CN109325915A - A kind of super resolution ratio reconstruction method for low resolution monitor video - Google Patents
A kind of super resolution ratio reconstruction method for low resolution monitor video Download PDFInfo
- Publication number
- CN109325915A CN109325915A CN201811056960.4A CN201811056960A CN109325915A CN 109325915 A CN109325915 A CN 109325915A CN 201811056960 A CN201811056960 A CN 201811056960A CN 109325915 A CN109325915 A CN 109325915A
- Authority
- CN
- China
- Prior art keywords
- image
- resolution
- feature
- layers
- convolutional
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 26
- 238000012544 monitoring process Methods 0.000 claims abstract description 23
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 21
- 238000012549 training Methods 0.000 claims abstract description 11
- 238000003475 lamination Methods 0.000 claims abstract description 9
- 230000003321 amplification Effects 0.000 claims description 7
- 238000003199 nucleic acid amplification method Methods 0.000 claims description 7
- 238000005457 optimization Methods 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 6
- 238000000605 extraction Methods 0.000 claims description 6
- 238000011478 gradient descent method Methods 0.000 claims description 2
- 238000011840 criminal investigation Methods 0.000 abstract description 5
- 230000000694 effects Effects 0.000 description 5
- 239000000284 extract Substances 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000011835 investigation Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000002401 inhibitory effect Effects 0.000 description 1
- 210000000056 organ Anatomy 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
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
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4046—Scaling of whole images or parts thereof, e.g. expanding or contracting using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/161—Detection; Localisation; Normalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/168—Feature extraction; Face representation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30196—Human being; Person
- G06T2207/30201—Face
Landscapes
- Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Multimedia (AREA)
- General Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of super resolution ratio reconstruction methods for low resolution monitor video, low resolution monitoring video frame is used for character representation using two various sizes of convolution kernels by this method, the feature that the two is extracted is combined as to next layer of input later, and it is more easier network training by residual error mode of learning, the reconstruction of super-resolution is carried out for the feature learnt with warp lamination, then convolutional neural networks are optimized using stochastic gradient descent algorithm, obtain the network model trained, low resolution monitoring picture to be reconstructed is input in trained network model again, it is monitored the super-resolution rebuilding of video.The present invention is not under the premise of improving hardware cost, improve the image resolution ratio of monitor video, characteristic information needed for making it possible to more obtain identification face, for assisting criminal investigation to determine suspect's identity, the accuracy rate and efficiency that suspect's identity is determined in criminal investigation are improved.
Description
Technical field
The present invention relates to computer vision methods field, specifically a kind of super-resolution for low resolution monitor video
Method for reconstructing.
Background technique
As the Chinese government is actively using advanced security and guard technology for maintaining social stability and ensureing people's lives and properties
Safety, establishes more perfect video monitoring system in National urban.Punishment of these video monitoring systems in public security organ
Important function has been played in thing investigation.But in actual monitoring, due to suspect apart from camera distance farther out or
Since imaging effect is bad in monitoring camera, cause in monitoring to be much low resolution image, it is difficult to provide needed for identification face
Characteristic information.Therefore for low resolution monitoring image carry out increase resolution processing come improve target can identification be
Starting point of the invention.
Image super-resolution rebuilding is a kind of technology of mode improving image quality by using software algorithm, it overcomes
High-definition picture disadvantage at high cost, in terms of improving image visual effect important in inhibiting are obtained by hardware.Make
Increase resolution is carried out come the monitoring image to low resolution with image super-resolution rebuilding technology, is not improving hardware cost
Under the premise of, the image resolution ratio of monitor video is improved, characteristic information needed for making it possible to more obtain identification face is used
Suspect's identity is determined in auxiliary criminal investigation.
Summary of the invention
The object of the present invention is to provide a kind of super resolution ratio reconstruction methods for low resolution monitor video, existing to solve
It is low resolution in many monitoring pictures, it is difficult to the problems such as characteristic information needed for recognizing target face is provided.
In order to achieve the above object, the technical scheme adopted by the invention is as follows:
A kind of super resolution ratio reconstruction method for low resolution monitor video, it is characterised in that: by the image of training
Feature extraction is carried out using the convolutional neural networks method connected containing convolutional layer with residual error, and passes through warp lamination for image
It rebuilds, improves photo resolution, instructed later using the optimization that stochastic gradient descent algorithm carries out convolutional neural networks
Experienced network model, then picture frame to be reconstructed is inputted into the network model trained, obtain reconstructed results；Its step is such as
Under:
(1), multiple pictures are chosen and are used as tranining database, comprising inputting the low resolution figure of network in tranining database
Picture and corresponding high-definition picture as supervised learning label；
(2), training sample is input to the training that convolutional neural networks carry out network, this convolutional neural networks includes more altogether
Layer convolutional layer is connected with multiple residual errors, wherein treatment process in convolutional layer are as follows:
First layer is 1 convolutional layer containing 3*3 size convolution kernel, for extracting the feature of image overall, layer later
For multiple parallel convolutional layers for possessing 2 different convolution kernels, for extracting different size of feature, first convolutional layer contains
Multiple sizes are the convolution kernel of 3*3, and second convolutional layer contain the convolution kernel that multiple sizes are 5*5 and be, respectively by convolution kernel with
Original image carries out discrete convolution and plus after bias term, obtains the characteristics of image after extracting by ReLU activation primitive, indicates
It is as follows:
Wherein l=1,2 ..., L represent the network number of plies, and i represents the position of pixel,Represent of image in l-1 layers
I pixel,Represent j-th of characteristics of image of h-th of convolutional layer in l layers, MjRepresent the set of all images of input, k
Convolution kernel is represented,I-th of value in j-th of convolution kernel in l layers is represented,Represent j-th of bias term in l layers.By
It include 2 parallel convolutional layers for possessing different convolution kernels in each layer in the present invention, therefore h=1,2, f (x) represent ReLU
Activation primitive is expressed as follows:
F (x)=max (0, x) (2),
Convolution merges the result of 2 parallel convolutional layers as a monolith image spy in merging layer after completing
Sign, is expressed as follows:
Wherein XLL layers of output is represented, [], which represents, merges conduct for the result of multiple parallel convolutional layers
The operation of one monolith characteristics of image.
(3), l layers of output is repeated into the calculating of the parallel-convolution layer in step (2), directly as l+1 layers of input
To the last layer for reaching network.The total L layers of convolutional layer of present networks, after L layers of convolution are complete, by L layers of output feature with
The input feature vector of first layer carries out residual error operation, is expressed as follows:
X=XL+X1 (4)
Wherein X represents the feature after residual error operation is completed.Later by X feature input into warp lamination by X feature
Size amplification, then by the feature of amplification by a convolutional layer containing 3*3 size convolution kernel, after finally obtaining increase resolution
Output image.
(4), by as the corresponding high-definition picture of supervised learning label and output image comparison, boarding steps are used
It spends descent method and optimizes convolutional neural networks, the network trained after at least 100,000 iteration；
(5), it has been known that there is a low resolution monitoring images to super-resolution rebuilding, and image to be reconstructed input progress is rapid
(4) in the network trained obtained in, by the high-resolution monitoring image after convolutional neural networks output oversubscription reconstruction.
A kind of super resolution ratio reconstruction method for low resolution monitor video, it is characterised in that: use convolution
Neural network carries out the feature extraction of low resolution monitoring image, wherein extracted not using the convolution kernel of different sizes
With the feature of size, the two feature is combined later, and using the training of the connection type of residual error optimization network, uses deconvolution
Layer is come the reconstruction for the feature learnt progress super-resolution, to obtain the reconstruction image of increase resolution.Finally use
Stochastic gradient descent method carries out the network that the network optimization has been trained, and then carries out the super-resolution of low resolution monitoring image
It rebuilds.
The present invention carries out super-resolution rebuilding by the monitoring image to low resolution, the high resolution graphics after being rebuild
Picture improves the image resolution ratio of monitor video under the premise of not improving hardware cost, makes it possible to more obtain identification
Characteristic information needed for face, for assisting criminal investigation to determine suspect's identity.
In the present invention, stochastic gradient descent algorithm is a kind of optimization algorithm, be relatively specific for control variable it is more, controlled system
The optimized control process for uniting more complicated.Target is to make the output result and correct result of network during training network
Error is minimum, by successive ignition, obtains the minimum of objective function.
The present invention carries out feature extraction and super-resolution rebuilding using the method for convolutional neural networks.This method is low from extracting
Level characteristics, which gradually arrive, extracts high-level abstract feature, and extracts difference by using the convolution kernel of different sizes
The feature of size improves the effect of reconstruction, and convolutional neural networks have to preferably be extracted effective characteristic information
Very high flexibility can carry out the adjustment of different parameters according to different actual conditions, reapply into different occasions.
The beneficial effects of the present invention are:
Super-resolution Reconstruction of the convolutional neural networks for picture is used to improve point of low resolution monitoring picture by the present invention
Resolution, so that characteristic information needed for more obtaining identification face, realizes the super-resolution rebuilding of picture applying to punishment
In thing investigation, the accuracy rate and efficiency that suspect's identity is determined in criminal investigation are improved.
Detailed description of the invention
Fig. 1 is implementation process frame diagram of the invention.
Fig. 2 is the convolutional network structure that the present invention uses.
Fig. 3 is the effect contrast figure in monitor video of the present invention.
Specific embodiment
As shown in Figure 1, a kind of super resolution ratio reconstruction method for low resolution monitor video, process are as follows:
(1), 700 pictures are chosen and is used as tranining database, comprising inputting the low resolution of network in tranining database
Image and corresponding high-definition picture as supervised learning label；
(2), training sample is input to the training that convolutional neural networks carry out network, this convolutional neural networks includes more altogether
Layer convolutional layer is connected with multiple residual errors, and convolution operation here includes the convolutional layer for possessing 2 different convolution kernels, wherein convolutional layer
Middle treatment process are as follows:
First layer is 1 convolutional layer containing 3*3 size convolution kernel, for extracting the feature of image overall, layer later
For multiple parallel convolutional layers for possessing 2 different convolution kernels, for extracting different size of feature, first convolutional layer contains
Multiple sizes are the convolution kernel of 3*3, and second convolutional layer contain the convolution kernel that multiple sizes are 5*5 and be, respectively by convolution kernel with
Original image carries out discrete convolution and plus after bias term, obtains the characteristics of image after extracting by ReLU activation primitive, indicates
It is as follows:
Wherein l=1,2 ..., L represent the network number of plies, and i represents the position of pixel,Represent of image in l-1 layers
I pixel,Represent j-th of characteristics of image of h-th of convolutional layer in l layers, MjRepresent the set of all images of input, k
Convolution kernel is represented,I-th of value in j-th of convolution kernel in l layers is represented,Represent j-th of bias term in l layers.By
It include 2 parallel convolutional layers for possessing different convolution kernels in each layer in the present invention, therefore h=1,2, f (x) represent ReLU
Activation primitive is expressed as follows:
F (x)=max (0, x) (2),
Convolution merges the result of 2 parallel convolutional layers as a monolith image spy in merging layer after completing
Sign, is expressed as follows:
Wherein XLL layers of output is represented, [], which represents, merges conduct for the result of multiple parallel convolutional layers
The operation of one monolith characteristics of image.
(3), l layers of output is repeated into the calculating of the parallel-convolution layer in step (2), directly as l+1 layers of input
To the last layer for reaching network.The total L layers of convolutional layer of present networks, after L layers of convolution are complete, by L layers of output feature with
The input feature vector of first layer carries out residual error operation, is expressed as follows:
X=XL+X1 (4)
Wherein X represents the feature after residual error operation is completed.Later by X feature input into warp lamination by X feature
Size amplification, then by the feature of amplification by a convolutional layer containing 3*3 size convolution kernel, after finally obtaining increase resolution
Output image.
(4), by as the corresponding high-definition picture of supervised learning label and output image comparison, boarding steps are used
It spends descent method and optimizes convolutional neural networks, the network trained after at least 100,000 iteration；
(5), the known one low resolution monitoring image to super-resolution rebuilding, the instruction that image to be reconstructed is inputted
In experienced network, by the high-resolution monitoring image after convolutional neural networks output oversubscription reconstruction.
It is the convolutional network structure that the present invention uses in Fig. 2, wherein the formula on the convolutional layer left side represents the size of convolution kernel,
This convolutional network structure connects to carry out feature extraction using the convolutional layer containing different size convolution kernels with residual error, wherein merging
The output feature of convolutional layer containing different size convolution kernels is merged into a monolith characteristics of image by layer, and add character represents residual error
Connection is later amplified characteristic size using warp lamination, the output image after finally obtaining increase resolution.(a) is in Fig. 3
Low resolution monitoring picture (b) is the high-resolution pictures after rebuilding, the low same size of picture differentiated after picture and reconstruction, with
Just more intuitively contrast effect, wherein the unified amplification in red frame face part is used to contrast effect.
Claims (2)
1. a kind of super resolution ratio reconstruction method for low resolution monitor video, it is characterised in that: make the image of training
Feature extraction is carried out with the convolutional neural networks method connected containing convolutional layer with residual error, and passes through warp lamination for image weight
It builds, improves photo resolution, trained later using the optimization that stochastic gradient descent algorithm carries out convolutional neural networks
Network model, then picture frame to be reconstructed is inputted into the network model trained, obtains reconstructed results；Its step is such as
Under:
(1), multiple pictures are chosen and are used as tranining database, in tranining database the low-resolution image comprising input network with
Corresponding high-definition picture as supervised learning label；
(2), training sample is input to the training that convolutional neural networks carry out network, this convolutional neural networks includes multilayer volume altogether
Lamination is connected with multiple residual errors, wherein treatment process in convolutional layer are as follows:
First layer is 1 convolutional layer containing 3*3 size convolution kernel, and for extracting the feature of image overall, layer later is more
A parallel convolutional layer for possessing 2 different convolution kernels, for extracting different size of feature, first convolutional layer contains multiple
Size is the convolution kernel of 3*3, and second convolutional layer contains the convolution kernel that multiple sizes are 5*5 and be, respectively by convolution kernel and original image
As carrying out discrete convolution and plus after bias term, the characteristics of image after extracting is obtained by ReLU activation primitive, is expressed as follows:
Wherein l=1,2 ..., L represent the network number of plies, and i represents the position of pixel,Represent i-th of picture of image in l-1 layers
Element,Represent j-th of characteristics of image of h-th of convolutional layer in l layers, MjThe set of all images of input is represented, k represents volume
Product core,I-th of value in j-th of convolution kernel in l layers is represented,Represent j-th of bias term in l layers；Due to this hair
It include 2 parallel convolutional layers for possessing different convolution kernels in each layer in bright, therefore h=1,2, f (x) represent ReLU activation letter
Number, is expressed as follows:
F (x)=max (0, x) (2),
Convolution merges the result of 2 parallel convolutional layers as a monolith characteristics of image in merging layer after completing,
It is expressed as follows:
Wherein XLL layers of output is represented, [] representative merges the result of multiple parallel convolutional layers as a monolith
The operation of characteristics of image；
(3), l layers of output is repeated into the calculating of the parallel-convolution layer in step (2), Zhi Daochuan as l+1 layers of input
To the last layer of network；The total L layers of convolutional layer of present networks, after L layers of convolution are complete, by L layers of output feature and first
The input feature vector of layer carries out residual error operation, is expressed as follows:
X=XL+X1 (4)
Wherein X represents the feature after residual error operation is completed；Later by X feature input into warp lamination by the size of X feature
Amplification, then by the feature of amplification by a convolutional layer containing 3*3 size convolution kernel, it is defeated after finally obtaining increase resolution
Image out；
(4), by as the corresponding high-definition picture of supervised learning label and output image comparison, using under stochastic gradient
Drop method optimizes convolutional neural networks, the network trained after at least 100,000 iteration；
(5), it has been known that there is a low resolution monitoring images to super-resolution rebuilding, will be in image to be reconstructed input progress rapid (4)
In the obtained network trained, by the high-resolution monitoring image after convolutional neural networks output oversubscription reconstruction.
2. a kind of super resolution ratio reconstruction method for low resolution monitor video according to claim 1, feature exist
In: the feature extraction of low resolution monitoring image is carried out using convolutional neural networks, wherein using the volume of different sizes
Core is accumulated to extract different size of feature, later combines the two feature, and uses the instruction of the on-link mode (OLM) of residual error optimization network
Practice, the reconstruction of super-resolution is carried out for the feature learnt using warp lamination, to obtain the reconstruction of increase resolution
Image；The network that the network optimization has been trained finally is carried out using stochastic gradient descent method, and then carries out low resolution monitoring
The super-resolution rebuilding of image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811056960.4A CN109325915B (en) | 2018-09-11 | 2018-09-11 | Super-resolution reconstruction method for low-resolution monitoring video |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811056960.4A CN109325915B (en) | 2018-09-11 | 2018-09-11 | Super-resolution reconstruction method for low-resolution monitoring video |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109325915A true CN109325915A (en) | 2019-02-12 |
| CN109325915B CN109325915B (en) | 2022-11-08 |
Family
ID=65264816
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811056960.4A Active CN109325915B (en) | 2018-09-11 | 2018-09-11 | Super-resolution reconstruction method for low-resolution monitoring video |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109325915B (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110647936A (en) * | 2019-09-20 | 2020-01-03 | 北京百度网讯科技有限公司 | Training method and device for video super-resolution reconstruction model and electronic equipment |
| CN111062867A (en) * | 2019-11-21 | 2020-04-24 | 浙江大华技术股份有限公司 | Video super-resolution reconstruction method |
| CN111915492A (en) * | 2020-08-19 | 2020-11-10 | 四川省人工智能研究院(宜宾) | Multi-branch video super-resolution method and system based on dynamic reconstruction |
| WO2020248706A1 (en) * | 2019-06-14 | 2020-12-17 | 深圳市中兴微电子技术有限公司 | Image processing method, device, computer storage medium, and terminal |
| CN112200152A (en) * | 2019-12-06 | 2021-01-08 | 中央广播电视总台 | Super-resolution method for aligning face images based on residual back-projection neural network |
| CN112580502A (en) * | 2020-12-17 | 2021-03-30 | 南京航空航天大学 | SICNN-based low-quality video face recognition method |
| CN113408347A (en) * | 2021-05-14 | 2021-09-17 | 桂林电子科技大学 | Method for detecting change of remote building by monitoring camera |
| CN113869282A (en) * | 2021-10-22 | 2021-12-31 | 马上消费金融股份有限公司 | Face recognition method, hyper-resolution model training method and related equipment |
| CN114500948A (en) * | 2022-01-25 | 2022-05-13 | 重庆卡佐科技有限公司 | Vehicle monitoring method, monitoring system and vehicle-mounted terminal |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016019484A1 (en) * | 2014-08-08 | 2016-02-11 | Xiaoou Tang | An apparatus and a method for providing super-resolution of a low-resolution image |
| CN106683067A (en) * | 2017-01-20 | 2017-05-17 | 福建帝视信息科技有限公司 | Deep learning super-resolution reconstruction method based on residual sub-images |
| CN107578377A (en) * | 2017-08-31 | 2018-01-12 | 北京飞搜科技有限公司 | A kind of super-resolution image reconstruction method and system based on deep learning |
| EP3319039A1 (en) * | 2016-11-07 | 2018-05-09 | UMBO CV Inc. | A method and system for providing high resolution image through super-resolution reconstruction |
-
2018
- 2018-09-11 CN CN201811056960.4A patent/CN109325915B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016019484A1 (en) * | 2014-08-08 | 2016-02-11 | Xiaoou Tang | An apparatus and a method for providing super-resolution of a low-resolution image |
| EP3319039A1 (en) * | 2016-11-07 | 2018-05-09 | UMBO CV Inc. | A method and system for providing high resolution image through super-resolution reconstruction |
| CN106683067A (en) * | 2017-01-20 | 2017-05-17 | 福建帝视信息科技有限公司 | Deep learning super-resolution reconstruction method based on residual sub-images |
| CN107578377A (en) * | 2017-08-31 | 2018-01-12 | 北京飞搜科技有限公司 | A kind of super-resolution image reconstruction method and system based on deep learning |
Non-Patent Citations (1)
| Title |
|---|
| 王爱丽等: "基于深度卷积神经网络的遥感图像超分辨率重建", 《黑龙江大学自然科学学报》 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020248706A1 (en) * | 2019-06-14 | 2020-12-17 | 深圳市中兴微电子技术有限公司 | Image processing method, device, computer storage medium, and terminal |
| CN110647936A (en) * | 2019-09-20 | 2020-01-03 | 北京百度网讯科技有限公司 | Training method and device for video super-resolution reconstruction model and electronic equipment |
| CN111062867A (en) * | 2019-11-21 | 2020-04-24 | 浙江大华技术股份有限公司 | Video super-resolution reconstruction method |
| CN112200152A (en) * | 2019-12-06 | 2021-01-08 | 中央广播电视总台 | Super-resolution method for aligning face images based on residual back-projection neural network |
| CN112200152B (en) * | 2019-12-06 | 2024-04-26 | 中央广播电视总台 | Super-resolution method for aligning face images based on residual back projection neural network |
| CN111915492A (en) * | 2020-08-19 | 2020-11-10 | 四川省人工智能研究院(宜宾) | Multi-branch video super-resolution method and system based on dynamic reconstruction |
| CN112580502A (en) * | 2020-12-17 | 2021-03-30 | 南京航空航天大学 | SICNN-based low-quality video face recognition method |
| CN113408347A (en) * | 2021-05-14 | 2021-09-17 | 桂林电子科技大学 | Method for detecting change of remote building by monitoring camera |
| CN113408347B (en) * | 2021-05-14 | 2022-03-15 | 桂林电子科技大学 | Method for detecting change of remote building by monitoring camera |
| CN113869282A (en) * | 2021-10-22 | 2021-12-31 | 马上消费金融股份有限公司 | Face recognition method, hyper-resolution model training method and related equipment |
| CN114500948A (en) * | 2022-01-25 | 2022-05-13 | 重庆卡佐科技有限公司 | Vehicle monitoring method, monitoring system and vehicle-mounted terminal |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109325915B (en) | 2022-11-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109325915A (en) | A kind of super resolution ratio reconstruction method for low resolution monitor video | |
| Wang et al. | Multistage attention network for image inpainting | |
| CN111062872B (en) | Image super-resolution reconstruction method and system based on edge detection | |
| CN109615582B (en) | Face image super-resolution reconstruction method for generating countermeasure network based on attribute description | |
| CN104978580B (en) | A kind of insulator recognition methods for unmanned plane inspection transmission line of electricity | |
| CN111861961B (en) | Single image super-resolution multi-scale residual error fusion model and restoration method thereof | |
| CN105069746B (en) | Video real-time face replacement method and its system based on local affine invariant and color transfer technology | |
| CN111127308B (en) | Mirror image feature rearrangement restoration method for single sample face recognition under partial shielding | |
| CN107977932A (en) | It is a kind of based on can differentiate attribute constraint generation confrontation network face image super-resolution reconstruction method | |
| CN109741256A (en) | Image super-resolution rebuilding method based on rarefaction representation and deep learning | |
| CN108961186A (en) | A kind of old film reparation recasting method based on deep learning | |
| CN107154023A (en) | Face super-resolution reconstruction method based on generation confrontation network and sub-pix convolution | |
| CN107220957B (en) | It is a kind of to utilize the remote sensing image fusion method for rolling Steerable filter | |
| CN111340696B (en) | Convolutional neural network image super-resolution reconstruction method fused with bionic visual mechanism | |
| CN109801232A (en) | A kind of single image to the fog method based on deep learning | |
| CN113628261B (en) | Infrared and visible light image registration method in electric power inspection scene | |
| CN114897742B (en) | Image restoration method with texture and structural features fused twice | |
| CN110490807A (en) | Image rebuilding method, device and storage medium | |
| CN113837956A (en) | Method for detecting unpaired supervision cloud and removing thick cloud in large area | |
| CN113822805A (en) | Image super-resolution reconstruction method and Chinese medicinal plant leaf disease diagnosis method and equipment | |
| CN115731597A (en) | Automatic segmentation and restoration management platform and method for mask image of face mask | |
| CN117391938B (en) | Infrared image super-resolution reconstruction method, system, equipment and terminal | |
| CN116664677B (en) | Sight estimation method based on super-resolution reconstruction | |
| CN108764287A (en) | Object detection method and system based on deep learning and grouping convolution | |
| CN117315336A (en) | Pollen particle identification method, device, electronic equipment and storage medium |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |