CN112150356A - Single compressed image super-resolution reconstruction method based on cascade framework - Google Patents
Single compressed image super-resolution reconstruction method based on cascade framework Download PDFInfo
- Publication number
- CN112150356A CN112150356A CN201910571747.5A CN201910571747A CN112150356A CN 112150356 A CN112150356 A CN 112150356A CN 201910571747 A CN201910571747 A CN 201910571747A CN 112150356 A CN112150356 A CN 112150356A
- Authority
- CN
- China
- Prior art keywords
- resolution
- network
- image
- super
- reconstruction
- 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 74
- 230000000694 effects Effects 0.000 claims abstract description 30
- 230000006837 decompression Effects 0.000 claims abstract description 27
- 238000012549 training Methods 0.000 claims abstract description 24
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims abstract description 13
- 230000004913 activation Effects 0.000 claims abstract description 10
- 238000007906 compression Methods 0.000 claims description 20
- 230000006835 compression Effects 0.000 claims description 19
- 230000000903 blocking effect Effects 0.000 claims description 5
- 238000012360 testing method Methods 0.000 description 12
- 230000006870 function Effects 0.000 description 10
- 238000006731 degradation reaction Methods 0.000 description 7
- 238000002474 experimental method Methods 0.000 description 6
- 238000000605 extraction Methods 0.000 description 6
- 230000008569 process Effects 0.000 description 6
- 238000005070 sampling Methods 0.000 description 6
- 230000015556 catabolic process Effects 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 5
- 238000012545 processing Methods 0.000 description 5
- 230000000007 visual effect Effects 0.000 description 5
- 238000010586 diagram Methods 0.000 description 3
- 230000006872 improvement Effects 0.000 description 3
- 238000013507 mapping Methods 0.000 description 3
- 238000013527 convolutional neural network Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 101100365548 Caenorhabditis elegans set-14 gene Proteins 0.000 description 1
- 235000002566 Capsicum Nutrition 0.000 description 1
- 241000283070 Equus zebra Species 0.000 description 1
- 241000758706 Piperaceae Species 0.000 description 1
- 241000287531 Psittacidae Species 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 239000000969 carrier Substances 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000009877 rendering Methods 0.000 description 1
Images
Classifications
-
- 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
- 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
- G06T3/4023—Scaling of whole images or parts thereof, e.g. expanding or contracting based on decimating pixels or lines of pixels; based on inserting pixels or lines of pixels
-
- 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
- G06T3/4076—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 using the original low-resolution images to iteratively correct the high-resolution images
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a super-resolution reconstruction method for a single compressed image based on a cascade framework. The invention discloses a super-resolution reconstruction method aiming at a single compressed image, which is characterized in that a decompression effect sub-network and a super-resolution sub-network are cascaded, and the super-resolution reconstruction method mainly comprises the following steps: constructing a decompression effect removing network based on a wide activation residual block for an input low-resolution compressed image to obtain a denoised low-resolution image; constructing a super-resolution reconstruction network based on a wide-activation residual block for the denoised low-resolution image to obtain a final high-resolution image; training the constructed decompression effect removing network and the super-resolution reconstruction network by utilizing a training image data set; in the image reconstruction stage, a low-resolution compressed image is used as input, and a final high-resolution image is reconstructed by using a trained network. The method can obtain good subjective and objective effects, and is an effective single-image super-resolution reconstruction method.
Description
Technical Field
The invention relates to an image resolution improvement technology, in particular to a super-resolution reconstruction method for a single compressed image based on a cascade framework, and belongs to the field of image processing.
Background
Images, which are one of the main information carriers, have a very important position in human production and life because of their intuitive and vivid characteristics. With the rapid development of the network science and technology era, the data volume of images and videos is increased explosively, and a serious load is brought to the storage and transmission of information. In addition, the demand for image quality and resolution is increasing, and especially in the application fields of medicine, military, remote sensing, astronomy, intelligent monitoring and the like, clearer, higher-resolution and better-quality images are required to be obtained, and the demand of the like presents huge challenges to the existing technologies and equipment. At this stage, images are often acquired, stored and transmitted in a low resolution form, and are usually compressed by a certain factor to save storage space and bandwidth resources. However, the improvement of the compression ratio may cause a serious distortion phenomenon in the image, and the quality of the image is reduced, and the reduction of the image acquisition resolution may also cause excessive loss of high-frequency detail information of the image, so that the acquired image may not well meet application requirements in different aspects.
The super-resolution reconstruction technology can improve the resolution of the collected degraded images and videos, and has the characteristics of low cost and strong practicability, so that the super-resolution reconstruction technology has very important theoretical and practical significance. However, most super-resolution reconstruction algorithms are directed at non-compressed images, that is, only a down-sampling or blurring degradation condition exists in a processed low-resolution image, and a compression effect introduced by a compressed image in a quantization stage has a strong correlation with information of the image, and particularly under the condition of a high compression multiple, the conventional super-resolution reconstruction algorithms for non-compressed images cannot effectively process the compressed image.
Disclosure of Invention
The invention aims to solve the problems and provide a super-resolution reconstruction method for a single compressed image based on a cascade framework. In the present invention, the super-resolution reconstruction method for the non-compressed image is collectively referred to as conventional super-resolution, and the super-resolution reconstruction method for the compressed image is referred to as compressed super-resolution. The method of the invention uses a convolutional neural network based on deep learning to decompose a compression hyper-division task into two sub-problems of decompression effect and super-resolution reconstruction, namely, decompression effect processing is firstly carried out, and then super-resolution reconstruction is carried out on the processed result, thereby constructing a cascaded super-resolution reconstruction frame.
The invention provides a super-resolution reconstruction method of a single compressed image based on a cascade framework, which mainly comprises the following operation steps:
(1) constructing a decompression effect removing network based on a wide activation residual block for an input low-resolution compressed image, and removing compression noise to obtain a denoised low-resolution image;
(2) training the decompression effect network constructed in the step (1) by utilizing a training image set;
(3) constructing a super-resolution reconstruction network based on a wide-activation residual block for the denoised low-resolution image so as to improve the image resolution and obtain a final high-resolution image;
(4) training the super-resolution reconstruction network constructed in the step (3) by using a training image data set;
(5) in the image reconstruction stage, the low-resolution compressed image is used as input, the network trained in the step (2) is used for reconstructing a low-resolution image, the obtained image is used as input, and the network trained in the step (4) is used for reconstructing a final high-resolution image.
Drawings
FIG. 1 is a schematic block diagram of a super-resolution reconstruction method of a single compressed image based on a cascade framework
FIG. 2 is a block diagram of a wide activation residual block used in the convolutional neural network of the present invention
FIG. 3 is a comparison graph of reconstruction results of a test image "Circuit" according to the present invention and six methods (super resolution reconstruction factor is 2, compressed image quality factor is 10): wherein, (a) is the original image, and (b) to (h) are bicubic interpolation, methods 1 to 5 and the reconstruction result of the invention respectively
Fig. 4 is a comparison diagram of the reconstruction result of the test image "Foreman" by the present invention and six methods (the super-resolution reconstruction factor is 2, and the compressed image quality factor is 20): wherein, (a) is the original image, and (b) to (h) are bicubic interpolation, methods 1 to 5 and the reconstruction result of the invention respectively
Fig. 5 is a graph comparing the reconstruction results of the test image "Ppt 3" according to the present invention and six methods (super resolution reconstruction factor of 2, compressed image quality factor of 30): wherein, (a) is the original image, and (b) to (h) are bicubic interpolation, methods 1 to 5 and the reconstruction result of the invention respectively
Detailed Description
The invention will be further described with reference to the accompanying drawings in which:
in fig. 1, the method for reconstructing the super-resolution of the single compressed image based on the cascade framework specifically includes the following five steps:
(1) constructing a decompression effect removing network based on a wide activation residual block for an input low-resolution compressed image, and removing compression noise to obtain a denoised low-resolution image;
(2) training the decompression effect network constructed in the step (1) by utilizing a training image set;
(3) constructing a super-resolution reconstruction network based on a wide-activation residual block for the denoised low-resolution image so as to improve the image resolution and obtain a final high-resolution image;
(4) training the super-resolution reconstruction network constructed in the step (3) by using a training image data set;
(5) and (3) in an image reconstruction stage, taking the low-resolution compressed image as input, reconstructing the low-resolution image by using the network trained in the step (2), taking the obtained image as input, and reconstructing a final high-resolution image by using the network trained in the step (4).
Specifically, in step (1), the reconstruction-based image super-resolution reconstruction method models a degradation process of an image, and a general degradation model can be represented as:
y＝DsHx+n (1)
wherein x represents the original high resolution image, DsThe down sampling matrix is represented, H represents the image blurring matrix, n is white Gaussian noise with the mean value of zero, and y is the observed low-resolution image.
However, super-resolution image reconstruction is an ill-posed problem, and the solution space needs to be normalized by the prior information of the image. The solution of the prior super-resolution reconstruction problem is introduced to be converted into:
the first term is the data fitting term, the second term ψ (x) is the regularization term, i.e. the introduced image prior, and λ is the regularization parameter.
While in low resolution compressed images, the main degradation factors are down-sampling and image compression. The degradation of the image in the compression process is mainly represented by introducing compression noise, and in the field of decompression effect, the image is generally modeled as Gaussian noise. So its degradation model can be expressed as:
where e represents compression noise and z represents a low resolution image that has been degraded by down-sampling, blurring, etc. In order to achieve the compression and over-division task, the above two degradation processes are usually solved separately, and the solution process can be formulated as:
in the formula z*And x*Respectively representing the optimal low resolution image and high resolution image solved by minimizing the errorResolution image, λdAnd λuRespectively representing the regularization parameter, # in the decompression process and the super-resolution reconstruction processd(. cndot.) and ψu(. cndot.) represents the image priors in the two solving processes, respectively.
In the step (1), the constructed decompression effect network utilizes the wide activation residual block shown in fig. 2 to perform channel expansion on the features before the linear rectification function (ReLU) activation, and performs channel restoration on the activated features, so that the network prediction performance can be effectively improved without introducing more network parameters and calculated amount. Under the condition that the parameter quantity and the network computing complexity are unchanged, the wide-activation network can obtain better performance.
Specifically, the constructed decompression effect network is shown in fig. 1. Firstly, stretching an image block into a tensor through image block extraction layer operation; then, directly connecting the output of the image block extraction layer to a network structure containing a wide active residual block, and carrying out nonlinear representation and extraction on redundant information; secondly, adding the information learned after the wide activation residual error network and original image block tensor information to form a global residual error; and finally, stretching and recombining the information through a single convolution layer, and outputting a final prediction result of the decompression effect network. In the image block extraction layer operation, different from JPEG, in the proposed network, the image blocks extracted by the block extraction operation are overlapped, and the overlapping block extraction mode increases the receptive field of the network on one hand, so that the prediction output of the network can utilize more input image information; on the other hand, the operation carries out repeated reconstruction on the pixel to be reconstructed to a certain extent, and can effectively eliminate the blocking effect while improving the network robustness.
In step (2), the high resolution images in the database are processed
Down-sampling to obtain low-resolution image
After compressing it, a low-resolution compressed image is obtained
Use of
And
the formed image sample pair trains a decompression effect network, and in the training process, a loss cost function of the network is defined as:
wherein, N represents the number of training sample pairs input during one forward prediction, i.e. the Size of Batch Size (Batch Size) during network training, and ΘARRepresenting trainable parameters in a decompression effect network, fAR(. is a mapping function of a decompression effect network, /)ARAs a function of the loss during the decompression effect.
In the step (3), a super-resolution reconstruction network based on the wide activation residual block is constructed for the low-resolution image obtained in the step (1) to improve the resolution of the image and obtain a final high-resolution image. In order to improve the training speed and performance of the network, in the super-resolution reconstruction subnetwork of this chapter, an deconvolution layer is used to improve the resolution of an image, that is, the penultimate layer of the network is an deconvolution layer, and the last layer is a convolution output layer.
In the step (4), after the decompression effect network training in the step (2) is finished, the trained network is used for processing the training sample
To obtain its corresponding uncompressed image
Is estimated optimally
Then will be
And
inputting the formed new training sample pairs into a super-resolution reconstruction sub-network to learn the mapping function f for improving the image resolutionSR(. to) the objective of the mapping function is to bring the network into input
Is as close as possible to the original high resolution image
The loss cost function in the super-resolution reconstruction sub-network training process is as follows:
in the formula (c) (-)SRRepresenting trainable parameters in a super resolution reconstruction network,/SRAs a function of the loss during the super-resolution reconstruction.
In the step (5), in the image reconstruction stage, the low-resolution compressed image is used as input, the network trained in the step (2) is used for reconstructing a low-resolution image, the obtained image is used as input, and the network trained in the step (4) is used for reconstructing a final high-resolution image.
To better illustrate the effectiveness of the method of the present invention, 10 test images were selected from the commonly used data sets Set12 and Set14, respectively, "Butterfly," Ppt3, "" Foreman, "" House, "" Leaves, "" Circuit, "" Woman, "" Zebra, "" Peppers, "and" Parrot. The compressed low-resolution image is generated in the following way: for a color image in a test data set, the color image is converted into a YCbCr space, then a brightness channel Y is extracted as an original high-resolution image, then an imresize function in MATLAB is used for carrying out double-three down-sampling on the Y channel of the original image by 2 times, and then an imwrite function is used for carrying out JPEG compression on the down-sampled image so as to generate a low-resolution compressed image to be tested. In the experiment, a bicubic interpolation method and five compressed image super-resolution reconstruction algorithms are selected as comparison methods.
The five contrast compressed image super-resolution methods are as follows:
the method comprises the following steps: the method proposed by Xiao et al, references "Xiao J, Wang C, Hu X," Single Image super-resolution in compressed domain based field of the expert prior, "IEEE International consistency on Image and Signal Processing,2012:607 and 611"
The method 2 comprises the following steps: methods proposed by Li et al, references "Li T, He X, Qing L," iterative frame of blocked and super-resolution for compression maps, "IEEE Transactions on multimedia,2018,20(6): 1305-"
The method 3 comprises the following steps: methods proposed by Kim et al, references "Kim J, Kwon Lee J, Mu Lee K," Accurate image super-resolution using deep relational networks, "IEEE Conference on Computer Vision and Pattern registration, 2016: 1646-"
The method 4 comprises the following steps: the methods proposed by Zhang et al, references "Zhang K, Zuo W, Chen Y," Beyond a gaussian noise ": Residual leading of deep cnn for Image noise" ("IEEE Transactions on Image Processing", 2017,26(7):3142- "
The method 5 comprises the following steps: the method proposed by Chen et al, references "Chen H, He X, Ren C," CISRDCNN: Super-resolution of compressed images using deep connected network networks, "neuro-rendering, 2018,285: 204-219"
The contents of the comparative experiment are as follows:
experiment 1, low-resolution compressed images generated by simulating 10 test images are respectively reconstructed by 2 times by bicubic interpolation, methods 1 to 5 and the method of the invention, and the compression quality factor QF is 10. The PSNR (Peak Signal to Noise Ratio, unit: dB) parameters and the average value of the reconstruction results of each method are shown in the table I. In addition, for visual comparison, the results of the "Circuit" image are given, as shown in fig. 3. Fig. 3(a) is a "Circuit" original image, fig. 3(b) to 3(g) are reconstruction results of methods 1 to 5, and fig. 3(h) is a reconstruction result of the present invention.
Watch 1
Experiment 2, low-resolution compressed images generated by simulating 10 test images are respectively reconstructed by 2 times by bicubic interpolation, methods 1 to 5 and the method of the invention, and the compression quality factor QF is 20. The PSNR (Peak Signal to Noise Ratio, unit: dB) parameters and the average values of the reconstruction results of the methods are shown in Table II. In addition, for visual comparison, the results of the "Foreman" image are given, as shown in fig. 4. Fig. 4(a) is a "Foreman" original image, fig. 4(b) to 4(g) are reconstruction results of methods 1 to 5, and fig. 4(h) is a reconstruction result of the present invention.
Watch two
Experiment 3, low-resolution compressed images generated by simulating 10 test images are respectively reconstructed by 2 times by bicubic interpolation, methods 1 to 5 and the method of the invention, and the compression quality factor QF is 30. Table III shows the PSNR (Peak Signal to Noise Ratio, unit: dB) parameters and the average value of the reconstruction results of each method for ten test charts. In addition, for visual comparison, the results of the "Ppt 3" image are given, as shown in fig. 5. Fig. 5(a) shows the original image "Ppt 3", fig. 5(b) to 5(g) show the reconstruction results of methods 1 to 5, and fig. 5(h) shows the reconstruction results of the present invention.
Watch III
Experiment 4, low-resolution compressed images generated by simulating 10 test images are respectively reconstructed by 2 times by bicubic interpolation, methods 1 to 4 and the method of the invention, and the compression quality factor QF is 40. The PSNR (Peak Signal to Noise Ratio, unit: dB) parameters and the average values of the reconstruction results of each method for ten test charts are shown in Table four. The fifth table shows the average ssim (structure Similarity index) and ifc (information Fidelity criterion) parameters of each reconstruction result for ten test charts under different compression quality factors.
Watch four
Watch five
As can be seen from the experimental results shown in fig. 3, 4 and 5, the results of the bicubic interpolation method and the method 1 contain relatively obvious blocking effect and residual noise, and the image visual effect is relatively poor; although the methods 2 to 5 remove the blocking effect in the compressed image to some extent, the details of the reconstruction result are not rich enough. In contrast, the method provided by the invention can effectively inhibit the compression effects such as blocking effect and artifact existing in the compressed image while improving the resolution of the compressed image. The reconstructed result image has richer detail information, and no new error information is introduced into the reconstructed result. As in fig. 3, compared with the rest of the comparison method, the texture information of the reconstructed result is richer and the character is clearer. Similarly, in fig. 4, the boundary between the boards of the image reconstructed by the method of the present invention is clearer, and the texture information on the boards is also recovered to a certain extent, so as to avoid the phenomenon that the reconstruction result is too smooth. In fig. 5, the texture information of the reconstructed result of the method of the present invention is rich, the lines of the grid area are clear, and the artifacts in the characters are obviously removed. In addition, from the PSNR, SSIM and IFC parameters given in tables one to five, the present invention obtains the highest values in all three indexes, and the improvement is obvious. Therefore, the subjective visual effect and the objective parameters of the reconstruction results of the methods are comprehensively compared, so that the method has better reconstruction effect and is suitable for low-resolution compressed images. In conclusion, the invention is an effective super-resolution reconstruction method for a single compressed image.
Claims (5)
1. The super-resolution reconstruction method of the single compressed image based on the cascade framework is characterized by comprising the following steps of:
the method comprises the following steps: constructing a decompression effect removing network based on a wide activation residual block for an input low-resolution compressed image, and removing compression noise to obtain a denoised low-resolution image;
step two: training the decompression effect removing network constructed in the first step by using a training image set;
step three: constructing a super-resolution reconstruction network based on a wide-activation residual block for the denoised low-resolution image so as to improve the image resolution and obtain a final high-resolution image;
step four: training a super-resolution reconstruction network constructed in the third step by using a training image data set;
step five: in the image reconstruction stage, the low-resolution compressed image is used as input, the network trained in the step two is used for reconstructing a low-resolution image, the obtained image is used as input, and the network trained in the step four is used for reconstructing a final high-resolution image.
2. The super-resolution reconstruction method for single compressed images based on cascaded framework as claimed in claim 1, wherein the step one is a decompression effect network, which implements overlapped block fetching of images by setting fixed convolution layer, and this overlapped block fetching increases the receptive field of the network on one hand, so that the predicted output of the network can utilize more input image information; on the other hand, the operation carries out repeated reconstruction on the pixel to be reconstructed to a certain extent, and can effectively eliminate the blocking effect while improving the network robustness.
3. The super-resolution reconstruction method for single compressed images based on cascade framework as claimed in claim 1, wherein the super-resolution reconstruction network comprises a decompression effect network in the first step and a super-resolution reconstruction network in the third step, wherein the network utilizes wide activated residual blocks to perform channel expansion on the features before Linear rectification function (ReLU) activation, and performs channel restoration on the activated features, so that the network prediction performance can be effectively improved without introducing more network parameters and calculation amount; under the condition that the parameter quantity and the network computing complexity are unchanged, the wide-activation network can obtain better performance.
4. The method for reconstructing super-resolution of single compressed image based on cascaded framework as claimed in claim 1, wherein the input of the super-resolution reconstruction network in step three is the output of the decompression effect network in step one, so as to implement the cascade connection of the decompression effect network and the super-resolution network, and divide the super-resolution reconstruction problem of the low-resolution compressed image into two sub-links of decompression effect and super-resolution reconstruction.
5. The cascade-frame-based single compressed image super-resolution reconstruction method according to claim 1, wherein the super-resolution reconstruction network in the fourth step uses the image processed by the decompression effect network in the first step and the down-sampled low-resolution image as a training image pair when training the network, and the super-resolution reconstruction network trained in this way is more targeted and can effectively improve the performance of the final network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910571747.5A CN112150356A (en) | 2019-06-28 | 2019-06-28 | Single compressed image super-resolution reconstruction method based on cascade framework |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910571747.5A CN112150356A (en) | 2019-06-28 | 2019-06-28 | Single compressed image super-resolution reconstruction method based on cascade framework |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112150356A true CN112150356A (en) | 2020-12-29 |
Family
ID=73869013
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910571747.5A Pending CN112150356A (en) | 2019-06-28 | 2019-06-28 | Single compressed image super-resolution reconstruction method based on cascade framework |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112150356A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113096019A (en) * | 2021-04-28 | 2021-07-09 | 中国第一汽车股份有限公司 | Image reconstruction method, image reconstruction device, image processing equipment and storage medium |
| CN113506215A (en) * | 2021-06-22 | 2021-10-15 | 中国公路工程咨询集团有限公司 | Super-resolution image reconstruction method and device based on wide activation and electronic equipment |
| 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 |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106709875A (en) * | 2016-12-30 | 2017-05-24 | 北京工业大学 | Compressed low-resolution image restoration method based on combined deep network |
| CN107563965A (en) * | 2017-09-04 | 2018-01-09 | 四川大学 | Jpeg compressed image super resolution ratio reconstruction method based on convolutional neural networks |
| CN109214989A (en) * | 2018-09-04 | 2019-01-15 | 四川大学 | Single image super resolution ratio reconstruction method based on Orientation Features prediction priori |
-
2019
- 2019-06-28 CN CN201910571747.5A patent/CN112150356A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106709875A (en) * | 2016-12-30 | 2017-05-24 | 北京工业大学 | Compressed low-resolution image restoration method based on combined deep network |
| CN107563965A (en) * | 2017-09-04 | 2018-01-09 | 四川大学 | Jpeg compressed image super resolution ratio reconstruction method based on convolutional neural networks |
| CN109214989A (en) * | 2018-09-04 | 2019-01-15 | 四川大学 | Single image super resolution ratio reconstruction method based on Orientation Features prediction priori |
Non-Patent Citations (2)
| Title |
|---|
| HONGGANG CHEN等: ""CISRDCNN: Super-resolution of compressed images using deep convolutional neural networks"", 《NEUROCOMPUTING》 * |
| JIAHUI YU等: ""Wide activation for efficient and accurate image super-resolution"", 《COMPUTER VISION AND PATTERN RECOGNITON》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113096019A (en) * | 2021-04-28 | 2021-07-09 | 中国第一汽车股份有限公司 | Image reconstruction method, image reconstruction device, image processing equipment and storage medium |
| CN113506215A (en) * | 2021-06-22 | 2021-10-15 | 中国公路工程咨询集团有限公司 | Super-resolution image reconstruction method and device based on wide activation and electronic equipment |
| 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 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106709875B (en) | Compressed low-resolution image restoration method based on joint depth network | |
| CN107563965A (en) | Jpeg compressed image super resolution ratio reconstruction method based on convolutional neural networks | |
| CN109272452B (en) | Method for learning super-resolution network based on group structure sub-band in wavelet domain | |
| CN112801877B (en) | Super-resolution reconstruction method of video frame | |
| CN112150356A (en) | Single compressed image super-resolution reconstruction method based on cascade framework | |
| Luo et al. | Lattice network for lightweight image restoration | |
| CN112218094A (en) | JPEG image decompression effect removing method based on DCT coefficient prediction | |
| Qin et al. | Difficulty-aware image super resolution via deep adaptive dual-network | |
| CN113962882B (en) | JPEG image compression artifact eliminating method based on controllable pyramid wavelet network | |
| CN112188217B (en) | JPEG compressed image decompression effect removing method combining DCT domain and pixel domain learning | |
| CN110288529B (en) | Single image super-resolution reconstruction method based on recursive local synthesis network | |
| CN115526779A (en) | Infrared image super-resolution reconstruction method based on dynamic attention mechanism | |
| CN114926883A (en) | Face image processing method meeting various degradation models | |
| Amaranageswarao et al. | Joint restoration convolutional neural network for low-quality image super resolution | |
| CN108550111B (en) | Residual error example regression super-resolution reconstruction method based on multi-level dictionary learning | |
| Li et al. | RGSR: A two-step lossy JPG image super-resolution based on noise reduction | |
| CN112070676B (en) | Picture super-resolution reconstruction method of double-channel multi-perception convolutional neural network | |
| CN113837935A (en) | Compressed image super-resolution reconstruction method based on attention-enhancing network | |
| CN113628114A (en) | Image super-resolution reconstruction method of two-channel sparse coding | |
| CN110807746A (en) | Hyperspectral image sharpening method based on detail embedded injection convolutional neural network | |
| CN111402141A (en) | Tandem single-image super-resolution reconstruction method | |
| CN112308772A (en) | Super-resolution reconstruction method based on deep learning local and non-local information | |
| CN114862687B (en) | Self-adaptive compressed image restoration method driven by depth deblocking operator | |
| Zhou et al. | Cross-Scale Residual Network for Multiple Tasks: Image Super-resolution, Denoising, and Deblocking | |
| Fan et al. | Shallow and deep convolutional networks for image super-resolution |
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: 20201229 |