CN113781343A - Super-resolution image quality improvement method - Google Patents
Super-resolution image quality improvement method Download PDFInfo
- Publication number
- CN113781343A CN113781343A CN202111069054.XA CN202111069054A CN113781343A CN 113781343 A CN113781343 A CN 113781343A CN 202111069054 A CN202111069054 A CN 202111069054A CN 113781343 A CN113781343 A CN 113781343A
- Authority
- CN
- China
- Prior art keywords
- image
- quality
- low
- noise
- resolution
- 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
- 230000006872 improvement Effects 0.000 title claims abstract description 38
- 238000000034 method Methods 0.000 title claims abstract description 37
- 238000012549 training Methods 0.000 claims abstract description 26
- 238000012545 processing Methods 0.000 claims abstract description 23
- 238000005070 sampling Methods 0.000 claims description 28
- 238000006731 degradation reaction Methods 0.000 claims description 27
- 230000015556 catabolic process Effects 0.000 claims description 20
- 238000001914 filtration Methods 0.000 claims description 4
- 230000003321 amplification Effects 0.000 abstract description 3
- 238000003199 nucleic acid amplification method Methods 0.000 abstract description 3
- 238000004088 simulation Methods 0.000 description 7
- 230000008569 process Effects 0.000 description 6
- 238000010276 construction Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 5
- 238000000605 extraction Methods 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 230000009467 reduction Effects 0.000 description 3
- 238000013527 convolutional neural network Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 230000001133 acceleration Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 238000004321 preservation Methods 0.000 description 1
- 238000011002 quantification Methods 0.000 description 1
- 230000000007 visual effect 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
- 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
- G06T3/4007—Scaling of whole images or parts thereof, e.g. expanding or contracting based on interpolation, e.g. bilinear interpolation
-
- 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
- G06T3/4046—Scaling of whole images or parts thereof, e.g. expanding or contracting using neural networks
-
- 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
- G06T3/4053—Scaling of whole images or parts thereof, e.g. expanding or contracting based on super-resolution, i.e. the output image resolution being higher than the sensor resolution
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/73—Deblurring; Sharpening
-
- 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
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- 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)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (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 super-resolution image quality improving method, which comprises the following steps: step S1: collecting a data set, and respectively collecting a high-quality image and a low-quality image; step S2: extracting a quality-reducing kernel, constructing a quality-reducing kernel library, collecting a noise block from a low-quality image, and constructing a noise library; step S3: generating a low-quality image corresponding to the high-quality image, thereby constructing a training data set; step S4: and training the image quality improvement network with the super-resolution of any size to obtain an image quality improvement model with the super-resolution of any size. By using the method, the quality improvement of the real low-quality image can be met, the noise removal, sharpening and any resolution amplification of the low-quality image are realized, and the processing of details is not lost.
Description
Technical Field
The invention relates to the field of computer vision, deep learning and digital image processing, in particular to a super-resolution image quality improving method.
Background
The image quality improvement technology carries out a series of processing operations on the digital image, and completes the tasks of noise reduction, sharpening and resolution improvement on the image under the condition of not changing the image semantics and reserving the image details. The image quality improvement is widely applied in the fields of entertainment, medical treatment, monitoring and the like.
Conventional image quality enhancement techniques rely on various forms or combinations of filters to process the image, for example, in de-noising the image, the original image may be manipulated using a variety of low pass filtering. However, the conventional method has many limitations in improving image quality, for example, parameters need to be adjusted to adapt to different images, blurring is caused in the process of amplifying the image size, and details are lost due to amplification noise. In conclusion, the traditional method is difficult to balance the problems of image quality improvement and detail preservation.
With the rapid development of deep learning technology in the field of computer vision in recent years, deep learning methods represented by convolutional neural networks are widely researched and applied in the field of image quality improvement. A general image quality improvement network usually uses low-quality images as input of the network, the low-quality images are represented by noise, blur and low resolution, and the corresponding high-quality images are used as truth values of the network to supervise training of the network. The traditional image quality improving network usually adopts an ideal image quality reducing process to construct a training data set, for example, Gaussian noise is added to a high-quality image to construct a low-quality image containing noise, and bicubic linear interpolation is adopted to construct a low-resolution image. In a real situation, the image degradation process is various and complex, so that the conventional image quality improvement network is poor in real data performance and difficult to apply practically. And most image quality improving networks can only specify times to amplify images, such as 2 times and 4 times, due to the limitation of the convolutional neural network structure, and cannot amplify images in any size. This also limits the practical application of the image quality improvement method. Therefore, for practical application, an image quality improvement method which can realize super-resolution of any size and is obtained by simulating real image degradation process training is needed.
Therefore, those skilled in the art are dedicated to develop a super-resolution image quality improvement method, which adopts a degradation process for generating a confrontation network learning real image and extracts noise from a real noise-containing image to construct a training data set to simulate a real low-quality image; the super-resolution of any size is realized by adopting a full-connection network to simulate an image interpolation process; the acceleration of the image quality improvement processing is realized by carrying out model compression and Float16 quantification operations on the model and carrying out distributed deployment by utilizing multiple processes and shared memory.
Disclosure of Invention
In view of the above defects in the prior art, the technical problem to be solved by the present invention is to overcome the defects in the prior art, and provide an image quality improvement method which can realize super-resolution images of any size and can be obtained by training through a simulated real image degradation process.
In order to achieve the above object, the present invention provides a super-resolution image quality improving method, which comprises the following steps:
step S1: collecting a data set, and respectively collecting a high-quality image and a low-quality image, wherein the low-quality image has the characteristics of noise, blur and low resolution;
step S2: extracting a quality-reducing kernel by using a generation countermeasure network for each low-quality image by using the low-quality image data collected in the step S1, wherein the quality-reducing kernel is used for simulating blurring and downsampling, constructing a quality-reducing kernel library, and collecting a noise block from the low-quality image to construct a noise library;
step S3: generating a low-quality image corresponding to the high-quality image using the high-quality image acquired in step S1 and the degradation kernel library and the noise library constructed in step S2, thereby constructing a training data set;
step S4: and (5) training the image quality improvement network with any size super-resolution by using the training data set constructed in the step (S3) to obtain an image quality improvement model with any size super-resolution, and accelerating the trained image improvement model by using a network reasoning module.
Further, the step of extracting the degraded core includes:
step S211: the generator for generating the countermeasure network is only composed of a plurality of convolution layers, the input of the network is the low-quality image, and the output of the generator is the image after down sampling;
step S212: for the discriminator, the true value is the image block cut from the low-quality image, and the false value is the image cut from the down-sampled image output by the generator; simulating the degradation process of the image by a generator through training of each low-quality image;
s213: and obtaining a quality-reducing kernel by performing linear processing on the convolution kernel of the generator only consisting of the convolution layer, and obtaining the quality-reducing kernels with different down-sampling multiplying powers by adjusting the down-sampling multiplying power of the generator.
Further, the step of collecting noise blocks comprises:
if the noise block is extracted from the low-quality image, a low variance area in the low-quality image needs to be extracted, and the noise area is determined by comparing the pixel variance in the image block with a set variance threshold of the noise block:
σ(n_i)<μ
where σ (n _ i) is the pixel variance and μ is the variance threshold.
Further, the generating of the low-quality image corresponding to the high-quality image includes:
step S31: processing the high-quality image by using the quality-reducing kernel, performing filtering operation on the high-quality image I _ HR by using the quality-reducing kernel k randomly selected from the quality-reducing kernel library, and sampling the filtered image according to the appointed arbitrary multiplying power s to obtain a fuzzy and low-resolution image I _ D without noise:
I_D＝(I_HR*k)↓_s
step S32: adding noise to the image after the degradation kernel processing: randomly selecting a noise block n from the noise library, subtracting the mean value of the noise block, and adding the noise-free image I _ D after the noise block is subjected to the degradation kernel processing to obtain a noise-containing image I _ LR:
I_LR＝I_D+n。
furthermore, the image quality improvement network with the super-resolution of any size comprises an image feature coding module and an image up-sampling interpolation module, wherein the image feature coding module is formed by an RDN network without the up-sampling module; the image up-sampling interpolation module is realized by a full-connection network, the input of the image up-sampling interpolation module comprises the output of the image characteristic coding module and the size of the target image, and the output of the image up-sampling interpolation module is the RGB value of each pixel position in the target image.
Furthermore, the full-connection network predicts the RGB value corresponding to each pixel in the target image in a bilinear interpolation mode by using the image feature code corresponding to each pixel of the original image.
Furthermore, the image lifting model is accelerated by cutting the original image and sending the original image into the network in batches, and finally processing the original image in a mode of splicing the result.
The invention designs an image quality improving method capable of meeting super-resolution of any scale. By using the method, the quality improvement of the real low-quality image can be met, the noise removal, sharpening and any resolution amplification of the low-quality image are realized, and the processing of details is not lost. The image quality improving method which can be practically applied is really realized.
The conception, the specific structure and the technical effects of the present invention will be further described with reference to the accompanying drawings to fully understand the objects, the features and the effects of the present invention.
Drawings
FIG. 1 is a flow chart of an image quality improvement method according to a preferred embodiment of the present invention;
FIG. 2 is a block diagram of a method of the degradation simulation module according to a preferred embodiment of the present invention;
FIG. 3 is a diagram illustrating a degradation kernel obtained by a visual simulation according to a preferred embodiment of the present invention;
FIG. 4 is a diagram of noise block extraction from low quality images according to a preferred embodiment of the present invention;
fig. 5 is a block diagram of an image quality improvement network according to a preferred embodiment of the present invention.
Detailed Description
The technical contents of the preferred embodiments of the present invention will be more clearly and easily understood by referring to the drawings attached to the specification. The present invention may be embodied in many different forms of embodiments and the scope of the invention is not limited to the embodiments set forth herein.
In the drawings, structurally identical elements are represented by like reference numerals, and structurally or functionally similar elements are represented by like reference numerals throughout the several views. The size and thickness of each component shown in the drawings are arbitrarily illustrated, and the present invention is not limited to the size and thickness of each component. The thickness of the components may be exaggerated where appropriate in the figures to improve clarity.
As shown in the flow chart of the method of fig. 1, the method of the present patent includes the following steps:
step S1: and acquiring a data set, and respectively collecting a high-quality image and a low-quality image, wherein the low-quality image is represented by noise, blur, low resolution and the like. The collected high-quality images do not need to correspond one-to-one to the low-quality images.
Step S2: using the low-quality image data collected in step S1, a quality-reducing kernel is extracted using a generation countermeasure network for each low-quality image, the quality-reducing kernel is used to simulate blurring and downsampling, a quality-reducing kernel library is constructed, and noise blocks are collected from the low-quality image, and a noise library is constructed.
Step S3: and generating a low-quality image corresponding to the high-quality image by using the high-quality image acquired in the step S1 and the quality degradation kernel library and the noise library constructed in the step two, thereby constructing a training data set.
Step S4: and (5) finishing the training of the image quality improvement network with any size of super-resolution by using the training data set constructed in the step (S3) to obtain an image quality improvement model with any size of super-resolution.
Details of specific embodiments of the present invention will be described below.
The patent realizes an image quality improvement method for super-resolution of any size. As shown in the method flow diagram of fig. 1, high quality images and low quality images are first collected. And obtaining a quality degradation core library and a noise library by generating the learning of the countermeasure network on the low-quality image in a quality degradation simulation module. And the data processing module performs quality reduction processing on the high-quality image by using the quality reduction kernel library and the noise library to obtain a low-quality image corresponding to the high-quality image, and the construction of the training data set is completed. And the network training module finishes training the network by using the constructed training data. And the network reasoning module deploys by using the trained model, so that the image quality of the super-resolution image with any size is improved.
The implementation details are as follows:
1. construction and operation of a degradation simulation module:
the objective of the degradation simulation module is to extract the degradation kernel and noise from the low quality image, thereby constructing a degradation kernel library and a noise library. The function of the degradation kernel is to carry out fuzzy kernel down-sampling on a high-quality image; the effect of the noise is to add true noise to the high quality image.
A degraded nucleus extraction unit: the degraded kernel extracting unit extracts the degraded kernels from the low-quality image using the generative countermeasure network. The method is shown in figure 2. The generator for generating the countermeasure network is only composed of a plurality of convolution layers, the input of the network is a low-quality image, and the output of the generator is a downsampled image; the discriminator has a true value of an image block clipped from the low-quality image and a false value of an image clipped from the down-sampled image output from the generator. Through training of each low-quality image, the generator can simulate the image degradation process, the convolution kernel of the generator consisting of only convolution layers is subjected to linear processing to obtain a degradation kernel, and the down-sampling multiples of the generator are adjusted to obtain the degradation kernels with different down-sampling magnifications, so that two types of the double-time down-sampling kernel and the four-time down-sampling kernel can be obtained generally, as shown in fig. 3.
A noise extraction unit: noise in real images is not a single source and may be introduced during capture, storage, and post-processing. It is not reasonable to add a noise distribution conforming to a fixed probability model to the image, and therefore the noise extraction unit performs the simulation of the noise in a manner of extracting the noise from a large-scale low-quality noisy image to construct a noise library. In an image, noise is a high-frequency part, and a region with rich texture in the image also has high variance, so in order to extract a noise region from a low-quality image, a low-variance region needs to be extracted, so the noise region is determined by comparing the pixel variance σ (n _ i) in an image block with the set variance threshold μ of the noise block, as shown in the following:
σ(n_i)<μ
the extracted noise block is shown in fig. 4.
2. Construction and operation of the data processing module:
and the data processing module processes the high-quality image by using the quality-reducing kernel library and the noise library obtained by the quality-reducing simulation module to obtain the low-quality image corresponding to the high quality one by one. The treatment method comprises the following steps:
(1) processing high quality images with a degraded kernel: the nature of the degraded kernel k is a filter, the degraded kernel k randomly selected from the degraded kernel library is used for carrying out filtering operation on the high-quality image I _ HR, the filtered image is sampled according to the appointed arbitrary multiplying power s, a fuzzy and low-resolution image I _ D without noise is obtained, and the calculation is shown in the following formula;
I_D＝(I_HR*k)↓_s
(2) adding noise to the image after the degraded kernel processing: randomly selecting a noise block n from a noise library, subtracting the mean value of the noise block, and adding the noise-free image I _ D after the noise block is subjected to the degradation kernel processing to obtain a noise-containing image I _ LR, wherein the formula is as follows:
I_LR＝I_D+n
3. construction and operation of network training module
The structure of the image quality improvement network with super resolution of any scale is shown in fig. 5. The overall network structure comprises two parts. The image feature coding module E and the image up-sampling interpolation module I. The image feature coding module is composed of an RDN (remote data network) without an up-sampling module, and any network capable of finishing super-resolution of images can be used as the image feature coding module and is not limited to the RDN. The image up-sampling interpolation module is realized by a full-connection network, the output of the image up-sampling interpolation module is an RGB value of each pixel position in the target image, and the image up-sampling network is irrelevant to the size of the whole input feature map, but normalizes the target size image and the low-resolution image to the same size, and takes the feature of each point coordinate in the target size image and the nearest coordinate in the low-resolution image as the input for predicting the pixel value of the point, so that the interpolation of any size can be realized, and the part is also the key for realizing the super-resolution of any size. The input to the image upsampling interpolation module includes the output F of the image feature encoding module and the size of the target image. The full-connection network predicts the RGB value corresponding to each pixel in the target image by using the image characteristic coding corresponding to each pixel of the original image through a bilinear interpolation mode. In order to speed up network training and reduce the occupation of video memory, the input of the network is a block cut from an original picture. The loss function of the network training adopts an L1 loss function.
4. Construction and operation of network reasoning module
The network reasoning module uses TensorRt to accelerate the trained model. In order to reduce the video memory occupation, the original image is cut and sent to the network in batches, and finally the result is processed in a splicing mode. Network inference and picture processing are computed based on the Nvidia GPU.
The foregoing detailed description of the preferred embodiments of the invention has been presented. It should be understood that numerous modifications and variations could be devised by those skilled in the art in light of the present teachings without departing from the inventive concepts. Therefore, the technical solutions available to those skilled in the art through logic analysis, reasoning and limited experiments based on the prior art according to the concept of the present invention should be within the scope of protection defined by the claims.
Claims (7)
1. A super-resolution image quality improving method is characterized by comprising the following steps:
step S1: collecting a data set, and respectively collecting a high-quality image and a low-quality image, wherein the low-quality image has the characteristics of noise, blur and low resolution;
step S2: extracting a quality-reducing kernel by using a generation countermeasure network for each low-quality image by using the low-quality image data collected in the step S1, wherein the quality-reducing kernel is used for simulating blurring and downsampling, constructing a quality-reducing kernel library, and collecting a noise block from the low-quality image to construct a noise library;
step S3: generating a low-quality image corresponding to the high-quality image using the high-quality image acquired in step S1 and the degradation kernel library and the noise library constructed in step S2, thereby constructing a training data set;
step S4: and (5) training the image quality improvement network with any size super-resolution by using the training data set constructed in the step (S3) to obtain an image quality improvement model with any size super-resolution, and accelerating the trained image improvement model by using a network reasoning module.
2. The super-resolution image quality improvement method according to claim 1, wherein the step of extracting the degradation kernel includes:
step S211: the generator for generating the countermeasure network is only composed of a plurality of convolution layers, the input of the network is the low-quality image, and the output of the generator is the image after down sampling;
step S212: for the discriminator, the true value is the image block cut from the low-quality image, and the false value is the image cut from the down-sampled image output by the generator; simulating the degradation process of the image by a generator through training of each low-quality image;
s213: and obtaining a quality-reducing kernel by performing linear processing on the convolution kernel of the generator only consisting of the convolution layer, and obtaining the quality-reducing kernels with different down-sampling multiplying powers by adjusting the down-sampling multiplying power of the generator.
3. The super-resolution image quality improvement method according to claim 1, wherein the step of acquiring a noise block comprises:
if the noise block is extracted from the low-quality image, a low variance area in the low-quality image needs to be extracted, and the noise area is determined by comparing the pixel variance in the image block with a set variance threshold of the noise block:
σ(n_i)<μ
where σ (n _ i) is the pixel variance and μ is the variance threshold.
4. The super-resolution image quality improvement method according to claim 1, wherein the generating of the low-quality image corresponding to the high-quality image comprises:
step S31: processing the high-quality image by using the quality-reducing kernel, performing filtering operation on the high-quality image I _ HR by using the quality-reducing kernel k randomly selected from the quality-reducing kernel library, and sampling the filtered image according to the appointed arbitrary multiplying power s to obtain a fuzzy and low-resolution image I _ D without noise:
I_D＝(I_HR*k)↓_s
step S32: adding noise to the image after the degradation kernel processing: randomly selecting a noise block n from the noise library, subtracting the mean value of the noise block, and adding the noise-free image I _ D after the noise block is subjected to the degradation kernel processing to obtain a noise-containing image I _ LR:
I_LR＝I_D+n。
5. the super-resolution image quality improvement method according to claim 1, wherein the image quality improvement network for arbitrary size super-resolution comprises an image feature coding module and an image up-sampling interpolation module, wherein the image feature coding module is formed by an RDN network without the up-sampling module; the image up-sampling interpolation module is realized by a full-connection network, the input of the image up-sampling interpolation module comprises the output of the image characteristic coding module and the size of the target image, and the output of the image up-sampling interpolation module is the RGB value of each pixel position in the target image.
6. The super-resolution image quality improving method according to claim 5, wherein the full-connection network predicts the RGB value corresponding to each pixel in the target image by bilinear interpolation using image feature coding corresponding to each pixel of the original image.
7. The super-resolution image quality improvement method according to claim 1, wherein the image improvement model is accelerated by cutting the original image into the network in batches, and finally splicing the results.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111069054.XA CN113781343A (en) | 2021-09-13 | 2021-09-13 | Super-resolution image quality improvement method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111069054.XA CN113781343A (en) | 2021-09-13 | 2021-09-13 | Super-resolution image quality improvement method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113781343A true CN113781343A (en) | 2021-12-10 |
Family
ID=78843050
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111069054.XA Pending CN113781343A (en) | 2021-09-13 | 2021-09-13 | Super-resolution image quality improvement method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113781343A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114092337A (en) * | 2022-01-19 | 2022-02-25 | 苏州浪潮智能科技有限公司 | Method and device for super-resolution amplification of image at any scale |
| CN116342393A (en) * | 2023-04-11 | 2023-06-27 | 广州极点三维信息科技有限公司 | Image super-resolution method and system based on image noise prediction mechanism |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110223231A (en) * | 2019-06-06 | 2019-09-10 | 天津工业大学 | A kind of rapid super-resolution algorithm for reconstructing of noisy image |
| CN110276721A (en) * | 2019-04-28 | 2019-09-24 | 天津大学 | Image super-resolution rebuilding method based on cascade residual error convolutional neural networks |
| CN110363716A (en) * | 2019-06-25 | 2019-10-22 | 北京工业大学 | One kind is generated based on condition and fights network combined degraded image high quality method for reconstructing |
| CN110599499A (en) * | 2019-08-22 | 2019-12-20 | 四川大学 | MRI image heart structure segmentation method based on multipath convolutional neural network |
| CN110634108A (en) * | 2019-08-30 | 2019-12-31 | 北京工业大学 | Composite degraded live webcast video enhancement method based on element-cycle consistency countermeasure network |
| CN112365551A (en) * | 2020-10-15 | 2021-02-12 | 上海市精神卫生中心（上海市心理咨询培训中心） | Image quality processing system, method, device and medium |
| CN112435309A (en) * | 2020-12-07 | 2021-03-02 | 苏州深透智能科技有限公司 | Method for enhancing quality and resolution of CT image based on deep learning |
| WO2021042774A1 (en) * | 2019-09-04 | 2021-03-11 | 华为技术有限公司 | Image recovery method, image recovery network training method, device, and storage medium |
| CN112598579A (en) * | 2020-12-28 | 2021-04-02 | 苏州科达特种视讯有限公司 | Image super-resolution method and device for monitoring scene and storage medium |
| CN113160079A (en) * | 2021-04-13 | 2021-07-23 | Oppo广东移动通信有限公司 | Portrait restoration model training method, portrait restoration method and device |
| CN113379601A (en) * | 2021-06-08 | 2021-09-10 | 中国科学技术大学 | Real world image super-resolution method and system based on degradation variational self-encoder |
-
2021
- 2021-09-13 CN CN202111069054.XA patent/CN113781343A/en active Pending
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110276721A (en) * | 2019-04-28 | 2019-09-24 | 天津大学 | Image super-resolution rebuilding method based on cascade residual error convolutional neural networks |
| CN110223231A (en) * | 2019-06-06 | 2019-09-10 | 天津工业大学 | A kind of rapid super-resolution algorithm for reconstructing of noisy image |
| CN110363716A (en) * | 2019-06-25 | 2019-10-22 | 北京工业大学 | One kind is generated based on condition and fights network combined degraded image high quality method for reconstructing |
| CN110599499A (en) * | 2019-08-22 | 2019-12-20 | 四川大学 | MRI image heart structure segmentation method based on multipath convolutional neural network |
| CN110634108A (en) * | 2019-08-30 | 2019-12-31 | 北京工业大学 | Composite degraded live webcast video enhancement method based on element-cycle consistency countermeasure network |
| WO2021042774A1 (en) * | 2019-09-04 | 2021-03-11 | 华为技术有限公司 | Image recovery method, image recovery network training method, device, and storage medium |
| CN112365551A (en) * | 2020-10-15 | 2021-02-12 | 上海市精神卫生中心（上海市心理咨询培训中心） | Image quality processing system, method, device and medium |
| CN112435309A (en) * | 2020-12-07 | 2021-03-02 | 苏州深透智能科技有限公司 | Method for enhancing quality and resolution of CT image based on deep learning |
| CN112598579A (en) * | 2020-12-28 | 2021-04-02 | 苏州科达特种视讯有限公司 | Image super-resolution method and device for monitoring scene and storage medium |
| CN113160079A (en) * | 2021-04-13 | 2021-07-23 | Oppo广东移动通信有限公司 | Portrait restoration model training method, portrait restoration method and device |
| CN113379601A (en) * | 2021-06-08 | 2021-09-10 | 中国科学技术大学 | Real world image super-resolution method and system based on degradation variational self-encoder |
Non-Patent Citations (1)
| Title |
|---|
| YUNHE LI ET AL.: "Super-Resolution of Sentinel-2 Images at 10m Resolution without Reference Images", PREPRINTS 2021, 20 April 2021 (2021-04-20), pages 1 - 22 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114092337A (en) * | 2022-01-19 | 2022-02-25 | 苏州浪潮智能科技有限公司 | Method and device for super-resolution amplification of image at any scale |
| CN116342393A (en) * | 2023-04-11 | 2023-06-27 | 广州极点三维信息科技有限公司 | Image super-resolution method and system based on image noise prediction mechanism |
| CN116342393B (en) * | 2023-04-11 | 2023-09-26 | 广州极点三维信息科技有限公司 | Image super-resolution method and system based on image noise prediction mechanism |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112507997B (en) | Face super-resolution system based on multi-scale convolution and receptive field feature fusion | |
| CN110136063A (en) | A kind of single image super resolution ratio reconstruction method generating confrontation network based on condition | |
| CN111028163A (en) | Convolution neural network-based combined image denoising and weak light enhancement method | |
| CN112257766B (en) | Shadow recognition detection method in natural scene based on frequency domain filtering processing | |
| CN113284051B (en) | Face super-resolution method based on frequency decomposition multi-attention machine system | |
| CN116051428B (en) | Deep learning-based combined denoising and superdivision low-illumination image enhancement method | |
| CN113781343A (en) | Super-resolution image quality improvement method | |
| CN111861894A (en) | Image motion blur removing method based on generating type countermeasure network | |
| CN112801906B (en) | Cyclic iterative image denoising method based on cyclic neural network | |
| CN112164011A (en) | Motion image deblurring method based on self-adaptive residual error and recursive cross attention | |
| CN113344793A (en) | Image super-resolution reconstruction method, device, equipment and storage medium | |
| CN114723630A (en) | Image deblurring method and system based on cavity double-residual multi-scale depth network | |
| CN117575915B (en) | Image super-resolution reconstruction method, terminal equipment and storage medium | |
| CN116188272B (en) | Two-stage depth network image super-resolution reconstruction method suitable for multiple fuzzy cores | |
| CN115393191A (en) | Method, device and equipment for reconstructing super-resolution of lightweight remote sensing image | |
| CN115797176A (en) | Image super-resolution reconstruction method | |
| CN118742918A (en) | Method and system for high frequency attention network for high efficiency single image super resolution | |
| CN115131203A (en) | LR image generation method and real image super-resolution method based on uncertainty | |
| CN112927137A (en) | Method, device and storage medium for acquiring blind super-resolution image | |
| CN109993701B (en) | Depth map super-resolution reconstruction method based on pyramid structure | |
| CN114202460B (en) | Super-resolution high-definition reconstruction method, system and equipment for different damage images | |
| CN113096032B (en) | Non-uniform blurring removal method based on image region division | |
| GB2623387A (en) | Learnable image transformation training methods and systems in graphics rendering | |
| Van Noord et al. | Light-weight pixel context encoders for image inpainting | |
| CN112801909B (en) | Image fusion denoising method and system based on U-Net and pyramid module |
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 |