CN114937163A - Neural network image block reconstruction method based on clustering - Google Patents

Neural network image block reconstruction method based on clustering Download PDF

Info

Publication number
CN114937163A
CN114937163A CN202210552717.1A CN202210552717A CN114937163A CN 114937163 A CN114937163 A CN 114937163A CN 202210552717 A CN202210552717 A CN 202210552717A CN 114937163 A CN114937163 A CN 114937163A
Authority
CN
China
Prior art keywords
image
image block
neural network
images
clustering
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
Application number
CN202210552717.1A
Other languages
Chinese (zh)
Inventor
刘福涛
尹晓敏
曹维达
于依冉
鲁威志
张伟
王付奎
姜秋波
李振玲
万月忠
杨超尘
徐彪
李云龙
陈芳
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
State Grid Corp of China SGCC
Liaocheng Power Supply Co of State Grid Shandong Electric Power Co Ltd
Original Assignee
State Grid Corp of China SGCC
Liaocheng Power Supply Co of State Grid Shandong Electric Power Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by State Grid Corp of China SGCC, Liaocheng Power Supply Co of State Grid Shandong Electric Power Co Ltd filed Critical State Grid Corp of China SGCC
Priority to CN202210552717.1A priority Critical patent/CN114937163A/en
Publication of CN114937163A publication Critical patent/CN114937163A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social 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
    • G06N3/04—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
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/10—Image acquisition
    • G06V10/16—Image acquisition using multiple overlapping images; Image stitching
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Multimedia (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Databases & Information Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • General Engineering & Computer Science (AREA)
  • Molecular Biology (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Image Processing (AREA)

Abstract

The invention relates to a neural network image blocking reconstruction method based on clustering, which comprises the following steps of firstly, simply reconstructing a preprocessed image, and respectively blocking the preprocessed image and the reconstructed image to obtain paired image blocks at corresponding positions; clustering the image blocks based on the K mean value; and training a refined convolutional neural network for each type of image block, taking the reconstructed image block as input, and taking the target output as an estimated value of the corresponding original image block. Splicing the estimated values of the image blocks back to the original positions of the corresponding images, and finally performing noise reduction processing on the images by using a noise remover to obtain the reconstructed images. The method realizes the method for reconstructing the image block by using a plurality of refined neural networks, performs block reconstruction on the whole image, and improves the reconstruction precision.

Description

Neural network image block reconstruction method based on clustering
Technical Field
The invention relates to a neural network image blocking reconstruction method based on clustering, and belongs to the technical field of image processing.
Background
The compressed sensing has important application in the field of image reconstruction, and the theory breaks through the limit of the traditional Nyquist sampling theorem, so that the image signal is compressed while being sampled, and the original signal is accurately reconstructed by a small number of sampling points. Therefore, the method has great advantages in storage, transmission, analysis and processing of images, and becomes a research hotspot in recent years.
On the basis, Lu Gan and the like put forward a blocking compression sensing theory for image signals, carry out blocking operation on the images, and independently process each image block, so that the required storage space is reduced, meanwhile, a coding end does not need to wait for the whole image to be coded after observation is finished, coding transmission can be carried out after the image blocks are projected to an observation matrix, and the time for data sampling and reconstruction is shortened. In addition, the size of the observation matrix corresponding to the blocked image is reduced, and the calculation complexity is reduced.
The existing image block reconstruction method is to input all image blocks into the same neural network, and the difference between the image blocks is not reflected.
With the rapid development of deep learning in recent years, convolutional neural networks become an important component of deep learning and are widely applied in the visual field such as image processing. Therefore, on the basis of the traditional compressed sensing method, the deep learning framework is combined, the image reconstruction and the convolutional neural network are combined, and the efficiency and the precision of the image reconstruction are improved.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a neural network image block reconstruction method based on clustering;
the invention combines the image block reconstruction technology with the clustering method, and can better utilize the difference of image structure information and image block differences to realize the classification reconstruction of the image blocks.
Interpretation of terms:
partitioning: the image is divided into m x n non-overlapping sub-areas, m and n are the number of the image horizontal and vertical partitions respectively, and each sub-area represents an image block.
The technical scheme of the invention is as follows:
a neural network image blocking reconstruction method based on clustering comprises the following steps:
1) image preprocessing: performing graying processing on the input m images, and adjusting the pixel size to p1 × p 1; p1 is a multiple of 32;
2) simply reconstructing the m images preprocessed in the step 1) to obtain reconstructed images;
3) respectively blocking the m preprocessed images in the step 1) and the reconstructed images obtained in the step 2) to obtain paired image blocks at corresponding positions;
the pixel size of each image block is set as p2 × p2, and the value of p2 is 32; generating (p1/p2) ^2 image blocks for each image, then generating m (p1/p2) ^2 image blocks for m images, partitioning the m images preprocessed in the step 1) to obtain an original image block, marking the original image block as C1, and partitioning the reconstructed image obtained in the step 2) to obtain a reconstructed image block, marking the reconstructed image block as C2;
4) for each image block x in C1 i Performing K-means clustering, i is more than or equal to 1 and less than or equal to m (p1/p2) ^2, x i E to C1, wherein the total number of the classes is k, each image block is distributed into a subclass, and the clustering result of the ith image block is n i ，1≤n i ≤k，1≤i≤m*(p1/p2)^2；
5) Training a refined convolutional neural network on the basis of k classes into which the image block C1 is divided, and obtaining k trained convolutional neural networks; by image blocks
Figure BDA0003651171440000021
As input, the target output is
Figure BDA0003651171440000022
(as x) i E C1 estimate);
6) the test was performed using n images: sequentially carrying out image preprocessing in step 1), simple reconstruction in step 2) and blocking processing in step 3) on the n images, and storing obtained image blocks in a list, wherein the index sequence number of each image block is represented by (0, 0), (0, 1), (1.), (i, j), (1., (n-1, m-1);
7) judging which category of the k categories each image block in the list in the step 6) belongs to by using a k nearest neighbor algorithm, inputting the image block into the trained convolutional neural network corresponding to the category, splicing the image blocks generated by the trained convolutional neural network back to the original position by using the index sequence number of the image block in the list, and performing noise reduction processing on the image by using a noise remover to obtain a synthetic image. And calculating the average peak signal-to-noise ratio (PSNR) of the generated synthesized image, and evaluating the quality of an image after being compressed compared with the original image by using the PSNR, wherein the higher the PSNR is, the smaller the distortion after being compressed is.
According to the invention, the specific implementation process of step 2) is as follows:
a. setting the size of a matrix phi as M × N, M < < N, wherein the matrix phi is formed by 0, 1 and-1 randomly;
b. performing Discrete cosine transform (Discrete cosine transform-DCT) on any one of the m images z preprocessed in the step 1) to obtain: s ═ dct (z); DCT is a method for image processing, and is a block transform method for image data compression.
c. Compressing observation, and solving an observation vector: y Φ ═ s Φ × dct (z), where y is an observation vector M × 1;
d. and (3) reconstructing to obtain:
Figure BDA0003651171440000023
e. to pair
Figure BDA0003651171440000024
Performing inverse discrete cosine transform to obtain:
Figure BDA0003651171440000025
Figure BDA0003651171440000026
is the reconstructed image.
According to the invention, step 4) is preferably implemented as follows:
f. randomly selecting k image blocks from all the image blocks as initial clustering centers;
g. calculating the Euclidean distance from each image block to each clustering center, and classifying each image block into the category corresponding to the minimum Euclidean distance;
h. after all the image blocks are classified once, taking the mean value of all the current image blocks as a new clustering center for each class;
i. and g, repeating the step g and the step h until the clustering center is not changed any more or the set iteration times are reached.
Preferably according to the invention, the convolutional neural network comprises three convolutional layers, the first convolutional layer comprising 64 filters, each filter comprising 1 channel of size 11 x 11; the second convolutional layer comprises 32 filters, each filter comprising 64 channels with a size of 11 × 11; the third convolutional layer comprises 1 filter comprising 32 channels of size 11 x 11;
each convolution layer applies Relu function to the output of the convolution layer to carry out nonlinear activation;
using the mean square error as a loss function, as shown in equation (i):
Figure BDA0003651171440000031
in the formula (I), l is the number of image blocks in each class;
the weights and bias values for each neuron are updated using a back-propagation algorithm.
The initial reconstructed image blocks in each class
Figure BDA0003651171440000032
Inputting corresponding convolutional neural network, and outputting the target as
Figure BDA0003651171440000033
And calculating a loss function value corresponding to the convolutional neural network, updating the weight and the offset value of each neuron by using a back propagation algorithm, repeating iteration until the loss function value is less than a threshold value, and finishing training.
A computer device comprising a memory storing a computer program and a processor implementing the steps of a cluster-based neural network image block reconstruction method when executing the computer program.
A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of a cluster-based neural network image block reconstruction method.
The invention has the beneficial effects that:
1. according to the image block classification and reconstruction method, the image blocks are classified and respectively reconstructed according to different structural information of the image blocks corresponding to different positions of the image, and finally splicing and denoising are carried out, so that the accuracy degree of the reconstructed image is higher, and the image block classification and reconstruction can be better realized by utilizing the difference of the image structural information and the image block difference.
2. The invention carries out two times of reconstruction, wherein one time is the initial reconstruction of the whole image; and once, based on the nonlinear block depth reconstruction of the initial reconstructed image, the initial reconstructed image is further refined, so that the accuracy of the reconstructed image is higher.
3. The sampling matrix for initial reconstruction is a ternary matrix composed of { -1, 0, 1}, and compared with the traditional sampling matrix, the complexity of operation and the cost of storage are greatly reduced, and the sampling matrix can be expanded to be applied to deep reconstruction networks with various structures in the future, so that the application range is expanded.
Drawings
FIG. 1 is a block flow diagram of a neural network image block reconstruction method based on clustering proposed by the present invention;
fig. 2 is a schematic diagram of the structure of the convolutional neural network of the present invention.
Detailed Description
The invention is further defined in the following description, without being limited thereto, by reference to the drawings and examples.
Example 1
A neural network image block reconstruction method based on clustering is disclosed, as shown in FIG. 1, and comprises the following steps:
1) image preprocessing: 100 images are selected, wherein 90 images are used as a training set, 10 images are used as a test set, and the training set images are all subjected to graying processing, and in the test, if the images are color images, the images can be divided into R, G, B channels to be sequentially and independently tested. And adjusting the pixel size of the image to a multiple of 32;
2) simply reconstructing the 90 preprocessed images in the step 1) to obtain reconstructed images; the specific implementation process of the step 2) is as follows:
a. setting the size of a matrix phi as M × N, M < < N, wherein the matrix phi is randomly composed of 0, 1 and-1, and the compression rate is generally 0.25, 0.10, 0.04 or 0.01;
b. performing Discrete Cosine Transform (Discrete Cosine Transform-DCT) on any image z in the m images preprocessed in the step 1): s ═ dct (z);
s, DCT is a method for image processing, which is a block transform method for image data compression.
c. Compressing observation, and solving an observation vector: y Φ ═ s Φ × dct (z), where y is an observation vector M × 1;
d. and (3) reconstructing to obtain:
Figure BDA0003651171440000041
e. to pair
Figure BDA0003651171440000042
Performing inverse discrete cosine transform to obtain:
Figure BDA0003651171440000043
Figure BDA0003651171440000044
is the reconstructed image.
3) Respectively blocking the 90 preprocessed images in the step 1) and the 90 reconstructed images obtained in the step 2) to obtain paired image blocks at corresponding positions; setting the pixel size of each image block as 32 × 32, marking all original image blocks as C1, and marking all reconstructed image blocks as C2;
4) for each of C1An image block x i Performing K-means clustering, x i The image block belongs to C1, the total number of the classes is set to be 3, each image block is distributed into one subclass, and the clustering result of the ith image block is n i ，1≤n i Less than or equal to 3; the implementation process of the step 4) is as follows:
f. randomly selecting 3 image blocks from all the image blocks as initial clustering centers;
g. calculating the Euclidean distance from each image block to each clustering center, and classifying each image block into the category corresponding to the minimum Euclidean distance;
h. after all the image blocks are classified once, taking the mean value of all the current image blocks as a new clustering center for each class;
i. and g, repeating the step g and the step h until the clustering center is not changed any more or the set iteration times are reached.
5) Based on 3 types divided by the image block C1, training a refined convolutional neural network in each type to obtain 3 trained convolutional neural networks; by image blocks
Figure BDA0003651171440000045
As input, the target output is
Figure BDA0003651171440000046
(as x) i E.c 1);
6) the test was performed using 10 images: sequentially performing image preprocessing of the step 1), simple reconstruction of the step 2) and blocking processing of the step 3) on the 10 images, and storing obtained image blocks in a list, wherein index sequence numbers of each image block are (0, 0), (0, 1), (1.), (i, j), (1., (n-1, m-1);
7) judging which category of the 3 categories each image block in the list in the step 6) belongs to by using a k-nearest neighbor algorithm, inputting the image block into the trained convolutional neural network corresponding to the category, splicing the image blocks generated by the trained convolutional neural network back to the original position by using the index sequence number of the image block in the list, and performing noise reduction processing on the image by using a noise remover to obtain a synthetic image. And calculating the average peak signal-to-noise ratio (PSNR) of the generated synthetic image, and evaluating the quality of an image after being compressed compared with the original image by using the PSNR, wherein the higher the PSNR is, the smaller the distortion after being compressed is.
As shown in fig. 2, the convolutional neural network includes three convolutional layers as follows: the first layer had 64 filters, each filter having 1 channel of size 11 x 11, creating 64 signatures of size 32 x 32; the second layer has 32 filters, each filter has 64 channels with size 11 x 11, and 32 characteristic maps with size 32 x 32 are generated; the third layer has 1 filter with 32 channels of size 11 x 11, generating reconstructed image blocks of size 32 x 32. The initialized weight distribution is subjected to a gaussian distribution with a mean value of 0 and a variance of 0.01, and the initial values of the biases are all set to 0.
Each convolution layer applies Relu function to the output of the convolution layer to carry out nonlinear activation;
using the mean square error as a loss function, as shown in equation (i):
Figure BDA0003651171440000051
in the formula (I), l is the number of image blocks in each class;
the weights and bias values for each neuron are updated using a back-propagation algorithm.
The initial reconstructed image blocks in each class
Figure BDA0003651171440000052
Inputting corresponding convolutional neural network, and outputting the target as
Figure BDA0003651171440000053
And calculating a loss function value corresponding to the convolutional neural network, updating the weight and the offset value of each neuron by using a back propagation algorithm, repeating iteration until the loss function value is smaller than a threshold value, and finishing training.
Example 2
A computer device comprising a memory storing a computer program and a processor implementing the steps of the cluster-based neural network image block reconstruction method of embodiment 1 when the processor executes the computer program.
Example 3
A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of the cluster-based neural network image block reconstruction method of embodiment 1.

Claims (6)

1. A neural network image block reconstruction method based on clustering is characterized by comprising the following steps:
1) image preprocessing: performing graying processing on the input m images, and adjusting the pixel size to p1 × p 1;
2) simply reconstructing the m images preprocessed in the step 1) to obtain reconstructed images;
3) respectively blocking the m preprocessed images in the step 1) and the reconstructed images obtained in the step 2) to obtain paired image blocks at corresponding positions;
setting the pixel size of each image block as p2 × p2, generating (p1/p2) ^2 image blocks for each image, then generating m × (p1/p2) ^2 image blocks for m images, blocking the m images preprocessed in the step 1) to obtain an original image block, marking the original image block as C1, and blocking the reconstructed image obtained in the step 2) to obtain a reconstructed image block, marking the reconstructed image block as C2;
4) for each image block x in C1 i Performing K-means clustering, i is more than or equal to 1 and less than or equal to m (p1/p2) ^2, x i E to C1, wherein the total number of the classes is k, each image block is distributed into a subclass, and the clustering result of the ith image block is n i ，1≤n i ≤k，1≤i≤m*(p1/p2)^2；
5) Training a refined convolutional neural network on the basis of k classes into which the image block C1 is divided, and obtaining k trained convolutional neural networks; by image blocks
Figure FDA0003651171430000011
As input, the target output is
Figure FDA0003651171430000012
6) The test was performed using n images: sequentially carrying out image preprocessing in step 1), simple reconstruction in step 2) and blocking processing in step 3) on the n images, and storing obtained image blocks in a list, wherein the index sequence number of each image block is represented by (0, 0), (0, 1), (1.), (i, j), (1., (n-1, m-1);
7) judging which category of the k categories each image block in the list in the step 6) belongs to by using a k-nearest neighbor algorithm, inputting the image block into the trained convolutional neural network corresponding to the category, splicing the image blocks generated by the trained convolutional neural network back to the original positions by using the index sequence number of the image block in the list, and performing noise reduction processing on the image to obtain a synthetic image.
2. The neural network image block reconstruction method based on clustering according to claim 1, wherein the specific implementation process of step 2) is as follows:
a. setting the size of a matrix phi as M × N, M < < N, wherein the matrix phi is formed by 0, 1 and-1 randomly;
b. performing discrete cosine transform on any image z in the m images preprocessed in the step 1): s ═ dct (z);
c. compressing observation, and solving an observation vector: y Φ ═ s Φ × dct (z), where y is an observation vector M × 1;
d. and (3) reconstructing to obtain:
Figure FDA0003651171430000013
e. to pair
Figure FDA0003651171430000014
Performing inverse discrete cosine transform to obtain:
Figure FDA0003651171430000015
Figure FDA0003651171430000016
is the reconstructed image.
3. The method for reconstructing image blocks based on clustering according to claim 1, wherein the step 4) is implemented as follows:
f. randomly selecting k image blocks from all the image blocks as initial clustering centers;
g. calculating the Euclidean distance from each image block to each clustering center, and classifying each image block into the category corresponding to the minimum Euclidean distance;
h. after all the image blocks are classified once, taking the mean value of all the current image blocks as a new clustering center for each class;
i. and g, repeating the step g and the step h until the clustering center is not changed any more or the set iteration times are reached.
4. The method of claim 1, wherein the convolutional neural network comprises three convolutional layers, the first convolutional layer comprises 64 filters, each filter comprises 1 channel with a size of 11 × 11; the second convolutional layer comprises 32 filters, each filter comprising 64 channels of size 11 x 11; the third convolutional layer comprises 1 filter comprising 32 channels with a size of 11 × 11;
each convolution layer applies Relu function to the output of the convolution layer to carry out nonlinear activation;
using the mean square error as a loss function, as shown in equation (i):
Figure FDA0003651171430000021
in the formula (I), l is the number of image blocks in each class;
updating the weight and the bias value of each neuron by using a back propagation algorithm;
initial in each classReconstructing image blocks
Figure FDA0003651171430000022
Inputting corresponding convolutional neural network, and outputting the target as
Figure FDA0003651171430000023
And calculating a loss function value corresponding to the convolutional neural network, updating the weight and the offset value of each neuron by using a back propagation algorithm, repeating iteration until the loss function value is less than a threshold value, and finishing training.
5. A computer device comprising a memory and a processor, the memory storing a computer program, wherein the processor when executing the computer program implements the steps of the cluster-based neural network image block reconstruction method of any one of claims 1-4.
6. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method for cluster-based neural network image block reconstruction according to any one of claims 1 to 4.
CN202210552717.1A 2022-05-19 2022-05-19 Neural network image block reconstruction method based on clustering Pending CN114937163A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210552717.1A CN114937163A (en) 2022-05-19 2022-05-19 Neural network image block reconstruction method based on clustering

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210552717.1A CN114937163A (en) 2022-05-19 2022-05-19 Neural network image block reconstruction method based on clustering

Publications (1)

Publication Number Publication Date
CN114937163A true CN114937163A (en) 2022-08-23

Family

ID=82864193

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210552717.1A Pending CN114937163A (en) 2022-05-19 2022-05-19 Neural network image block reconstruction method based on clustering

Country Status (1)

Country Link
CN (1) CN114937163A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118097724A (en) * 2024-04-23 2024-05-28 江西百胜智能科技股份有限公司 Palm vein-based identity recognition method and device, readable storage medium and equipment

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118097724A (en) * 2024-04-23 2024-05-28 江西百胜智能科技股份有限公司 Palm vein-based identity recognition method and device, readable storage medium and equipment

Similar Documents

Publication Publication Date Title
CN109360156B (en) Single image rain removing method based on image block generation countermeasure network
CN106683048B (en) Image super-resolution method and device
Zhang et al. Adaptive residual networks for high-quality image restoration
Starovoytov et al. Comparative analysis of the SSIM index and the pearson coefficient as a criterion for image similarity
CN109003265B (en) No-reference image quality objective evaluation method based on Bayesian compressed sensing
Luo et al. Anti-forensics of JPEG compression using generative adversarial networks
Chen et al. Remote sensing image quality evaluation based on deep support value learning networks
CN110874563A (en) Method and apparatus for providing integrated feature maps through multiple image outputs of CNN
CN110276726A (en) A kind of image deblurring method based on the guidance of multichannel network prior information
CN112699899A (en) Hyperspectral image feature extraction method based on generation countermeasure network
CN114998958B (en) Face recognition method based on lightweight convolutional neural network
CN109949200B (en) Filter subset selection and CNN-based steganalysis framework construction method
CN111415323B (en) Image detection method and device and neural network training method and device
CN112183742A (en) Neural network hybrid quantization method based on progressive quantization and Hessian information
CN112967210A (en) Unmanned aerial vehicle image denoising method based on full convolution twin network
CN115880158A (en) Blind image super-resolution reconstruction method and system based on variational self-coding
CN114937163A (en) Neural network image block reconstruction method based on clustering
CN114283058A (en) Image super-resolution reconstruction method based on countermeasure network and maximum mutual information optimization
CN110503157B (en) Image steganalysis method of multitask convolution neural network based on fine-grained image
CN111008930A (en) Fabric image super-resolution reconstruction method
CN115760603A (en) Interference array broadband imaging method based on big data technology
Hernandez et al. Classification of color textures with random field models and neural networks
CN114581539A (en) Compressed sensing image reconstruction method, device, storage medium and system
CN110443755B (en) Image super-resolution method based on high-low frequency signal quantity
CN110827238A (en) Improved side-scan sonar image feature extraction method of full convolution neural network

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