CN112581367B - GPR image super-resolution reconstruction method based on random sample classification amplification - Google Patents
GPR image super-resolution reconstruction method based on random sample classification amplification Download PDFInfo
- Publication number
- CN112581367B CN112581367B CN202011384484.6A CN202011384484A CN112581367B CN 112581367 B CN112581367 B CN 112581367B CN 202011384484 A CN202011384484 A CN 202011384484A CN 112581367 B CN112581367 B CN 112581367B
- Authority
- CN
- China
- Prior art keywords
- image
- training
- resolution
- sample
- sample library
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 74
- 230000003321 amplification Effects 0.000 title claims abstract description 16
- 238000003199 nucleic acid amplification method Methods 0.000 title claims abstract description 16
- 238000012549 training Methods 0.000 claims abstract description 87
- 238000004364 calculation method Methods 0.000 claims abstract description 8
- 239000011159 matrix material Substances 0.000 claims description 17
- 238000012360 testing method Methods 0.000 claims description 6
- 230000000903 blocking effect Effects 0.000 claims description 5
- 238000012545 processing Methods 0.000 claims description 5
- 238000004422 calculation algorithm Methods 0.000 claims description 4
- 238000005516 engineering process Methods 0.000 description 5
- 230000003416 augmentation Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000005070 sampling Methods 0.000 description 3
- 230000001149 cognitive effect Effects 0.000 description 2
- 239000004973 liquid crystal related substance Substances 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 101100365548 Caenorhabditis elegans set-14 gene Proteins 0.000 description 1
- 241000282414 Homo sapiens Species 0.000 description 1
- 108010072610 N-acetyl-gamma-glutamyl-phosphate reductase Proteins 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000011840 criminal investigation Methods 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000013441 quality evaluation 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
- 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
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a GPR image super-resolution reconstruction method based on random sample classification amplification, which comprises the steps of constructing a training sample library, and classifying the training sample library to obtain K through random sample classification amplificationA training sample set; training a Gaussian process regression model corresponding to each sample set by adopting a training method of Gaussian process regression for each sample set; interpolation of the low resolution image to be measured to the desired size to obtain an image U I The method comprises the steps of carrying out a first treatment on the surface of the Calculation image U I The distance between the middle image block and the training sample set; from image U I And carrying out image reconstruction by using a Gaussian process regression model corresponding to the sample set with the closest distance. The invention can more effectively recover the effective detail information of the image, improves the training speed and saves the time cost.
Description
Technical Field
The invention belongs to an image super-resolution technology, and particularly relates to a GPR image super-resolution reconstruction method based on random sample classification amplification.
Background
The vision bears more than 80% of external cognitive information of human beings, and the computer vision derived from the external cognitive information becomes a key technology for artificial intelligence development and is widely researched and applied. Among them, a technique of reconstructing an unknown high-resolution image using one or more low-resolution images is called super-resolution technique. Super-resolution technology is widely applied to a plurality of fields of aerospace, medical detection, criminal investigation and the like in real life. In reality, due to limitations in hardware or cost, a significant portion of the acquired image has problems such as too low resolution, inability to provide satisfactory image details, and the like. For these low resolution images, how to reconstruct or predict the corresponding unknown high resolution images by the existing theory and technology has important practical significance and theoretical value for image processing and perception.
To improve efficiency, many existing self-learning Xi Chao resolution methods based on gaussian process regression are often based on local samples, which makes them unable to take full advantage of the widely existing block self-similarity of natural images themselves. To address this problem, the gaussian process regression-based document Fast single image super-resolution using sparse Gaussianprocess regression proposes a non-local, self-learning Xi Chao-resolved reconstruction framework that learns a non-local gaussian process regression model rather than a plurality of gaussian process regression models based on local images for super-resolution reconstruction. The method adopts an active learning strategy to heuristically select samples with more information to train regression parameters of the Gaussian process regression model, so that higher reconstruction quality can be maintained, and the calculation efficiency can be remarkably improved. However, this approach has certain limitations in that the randomness of the sampling can result in machine-learned mappings that may not be suitable for accurate reconstruction everywhere in the image, resulting in image local reconstruction bias.
Disclosure of Invention
The invention provides a GPR image super-resolution reconstruction method based on random sample classification amplification.
The technical scheme for realizing the purpose of the invention is as follows: a GPR image super-resolution reconstruction method based on random sample classification amplification comprises the following specific steps:
step 1: constructing a training sample library, wherein the training sample library comprises image pairs formed by original high-resolution images and low-resolution images with the same size as the high-resolution images;
step 2: classifying the training sample library to obtain K training sample sets through random classification amplification;
step 3: training a Gaussian process regression model corresponding to each sample set by adopting a training method of Gaussian process regression for each sample set;
step 4: interpolation of the low resolution image to be measured to the desired size to obtain an image U I ;
Step 5: calculation image U I Middle image block and training sample set D i Is a distance of (2);
step 6: from image U I Corresponding nearest sample set D i And (5) carrying out image reconstruction by a corresponding Gaussian process regression model.
Preferably, the construction method of the training sample library comprises the following steps: blurring and downsampling are carried out on the high-resolution image in the original training gallery, interpolation is carried out on the downsampled image so that the downsampled image is the same as the corresponding original high-resolution image in size, and the interpolation image and the corresponding original high-resolution image form an image pair;
blocking the images in the image pair to obtain an image matrixWherein P is i (I) Representing an image block matrix formed after the i-th interpolation image is blocked, P i (H) Representing an image block matrix formed after the i-th original high-resolution image block processing;
determining a training sample library D according to the image matrix, wherein the training sample library D comprises the following concrete steps:
where cen (·) represents the center pixel of the extracted image block matrix, cen (P) i (H) )-cen(P i (I) ) Representing the difference between the extracted high resolution image center pixel and the interpolated image center pixel.
Preferably, the formula for blurring and downsampling the high resolution image in the original training gallery is:
L k =(H*B k )↓+ε k ,k=1,2,…,K
wherein L is k For the kth downsampled image, H is the corresponding high resolution image, ε k Noise of B k For convolution operations, +..
Preferably, classifying the training sample library to obtain K training sample sets comprises random classification and classification augmentation, and the specific method is as follows:
random classification: randomly extracting a sample library S from the training sample library D, and simultaneously obtaining a residual sample library M=D\S; adopting a K-Means algorithm to perform initial classification, and dividing a sample library S into a sample library training subset L-type training subset;
and (3) classification amplification: for all image blocks x j E M, calculate image block x j And sample library training subset S l Distance d (x) j ,S l ) According to the distance d (x j ,S l ) To determine the size of the image block x j Where l=1, 2, …, L.
Preferably, according to the distance d (x j ,S l ) The size of (l=1, 2, …, L), the image block x is determined j The specific method of the attribution set of (a) is as follows:
if d (x) j ,S l ) T is less than or equal to T, x j Training the subset to the sample library with the smallest belonged distance, and then letting S i =S i ∪{x j T is a set threshold;
if d (x) j ,S l ) > T, a new training subset S is added L+1 ={x j -l+1→l simultaneously;
let k=l, obtain the final training sample set S 1 ,S 2 ,…,S L Wherein D is 1 =S 1 ,D 2 =S 2 ,…,D K =S L 。
Preferably, the set threshold is specifically:
T=1.5×max{ρ l |l=1,2,…,L}
wherein ρ is l Representing sample library training subset S l Is a set of polar paths.
Preferably, image block x j And sample library training subset S l The calculation formula of the distance of (2) is as follows:
wherein C (C) n )={c n ||c n -d|<r,n=1,2,…,k},x * Representing image blocks, C (C) n ) Represents a set of points of radius r from its cluster center D for sample set D, k being the set size, D (x * D) test image block x representing an input image * Average distance from sample set D, c n Indicating that for sample set D, it is no more than a point of radius r from its cluster center D.
Compared with the prior art, the invention has the remarkable advantages that: according to the invention, the multi-component learning is realized by carrying out random classification augmentation on the image blocks, classifying and discussing the image block pairs selected in the sample set for machine learning, and grouping the image block pairs into groups for subsequent training and testing stages, so that the comprehensiveness and reliability of the sampling process are improved, the finally obtained image is more vivid, the operation amount is reduced, and the training efficiency is improved;
according to the invention, the image block classification method is designed in a targeted manner by introducing the image super-resolution reconstruction technology, so that the image super-resolution reconstruction quality is improved on the basis of the original method, and the time spent on training a model is reduced.
The present invention will be described in further detail with reference to the accompanying drawings.
Drawings
FIG. 1 is a flow chart of the present invention.
Fig. 2 is a flow chart of random classification.
FIG. 3 is a flow chart of class augmentation.
Fig. 4 is a graph showing the comparison of the effect of the present invention based on the Butterfly image with the conventional image reconstruction method.
Fig. 5 is a graph comparing the reconstruction effect of the present invention based on the Lenna image with the existing image reconstruction method.
Fig. 6 is a graph comparing the reconstruction effect of the invention method with that of the prior image reconstruction method based on Kodim23 image.
Fig. 7 is a graph comparing the reconstruction effect of the present invention based on the Urban036 image with the existing image reconstruction method.
Detailed Description
A GPR image super-resolution reconstruction method based on random sample classification amplification comprises the following specific steps:
step 1: constructing a training sample library, wherein the training sample library comprises image pairs formed by original high-resolution images and low-resolution images with the same size as the high-resolution images;
in a further embodiment, the method for constructing the training sample library includes: for original training diagramBlurring and downsampling the high-resolution images in the library, interpolating the downsampled images to be the same size as the corresponding original high-resolution images, and forming an image pair from the interpolated images and the corresponding original high-resolution imagesWherein I is i Representing the ith interpolated image IR, H i Representing an ith high resolution image, N being the number of image pairs;
image is segmented to obtain an image matrixWherein P is i (I) Representing an image block matrix formed after the i-th interpolation image is blocked, P i (H) Representing the image block matrix formed after the i-th original high resolution image blocking process.
Determining a training sample library D according to the image matrix, wherein the training sample library D comprises the following concrete steps:
where cen (·) represents the center pixel of the extracted image block matrix, cen (P) i (H) )-cen(P i (I) ) Representing the difference between the extracted high resolution image center pixel and the interpolated image center pixel. For training sample library D, x i Stores an image block matrix formed after the IR blocking process of the ith interpolated image, and y i The residual of the center pixel of the i-th high resolution image and the center pixel of the i-th interpolation image is stored.
The specific formula for blurring and downsampling a high resolution image is:
L k =(H*B k )↓+ε k ,k=1,2,…,K (2)
wherein L is k For the kth downsampled image, H is the corresponding high resolution image, ε k Noise of B k In order for the convolution operation to be performed,and ∈r is a downsampling operator.
Step 2: classifying the training sample library to obtain K training sample sets D through random classification amplification 1 ,D 2 ,…D K ;
In a further embodiment, classifying the training sample library to obtain K training sample sets includes random classification and classification augmentation, and the specific method is as follows:
random classification: and randomly extracting a sample library S from the training sample library D, and simultaneously obtaining a residual sample library M=D\S. The K-Means algorithm is adopted to conduct initial classification, and the sample library S is divided into a sample library training subset S 1 ,S 2 ,…,S L Class L training subsets. For the remaining sample libraryM contains information of J tiles. Obviously, J < N.
And (3) classification amplification: for all image blocks x j E M, calculate image block x j And sample library training subset S l Distance d (x) j ,S l ) According to the distance d (x j ,S l ) To determine the size of the image block x j Where l=1, 2, …, L.
Further, the specific judging method comprises the following steps: let t=1.5×max { ρ l L=1, 2, …, L }, where ρ l Representing sample library training subset S l Is a set of polar paths. If d (x) j ,S l ) T is less than or equal to T, x j Training the subset by belonging to the sample library with the minimum distance; if d (x) j ,S l ) And > T, a new training subset is added.
If x j Belonging to a certain sample library training subset S i (i.e {1,2, …, L }), let S i =S i ∪{x j -a }; if x j Failing to attribute to any sample library training subset, the training subset S is added L+1 ={x j And simultaneously L+1→L. Repeating the above process for each image block x j And E, performing the judgment on the E M. Let k=l, obtain the final training sample set S 1 ,S 2 ,…,S L Wherein D is 1 =S 1 ,D 2 =S 2 ,…,D K =S L 。
In a further embodiment, image block x j And sample library training subset S l The calculation formula of the distance of (2) is as follows:
wherein C (C) n )={c n ||c n -d|<r,n=1,2,…,k},x * Representing image blocks, C (C) n ) Represents a set of points of radius r from its cluster center D for sample set D, k being the set size, D (x * D) test image block x representing an input image * Average distance from sample set D.
Step 3: training a Gaussian process regression model corresponding to each sample set by using a Gaussian process regression training method for each sample set obtained in the step (2);
using training dataset D i I=1, 2, …, K stored i-th interpolated image IR blocking process followed by forming an image block matrix x i And the residual y of the center pixel of the ith high resolution image and the center pixel of the ith interpolation image i GPR training was performed. GPR hypothesis training data satisfies the observation model y=f+epsilon, whereAnd the noise or error term is an independent co-distributed Gaussian random variable, i.e. noise +.>Using training sample set D 1 ,D 2 ,…,D K Training corresponding K Gaussian process regression models f 1 ,f 2 ,…,f K 。
Step 4: interpolation of the low resolution image to be measured to the desired size to obtain an image U I ;
Step 5: calculation image U I Middle image block x * And training sample set D i Distance d (x) * ,D i );
Step 6: from image U I Corresponding nearest sample set D i Corresponding GPR model f i And (5) performing image reconstruction. Predicting U using Gaussian process regression models I In the new high frequency details, and finally adding the predicted high frequency details to U I And obtaining super-resolution images.
In order to better measure the super-resolution reconstruction quality, the super-resolution reconstruction quality is measured by comparing the super-resolution reconstructed image with the original high-resolution image. As an image quality evaluation index, peak Signal-to-Noise Ratio (PSNR) and image structure similarity (Structural Similarity Image Measurement, SSIM) are adopted.
The definition of peak signal to noise ratio is as follows:
wherein I is the original high resolution image,for super-resolution reconstruction of predicted image, M, N are the size of the image, MAX I The maximum value of the pixel value in the image is taken. For example, the pixel value range in the image is [0,255 ]]MAX is MAX I 255.
Calculating local structural similarity on corresponding local areas of the two images:
wherein, the liquid crystal display device comprises a liquid crystal display device,for image->Is provided with a corresponding partial region of the (c),/>respectively->Mean value of->Respectively->Variance of->Is->Covariance of c 1 ,c 2 Is a parameter for adjusting the SSIM value.
And calculating all the local structural similarity and taking an average value to obtain the average structural similarity of the image.
The following examples and figures will provide those skilled in the art with a more complete understanding of the invention and are not intended to limit the invention in any way.
Examples
All programs of the embodiment are realized by Matlab 2016 software programming, and a hardware platform is PC: intel Core i5-8265U [email protected], 8.00GB memory. The data is selected from a set5 data set, a set14 data set, a kodak data set, a BSDS data set and other common reference test sets in the field of image super-resolution reconstruction, and related experimental results and analysis are as follows:
in this embodiment, the original image set is firstly subjected to a gaussian blur degradation process with a size of 7×7 and a standard deviation of 1.1, and then downsampled by 3 times to obtain a low-resolution image. Comparing the present invention (KGPR) with the prior art, the prior art comprises:
(1) Bicubic, an image interpolation restoration method, is also one of the most basic algorithms in the field of image reconstruction.
(2) SCSR sparse representation image reconstruction method (J.Yang, J.Wright, T.Huang, and y.ma,
"Image super-resolution via sparse characterizeation,"IEEE Trans.ImageProcess.,vol.19,no.11,pp.2861-2873,Nov.2010.)
(3) SRGPR local GPR image reconstruction method (HeH, siu W.Single image super-resolution using Gaussian process regression [ C ]// Computer Vision ]
Pattern Recognition.IEEE,2011)
(4) AGPR is a Gaussian process image super-resolution reconstruction method based on active sampling (Wang, haijun, li).
Fast single image super-resolution using sparse Gaussian processregression[J].Signal Processing the Official Publication of the EuropeanAssociation for Signal Processing,2017.)
According to tables 1 and 2, the PSNR value and the SSIM value under the different image reconstruction methods are compared, and the method has remarkable advantages that the parameter values are higher than those of other methods. According to table 3, the invention also significantly reduces the time of the training process, improves the training speed and saves the time cost.
As can be seen from comparison of the restored images of fig. 4 to 7, the present invention better restores the image information for different test images. The visual effect of the image obtained by the method is better, and the comparison of peak signal to noise ratio (PSNR) can also show that the peak signal to noise ratio of the image obtained by the method is relatively higher.
For example, as can be seen from the comparison of the restored images in fig. 5, the Gauss process image super-resolution reconstruction method after the image block clustering process can reconstruct the details of the images better, the restoration of hair in the Lenna image is better, and the pseudo-structure is less likely to appear in the smooth area, so that the image is smoother, and the visual effect of the restored images is enhanced.
Table 1 comparison table of PSNR values for different image reconstruction methods
Table 2 comparison table of SSIM values for different image reconstruction methods
Table 3 comparison table of GPU run times under different image reconstruction methods
Claims (3)
1. A GPR image super-resolution reconstruction method based on random sample classification amplification is characterized by comprising the following specific steps:
step 1: constructing a training sample library, wherein the training sample library comprises image pairs formed by original high-resolution images and low-resolution images with the same size as the high-resolution images;
step 2: the training sample library is classified to obtain K training sample sets through random classification and amplification, and the specific method is as follows:
random classification: randomly extracting a sample library S from the training sample library D, and simultaneously obtaining a residual sample library M=D\S; adopting a K-Means algorithm to perform initial classification, and dividing a sample library S into a sample library training subset L-type training subset;
and (3) classification amplification: for all image blocks x j E M, calculate image block x j And sample library training subset S l Distance d (x) j ,S l ) According to the distance d (x j ,S l ) To determine the size of the image block x j Wherein, l=1, 2, …, the specific method of L is:
if d (x) j ,S l ) T is less than or equal to T, x j Training the subset to the sample library with the smallest belonged distance, and then letting S i =S i ∪{x j T is a set threshold; t=1.5×max { ρ l |l=1,2,…,L}
Wherein ρ is l Representing sample library training subset S l Is a polar diameter set;
if d (x) j ,S l ) > T, a new training subset S is added L+1 ={x j -l+1→l simultaneously;
let k=l, obtain the final training sample set S 1 ,S 2 ,…,S L Wherein D is 1 =S 1 ,D 2 =S 2 ,…,D K =S L ;
Image block x j And sample library training subset S l The calculation formula of the distance of (2) is as follows:
wherein C (C) n )={c n ||c n -d|<r,n=1,2,…,k},x * Representing image blocks, C (C) n ) Represents a set of points of radius r from its cluster center D for sample set D, k being the set size, D (x * D) test image block x representing an input image * Average distance from sample set D, c n Representing that for sample set D, the point from its cluster center D is no more than a radius r;
step 3: training a Gaussian process regression model corresponding to each sample set by adopting a Gaussian process regression training method for each sample set;
step 4: interpolation of the low resolution image to be measured to the desired size to obtain an image U I ;
Step 5: calculation image U I The distance between the middle image block and the training sample set;
step 6: from image U I And carrying out image reconstruction by using a Gaussian process regression model corresponding to the sample set with the closest distance.
2. The GPR image super-resolution reconstruction method based on random sample classification amplification according to claim 1, wherein the training sample library is constructed by the following steps: blurring and downsampling are carried out on the high-resolution image in the original training gallery, interpolation is carried out on the downsampled image so that the downsampled image is the same as the corresponding original high-resolution image in size, and the interpolation image and the corresponding original high-resolution image form an image pair;
blocking the images in the image pair to obtain an image matrixWherein P is i (I) Representing an image block matrix formed after the i-th interpolation image is blocked, P i (H) Representing an image block matrix formed after the i-th original high-resolution image block processing;
determining a training sample library D according to the image matrix, wherein the training sample library D comprises the following concrete steps:
where cen (·) represents the center pixel of the extracted image block matrix, cen (P) i (H) )-cen(P i (I) ) Representing the difference between the extracted high resolution image center pixel and the interpolated image center pixel.
3. The GPR image super-resolution reconstruction method based on random sample classification amplification according to claim 2, wherein the formula for blurring and downsampling the high-resolution image in the original training gallery is:
L k =(H*B k )↓+ε k ,k=1,2,…,K
wherein L is k For the kth downsampled image, H is the corresponding high resolution image, ε k Noise of B k For convolution operations, +..
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011384484.6A CN112581367B (en) | 2020-12-01 | 2020-12-01 | GPR image super-resolution reconstruction method based on random sample classification amplification |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011384484.6A CN112581367B (en) | 2020-12-01 | 2020-12-01 | GPR image super-resolution reconstruction method based on random sample classification amplification |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112581367A CN112581367A (en) | 2021-03-30 |
CN112581367B true CN112581367B (en) | 2023-06-30 |
Family
ID=75126548
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202011384484.6A Active CN112581367B (en) | 2020-12-01 | 2020-12-01 | GPR image super-resolution reconstruction method based on random sample classification amplification |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112581367B (en) |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105488759A (en) * | 2015-12-09 | 2016-04-13 | 南京邮电大学 | Image super-resolution reconstruction method based on local regression model |
CN109712073A (en) * | 2018-12-19 | 2019-05-03 | 重庆邮电大学 | A kind of image super-resolution rebuilding method returned based on Gaussian process |
CN110097499A (en) * | 2019-03-14 | 2019-08-06 | 西安电子科技大学 | The single-frame image super-resolution reconstruction method returned based on spectrum mixed nucleus Gaussian process |
-
2020
- 2020-12-01 CN CN202011384484.6A patent/CN112581367B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105488759A (en) * | 2015-12-09 | 2016-04-13 | 南京邮电大学 | Image super-resolution reconstruction method based on local regression model |
CN109712073A (en) * | 2018-12-19 | 2019-05-03 | 重庆邮电大学 | A kind of image super-resolution rebuilding method returned based on Gaussian process |
CN110097499A (en) * | 2019-03-14 | 2019-08-06 | 西安电子科技大学 | The single-frame image super-resolution reconstruction method returned based on spectrum mixed nucleus Gaussian process |
Also Published As
Publication number | Publication date |
---|---|
CN112581367A (en) | 2021-03-30 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109325550B (en) | No-reference image quality evaluation method based on image entropy | |
Lei et al. | Scale insensitive and focus driven mobile screen defect detection in industry | |
Cao et al. | Collaborative discrepancy optimization for reliable image anomaly localization | |
CN110930378B (en) | Emphysema image processing method and system based on low data demand | |
CN112560966B (en) | Polarized SAR image classification method, medium and equipment based on scattering map convolution network | |
Yang et al. | A novel intelligent fault diagnosis method of rolling bearings with small samples | |
CN111667445A (en) | Image compressed sensing reconstruction method based on Attention multi-feature fusion | |
CN114170088A (en) | Relational reinforcement learning system and method based on graph structure data | |
CN113627375A (en) | Planetary gear fault diagnosis method and system, storage medium and computing device | |
Lai et al. | Industrial anomaly detection and one-class classification using generative adversarial networks | |
Wang et al. | Denoising auto-encoding priors in undecimated wavelet domain for MR image reconstruction | |
Lequyer et al. | Noise2Fast: fast self-supervised single image blind denoising | |
Zhu et al. | Deep image refinement method by hybrid training with images of varied quality in electrical capacitance tomography | |
CN110097499B (en) | Single-frame image super-resolution reconstruction method based on spectrum mixing kernel Gaussian process regression | |
Su et al. | High-Similarity-Pass Attention for Single Image Super-Resolution | |
Meng et al. | A gear fault diagnosis method based on improved accommodative random weighting algorithm and BB-1D-TP | |
CN113378472A (en) | Mixed boundary electromagnetic backscattering imaging method based on generation countermeasure network | |
CN112581367B (en) | GPR image super-resolution reconstruction method based on random sample classification amplification | |
Van Veen et al. | Scale-agnostic super-resolution in mri using feature-based coordinate networks | |
CN108492283B (en) | Hyperspectral image anomaly detection method based on band-constrained sparse representation | |
CN113298827B (en) | Image segmentation method based on DP-Net network | |
Xiang et al. | Pseudo light field image and 4D Wavelet-transform-based reduced-reference light field image quality assessment | |
Shen et al. | Graph-Represented Distribution Similarity Index for Full-Reference Image Quality Assessment | |
CN113435321A (en) | Method, system and equipment for evaluating state of main shaft bearing and readable storage medium | |
Ooi et al. | Enhanced dense space attention network for super-resolution construction from single input image |
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 | ||
CB03 | Change of inventor or designer information |
Inventor after: Zhang Jun Inventor after: Yang Mingbo Inventor after: Peng Peng Inventor after: Wu Zhenyu Inventor after: Guo Chao Inventor before: Yang Mingbo Inventor before: Peng Peng Inventor before: Wu Zhenyu Inventor before: Guo Chao Inventor before: Zhang Jun |
|
CB03 | Change of inventor or designer information | ||
GR01 | Patent grant | ||
GR01 | Patent grant |