize.
The method also comprises the following steps of: for a test data set, adding a mosaic by using the same processing method as training data, then independently taking out a generator in the trained GAN as a demosaicer, inputting the test data containing the mosaic into the generator, considering the output image as an image after demosaicing, and measuring the demosaicing effect by calculating the peak power Signal-to-noise ratio (PSNR) (Peak Signal to noise ratio) between the output image after demosaicing and an original image without the mosaic.
In one example, specifically:
training the discriminator:
when the discrimination network is trained, inputting 32 images into the network as a batch, connecting each image containing mosaic with an original image without mosaic into a six-channel image, and adding a label of '0'; inputting the image containing the mosaic into a generator to obtain a demosaiced output image, connecting the image containing the mosaic and the demosaiced image obtained by the processing of the generator into a six-channel image, and adding a label of '1'; the six-channel image is input into a discriminator to obtain discrimination output, loss is calculated by utilizing binary mutual entropy loss, and the definition formula of the binary mutual entropy loss is as follows:
in the formula oiActual labels, t, representing the second pair of imagesiRepresenting the output discrimination label of the ith image pair, and carrying out reverse back BP (Back propagation) on the calculated average value of the loss value to realize the training of the network;
training the generator:
in training for generating a network, firstly, an image is selected from a batch of read images containing mosaics, then the images are input into a generator, a demosaiced image is output, and the output image and the image without mosaics are subjected to two-norm loss calculation, which is defined as follows:
wherein xiIs an original image without mosaic, yiIs a demosaiced image;
training the generated network:
in the training process of the generation network, firstly, inputting a read-in image containing noise into the generation network, outputting a denoised image, and performing one-norm loss on the output denoised image and an image without noise in the image pair, wherein the definition is as follows:
then connecting the image containing mosaic and the demosaiced image into a six-channel image to be input into a discriminator to obtain an output label, and performing binary lake loss on the output label and the label '1' to obtain LEAnd weighting and summing the two loss values to obtain the final generator loss:
LG＝LE+λL1
in the above equation λ is set to 100, resulting in a generator loss LGAnd (4) returning by using a BP algorithm to realize the training of the generator.
The invention has the characteristics and beneficial effects that:
the image demosaicing technology based on the GAN realizes feature extraction by utilizing the nonlinear modeling of the convolutional neural network, generates redundant high-frequency information, and fills the high-frequency and low-frequency information while removing mosaic. The invention relates to the interdisciplinary field of image processing and deep learning, and has great research value and significance.
Description of the drawings:
FIG. 1. generating an architecture diagram of a countermeasure network.
Detailed Description
The method and the device realize mosaic removal of the digital image by using the generation countermeasure network in the deep learning algorithm, and improve the visual quality of the digital image. The following is a brief description of the steps of carrying out the invention:
1. and constructing a data set and adding a mosaic. The invention selects the ImageNet dataset as the training for generating the countermeasure network. Randomly selecting 100 kinds of images in the ImageNet data set, randomly selecting 100 images in each kind as training data, and using 20 images as test data to construct the data set. And adding mosaics at random positions on 10000 training images by utilizing Matlab, and randomly selecting the sizes and the degrees of the mosaics.
2. And generating construction of the confrontation network model. The generation of the countermeasure network consists of two parts: a generator and a discriminator. The invention utilizes a convolutional neural network to respectively form two parts. A generator is constructed on the basis of classical AlexNet by researching a classical convolutional neural network structure. The first two layers of the network in the generator are designed as convolutional layers of convolutional kernels of sizes 4 x 4 and 2 x 2 and are convolved with step sizes consistent with the sizes of the convolutional kernels, thereby realizing downsampling of the feature map, compressing the spatial resolution of the feature map and facilitating feature extraction. In addition, the invention removes the full connection layer in AlexNet and adds the deconvolution layer to the network to form a full convolution network, and the sizes of the input image and the output image are kept unchanged. The discriminator is composed of a convolution layer of five layers of 4 x 4 convolution kernels.
3. Training against the network is generated. And inputting the constructed training data set into a generation countermeasure network. For training data, the network randomly selects a batch of images for learning each time, and alternately trains the discriminator and the generator respectively. The training process uses a Stochastic Gradient Descent (SGD) method, and an adaptive moment estimation (Adam) optimizer for optimization. The network reaches convergence over 200 training periods.
4. And testing the network performance. For the test data set, a mosaic is added by using the same processing method as the training data, and then the generators in the trained GAN are taken out separately to be used as a demosaic device. The test data containing the mosaic is input into the generator and the output image is considered to be a demosaiced image. The demosaicing effect is measured by calculating the Peak Signal to Noise Ratio (PSNR) between the demosaiced output image and the original image without mosaic. The higher the PSNR value, the better the demosaicing effect.
In order to further clarify the present invention, each of the implementation steps of the present invention will be described in detail:
1. and constructing a data set and adding a mosaic.
The data set employed by the present invention is from the ImageNet data set, which contains over 20000 categories of 1400 or more than ten thousand pictures. The ImageNet data set is widely applied to the fields related to computer vision such as image classification, target detection and the like, plays a great role in promoting the development of machine learning and deep learning algorithms, and becomes a standard data set in computer vision.
The method randomly selects 100 kinds of images from the ImageNet data set, wherein 100 images are randomly selected for each kind to serve as a training set, and 20 images are selected for each kind to serve as a testing set. That is, the training set contains 10000 images and the test set contains 2000 images. The image is resized to 256 x 256 pixels using a Matlab or Python program and a mosaic of random degree and size is added at random locations in the image.
2. And generating the construction of the countermeasure network.
The generation of the countermeasure network is a novel network architecture in the deep learning algorithm in recent years, and the design inspiration of the network architecture is derived from binary zero sum game. The generation of a countermeasure network comprises two parts: a discriminator and a generator, both parts often constructed using convolutional neural networks. The input of the generator is an image containing mosaic, and the purpose of the generator is to remove the mosaic in the image and output a visually clear image. The input of the discriminator is a six-channel image, the six-channel image is formed by connecting two images in series, the two images can be an image containing mosaic and an original image without mosaic, and can also be an image containing mosaic and a demosaiced image processed by the generator. The output of the discriminator is the discrimination of the image: the '0' representative image is an original image without mosaic, and the '1' representative image is a demosaiced image processed by the generator. The purpose of the discriminator is to discriminate the image as accurately as possible. The network is generated to generate a realistic image as much as possible to fool the discriminator. The architecture for generating a countermeasure network is shown in fig. 1.
The invention constructs a generator in the GAN based on AlexNet. The convolution kernel sizes of the first two convolution layers of AlexNet are changed into 4 x 4 and 2 x 2, and the full connection layers in the network are deleted and replaced by the deconvolution layers to construct a symmetrical full convolution network.
3. Training against the network is generated.
In order to learn the mapping between the mosaic-containing image and the mosaic-free original image, the generation of the countermeasure network needs to be sufficiently trained by using the constructed training data set, and the training process is as follows:
(1) and training the discriminator.
When training the discrimination network, 32 images are input to the network as a batch. Connecting each image containing mosaic and original image without mosaic into a six-channel image, and adding a label of '0'; inputting the image containing mosaic into a generator to obtain a demosaiced output image, connecting the image containing mosaic and the demosaiced image obtained by the processing of the generator into a six-channel image, and adding a label of '1'. And inputting the six-channel image into a discriminator to obtain discrimination output, and calculating loss by using binary mutual entropy loss. The definition formula of the binary mutual entropy loss is as follows:
in the formula oiActual labels, t, representing the second pair of imagesiAn output representative of the ith image pair identifies the label. And (4) carrying out Backward Propagation (BP) on the calculated average value of the loss values to realize the training of the network.
(2) The generator is trained.
In training for generating a network, an image is selected from a batch of images containing mosaics, and the images are input to a generator to output a demosaiced image. And performing two-norm loss calculation on the output image and the image without mosaic, wherein the two-norm loss calculation is defined as follows:
(3) and training the generated network.
In the training process of the generation network, firstly, the read-in radar image containing noise is input into the generation network, and a denoised radar image is output. And performing a norm loss on the output denoised radar image and the radar image without noise in the image pair, wherein the norm loss is defined as follows:
wherein xiIs an original image without mosaic, yiIs a demosaiced image. Then connecting the image containing mosaic and the demosaiced image into a six-channel image to be input into a discriminator to obtain an output label, and performing binary lake loss on the output label and the label '1' to obtain LEAnd weighting and summing the two loss values to obtain the final generator loss:
LG＝LE+λL1
in the above equation λ is set to 100, resulting in a generator loss LGAnd (4) returning by using a BP algorithm to realize the training of the generator.
4. And testing the network performance.
2000 images in the test data set were similarly resized to 256 x 256 pixels and randomly added to the mosaic in the same manner as the training set was processed. And (4) taking out the generators in the trained generation countermeasure network individually as demosaic devices. The test data containing the mosaic is input into a demosaicer, and the resulting image is a demosaiced image. The evaluation of the demosaicing performance can be achieved by calculating the peak power signal-to-noise ratio between the original image without mosaic and the demosaiced image. PSNR is defined as follows:
where m × n × c is the size of the image, 256 × 256 × 3 in the present invention; x is the original image without mosaic, y is the demosaiced image, MAXIIs the pixel maximum, which is 255.
Claims (3)
1. An image demosaicing method based on a generation countermeasure network is characterized by comprising the following steps:
(1) and (3) construction of a data set: selecting an ImageNet data set as training for generating a countermeasure network, randomly selecting m types of images in the ImageNet data set, randomly selecting n images in each type as training data, and selecting images except the n images in each type as test data to construct a data set;
(2) adding a mosaic: adding mosaics at random positions on 10000 training images by utilizing Matlab, and randomly selecting the sizes and the degrees of the mosaics;
(3) construction of a generation countermeasure network model: the generation of the countermeasure network consists of two parts: a generator and a discriminator; respectively forming a generator and a discriminator by using a convolutional neural network, constructing the generator on the basis of AlexNet, designing the first two layers of networks in the generator into convolutional layers of convolutional kernels with sizes of 4 multiplied by 4 and 2 multiplied by 2, and carrying out convolution by a step length consistent with the size of the convolutional kernels, thereby realizing down-sampling of a feature map, compressing the spatial resolution of the feature map and facilitating feature extraction, in addition, removing all connecting layers in the AlexNet, adding a deconvolution layer to the networks to form a full convolutional network, and keeping the sizes of an input image and an output image unchanged;
(4) training to generate an antagonistic network: inputting the constructed training data set into a generated countermeasure network, for training data, randomly selecting a batch of images for learning by the network each time, and alternately training a discriminator and a generator respectively, wherein the training process adopts a random gradient descent (SGD) method and an adaptive optimization device for optimization.
2. The image demosaicing method based on generation of competing network as claimed in claim 1, further comprising the step of testing the performance of the network: for a test data set, adding a mosaic by using the same processing method as training data, then independently taking out a generator in the trained GAN as a demosaicer, inputting the test data containing the mosaic into the generator, considering the output image as an image after demosaicing, and measuring the demosaicing effect by calculating the peak power Signal-to-Noise ratio (PSNR) (Peak Signal to Noise ratio) between the output image after demosaicing and an original image without the mosaic.
3. The image demosaicing method based on generative countermeasure network as claimed in claim 1, wherein the detailed steps are detailed as follows:
training the discriminator:
when the discrimination network is trained, inputting 32 images into the network as a batch, connecting each image containing mosaic with an original image without mosaic into a six-channel image, and adding a label of '0'; inputting the image containing the mosaic into a generator to obtain a demosaiced output image, connecting the image containing the mosaic and the demosaiced image obtained by the processing of the generator into a six-channel image, and adding a label of '1'; the six-channel image is input into a discriminator to obtain discrimination output, loss is calculated by utilizing binary mutual entropy loss, and the definition formula of the binary mutual entropy loss is as follows:
in the formula oiRepresenting the reality of the second image pairLabel, tiRepresenting the output discrimination label of the ith image pair, and carrying out reverse back BP (Back propagation) on the calculated average value of the loss value to realize the training of the network;
training the generator:
in training for generating a network, firstly, an image is selected from a batch of read images containing mosaics, then the images are input into a generator, a demosaiced image is output, and the output image and the image without mosaics are subjected to two-norm loss calculation, which is defined as follows:
wherein xiIs an original image without mosaic, yiIs a demosaiced image;
training the generated network:
in the training process of the generation network, firstly, inputting a read-in image containing noise into the generation network, outputting a denoised image, and performing one-norm loss on the output denoised image and an image without noise in the image pair, wherein the definition is as follows:
then connecting the image containing mosaic and the demosaiced image into a six-channel image to be input into a discriminator to obtain an output label, and performing binary lake loss on the output label and the label '1' to obtain LEAnd weighting and summing the two loss values to obtain the final generator loss:
LG＝LE+λL1
in the above equation λ is set to 100, resulting in a generator loss LGAnd (4) returning by using a BP algorithm to realize the training of the generator.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910804744.1A CN110634103A (en) | 2019-08-28 | 2019-08-28 | Image demosaicing method based on generation of countermeasure network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910804744.1A CN110634103A (en) | 2019-08-28 | 2019-08-28 | Image demosaicing method based on generation of countermeasure network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110634103A true CN110634103A (en) | 2019-12-31 |
Family
ID=68969493
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910804744.1A Pending CN110634103A (en) | 2019-08-28 | 2019-08-28 | Image demosaicing method based on generation of countermeasure network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110634103A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112508929A (en) * | 2020-12-16 | 2021-03-16 | 奥比中光科技集团股份有限公司 | Method and device for training generation of confrontation network |
| CN113095219A (en) * | 2021-04-12 | 2021-07-09 | 中国工商银行股份有限公司 | Reticulate pattern face recognition method and device |
| CN113824945A (en) * | 2021-11-22 | 2021-12-21 | 深圳深知未来智能有限公司 | Rapid automatic white balance and color correction method based on deep learning |
| CN113989115A (en) * | 2021-10-21 | 2022-01-28 | 河北师范大学 | Global mosaic removal method based on contrast learning |
| CN114528920A (en) * | 2022-01-19 | 2022-05-24 | 西北大学 | Terracotta soldiers and horses color restoration method based on generation of confrontation network |
| CN114972073A (en) * | 2022-04-24 | 2022-08-30 | 武汉大学 | Image demosaicing method for generating confrontation network SRGAN based on super-resolution |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105993169A (en) * | 2014-09-15 | 2016-10-05 | 深圳市大疆创新科技有限公司 | System and method for image demosaicing |
| CN108492258A (en) * | 2018-01-17 | 2018-09-04 | 天津大学 | A kind of radar image denoising method based on generation confrontation network |
| WO2019057944A1 (en) * | 2017-09-22 | 2019-03-28 | F. Hoffmann-La Roche Ag | Artifacts removal from tissue images |
-
2019
- 2019-08-28 CN CN201910804744.1A patent/CN110634103A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105993169A (en) * | 2014-09-15 | 2016-10-05 | 深圳市大疆创新科技有限公司 | System and method for image demosaicing |
| WO2019057944A1 (en) * | 2017-09-22 | 2019-03-28 | F. Hoffmann-La Roche Ag | Artifacts removal from tissue images |
| CN108492258A (en) * | 2018-01-17 | 2018-09-04 | 天津大学 | A kind of radar image denoising method based on generation confrontation network |
Non-Patent Citations (1)
| Title |
|---|
| 石光明等: "压缩感知理论的工程应用方法", 西安电子科技大学出版社, pages: 63 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112508929A (en) * | 2020-12-16 | 2021-03-16 | 奥比中光科技集团股份有限公司 | Method and device for training generation of confrontation network |
| CN113095219A (en) * | 2021-04-12 | 2021-07-09 | 中国工商银行股份有限公司 | Reticulate pattern face recognition method and device |
| CN113989115A (en) * | 2021-10-21 | 2022-01-28 | 河北师范大学 | Global mosaic removal method based on contrast learning |
| CN113989115B (en) * | 2021-10-21 | 2024-04-02 | 河北师范大学 | Global mosaic removing method based on contrast learning |
| CN113824945A (en) * | 2021-11-22 | 2021-12-21 | 深圳深知未来智能有限公司 | Rapid automatic white balance and color correction method based on deep learning |
| CN114528920A (en) * | 2022-01-19 | 2022-05-24 | 西北大学 | Terracotta soldiers and horses color restoration method based on generation of confrontation network |
| CN114528920B (en) * | 2022-01-19 | 2023-08-18 | 西北大学 | Terra-cotta warriors color restoration method based on generation countermeasure network |
| CN114972073A (en) * | 2022-04-24 | 2022-08-30 | 武汉大学 | Image demosaicing method for generating confrontation network SRGAN based on super-resolution |
| CN114972073B (en) * | 2022-04-24 | 2024-04-30 | 武汉大学 | Image demosaicing method for generating countermeasure network SRGAN based on super resolution |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110969577B (en) | Video super-resolution reconstruction method based on deep double attention network | |
| CN110634103A (en) | Image demosaicing method based on generation of countermeasure network | |
| WO2022267641A1 (en) | Image defogging method and system based on cyclic generative adversarial network | |
| Dong et al. | Deep spatial–spectral representation learning for hyperspectral image denoising | |
| CN109064396B (en) | Single image super-resolution reconstruction method based on deep component learning network | |
| CN107154023B (en) | Based on the face super-resolution reconstruction method for generating confrontation network and sub-pix convolution | |
| Liu et al. | Robust video super-resolution with learned temporal dynamics | |
| CN109523470B (en) | Depth image super-resolution reconstruction method and system | |
| WO2021048607A1 (en) | Motion deblurring using neural network architectures | |
| CN107123089A (en) | Remote sensing images super-resolution reconstruction method and system based on depth convolutional network | |
| CN109146788A (en) | Super-resolution image reconstruction method and device based on deep learning | |
| CN111445418A (en) | Image defogging method and device and computer equipment | |
| CN114746895A (en) | Noise reconstruction for image denoising | |
| CN113222825B (en) | Infrared image super-resolution reconstruction method based on visible light image training and application | |
| CN103455988A (en) | Super-resolution image reconstruction method based on structure self-similarity and sparse representation | |
| CN113379606B (en) | Face super-resolution method based on pre-training generation model | |
| CN110809126A (en) | Video frame interpolation method and system based on adaptive deformable convolution | |
| CN113674154B (en) | Single image super-resolution reconstruction method and system based on generation countermeasure network | |
| CN117173025A (en) | Single-frame image super-resolution method and system based on cross-layer mixed attention transducer | |
| CN115713462A (en) | Super-resolution model training method, image recognition method, device and equipment | |
| CN108550111B (en) | Residual error example regression super-resolution reconstruction method based on multi-level dictionary learning | |
| Zhang et al. | Enhanced visual perception for underwater images based on multistage generative adversarial network | |
| Xiao et al. | Effective PRNU extraction via densely connected hierarchical network | |
| CN113538402A (en) | Crowd counting method and system based on density estimation | |
| CN114821239B (en) | Method for detecting plant diseases and insect pests in foggy environment |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191231 |