CN107958246A - A kind of image alignment method based on new end-to-end human face super-resolution network - Google Patents
A kind of image alignment method based on new end-to-end human face super-resolution network Download PDFInfo
- Publication number
- CN107958246A CN107958246A CN201810045095.7A CN201810045095A CN107958246A CN 107958246 A CN107958246 A CN 107958246A CN 201810045095 A CN201810045095 A CN 201810045095A CN 107958246 A CN107958246 A CN 107958246A
- Authority
- CN
- China
- Prior art keywords
- network
- image
- resolution
- super
- human face
- 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.)
- Withdrawn
Links
Classifications
-
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformation in the plane of the image
- G06T3/40—Scaling the whole image or part thereof
- G06T3/4007—Interpolation-based scaling, e.g. bilinear interpolation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformation in the plane of the image
- G06T3/40—Scaling the whole image or part thereof
- G06T3/4053—Super resolution, i.e. output image resolution higher than sensor resolution
Abstract
The present invention proposes a kind of image alignment method based on new end-to-end human face super-resolution network, its main contents includes：Human face super-resolution network, human face super-resolution generation confrontation network, its process is, coarseness super-resolution (SR) network is constructed to recover the high-resolution of coarseness (HR) image, coarseness HR images are separately sent in fine granularity SR encoders and prior information estimation network, fine granularity SR encoders are used for extracting characteristics of image, prior information estimation network is used for estimating key point hotspot graph, recover HR images using characteristics of image and prior information in a decoder, in order to further produce face true to nature, propose human face super-resolution generation confrontation network, and human face super-resolution network is included into confrontation loss.The present invention carries out image alignment using the network being made of human face super-resolution network and human face super-resolution generation confrontation network, can effectively improve image quality, simplify image alignment method.
Description
Technical field
The present invention relates to image processing field, and new end-to-end human face super-resolution network is based on more particularly, to a kind of
Image alignment method.
Background technology
Image alignment is a basic problem in image procossing, is widely used in public security, military affairs are investigated, at medical image
The various fields such as reason.Specifically, in police field, the monitoring image collected, extraction two width prison can be handled in case investigation
The respective characteristic point of image is controlled, is alignd to the feature point set of two images, so as to fulfill image mosaic and target identification,
A suspect effectively in identification monitoring, further enhances security protection ability.In military surveillance field, in face of complicated military ring
Border, such as wild environment operation, the investigation of aerial unmanned plane, using image alignment, can improve the identification of enemy in Reconnaissance system
Rate and the precision and reliability of tracking.And in field of medical image processing, it is that lesion detection, lesion positioning, haemocyte are micro-
The medical image analysis such as image classification are provided convenience.Although the research to human face super-resolution is a lot of, due to face
Complexity, existing method are largely trained using the multistage rather than end to end so that and method is excessively cumbersome and complicated, because
This will further simplify research method, even there are certain challenge on the premise of super-resolution image quality is ensured.
The present invention proposes a kind of image alignment method based on new end-to-end human face super-resolution network, constructs first
One coarseness super-resolution (SR) network recovers coarseness high-resolution (HR) image, then distinguishes coarseness HR images
It is sent in fine granularity SR encoders and prior information estimation network, fine granularity SR encoders are used for extracting characteristics of image, priori
Information estimation network is used for estimating and parsing key point hotspot graph, is believed in fine granularity SR decoders using characteristics of image and priori
Breath recovers HR images.In order to further produce face true to nature, human face super-resolution generation confrontation network is proposed, and will be to damage-retardation
Human face super-resolution network is included in mistake.The present invention proposes a kind of image pair based on new end-to-end human face super-resolution network
Neat method, image pair is carried out using the network being made of human face super-resolution network and human face super-resolution generation confrontation network
Together, image quality can be effectively improved, simplifies image alignment method.
The content of the invention
For image alignment, the present invention proposes a kind of image alignment based on new end-to-end human face super-resolution network
Method, constructs coarseness super-resolution (SR) network to recover the high-resolution of coarseness (HR) image, by coarseness HR
Image is separately sent in fine granularity SR encoders and prior information estimation network, and fine granularity SR encoders are used for extracting image spy
Sign, prior information estimation network are used for estimating key point hotspot graph, are recovered in a decoder using characteristics of image and prior information
HR images, in order to further produce face true to nature, it is proposed that human face super-resolution generation confrontation network, and confrontation loss is received
Enter human face super-resolution network.
To solve the above problems, propose a kind of image alignment side based on new end-to-end human face super-resolution network
Method, its main contents include：
(1) human face super-resolution network；
(2) human face super-resolution generation confrontation network.
Wherein, the human face super-resolution network, human face super-resolution network (FSRNet) is by coarseness super-resolution
Network and fine granularity super-resolution network composition, it includes fine granularity SR encoders, prior estimate network and fine granularity SR solutions
Code device, low resolution input picture is represented with x, represents high-resolution (HR) image by FSRNet recoveries with y, p represents logical
The prior information of FSRNet estimations is crossed, since the low-resolution image clarity inputted in prior estimate is relatively low, therefore constructs one
A coarseness SR networks recover coarseness SR images, are given by below equation：
Wherein, x is low resolution (LR) image, ycIt is coarseness SR images,It is coarseness super-resolution network from image
X to image ycMapping, ycIt is respectively sent to prior estimate networkWith fine granularity SR encodersOn, obtain below equation：
Wherein, f is by encoderThe feature of extraction, after coding is completed, utilizes the decoder of SRConnection figure picture is special
F and prior information p is levied to recover SR images, is expressed by the following equation：
Give a training setWherein N is the quantity of institute's training image, x(i)It is super positioned at calibration
Low-resolution image in image in different resolution,It is the prior information of calibration, the loss function of FSRNet is given by：
Wherein, Θ represent parameter set, λ be priori loss weight, y(i)And p(i)Be recover HR images, p(i)It is i-th
The prior information of a Image estimation.
Further, the coarseness super-resolution network, probably recovers coarseness HR figures using coarseness SR networks
Picture, mitigates the difficulty of estimation prior information with this, coarseness SR network architecture is by 3 × 3 convolutional layer, followed by 3
A residual error module, finally rebuilds coarseness HR images using another 3 × 3 convolutional layer.
Further, the fine granularity super-resolution network, in fine granularity SR networks, coarseness HR images are sent
Into prior estimate network and fine granular encoded device network, it is respectively intended to carry out facial prior information estimation and feature extraction, so
Decoder recovers fine granularity HR images using the two results afterwards.
Further, the prior estimate network, estimates facial markers using hourglass configuration in prior estimate network
The crucial figure and analysis diagram of point, since the two prior informations represent the face shape of two dimension, therefore in prior estimate network, are removed
Outside last layer, whole features between the two tasks are shared, in order to merge feature effectively across scale and preserve different rulers
The spatial information of degree, hourglass module, using jump connection mechanism, then handle institute between symmetrical layers with 1 × 1 convolutional layer
The feature of acquisition, the feature being shared connect two single 1 × 1 convolutional layers, generate key point hotspot graph and analysis diagram.
Further, the fine granular encoded device, for fine granularity SR encoders, carries out feature using residual error module and carries
Take, the sizes of priori features is downsampled as 64 × 64, and in order to keep the uniformity of characteristic size, fine granularity SR encoders are from step
A length of 23 × 3 convolutional layers start, and make characteristic pattern be down sampled to 64 × 64, then utilize residual error structure extraction characteristics of image.
Further, the decoder, fine granularity SR decoders recover final using feature and prior information at the same time
Fine granularity HR images, first, the input that priori features p and characteristics of image f are connected as decoder, then volume 3 × 3
The quantity of Feature Mapping is reduced to 64 by lamination, is upsampled to the size of Feature Mapping using the uncoiling lamination of one 4 × 4
128 × 128, then feature is decoded using 3 residual error modules, finally recovers fine granularity HR figures using 3 × 3 convolutional layers
Picture.
Wherein, human face super-resolution generation confrontation network, in order to generate high-resolution human face true to nature, existing
Network (GAN) is resisted using a generation in model, obtains human face super-resolution generation confrontation network (FSRGAN), resists network
The object function of C is expressed as：
Wherein, C output probability attribute be it is true,It is the desired value of probability distribution, except antagonism is lostOutside, make
The advanced features figure trained in advance with network introduces one and perceives loss, for assessing the correlation of perception, passes through below equation
It is given：
Wherein, φ represents the network model in fixing point pre-training, and by image y orIt is mapped to feature space, FSRGAN
Final goal function be：
Wherein, γCIt is to generate the weight for resisting network, γPIt is the perception loss of generation confrontation network.
Further, the training, uses an amount of images to be tested for 2300 data set, in data set most
50 pictures are used to test afterwards, other pictures are used to train, and data enhancing is performed on training image, original image is revolved respectively
Turn 90 °, 180 °, 270 ° and to carry out flip horizontal, make each 7 enhancing images of original image affix, every width in data set
Image all demarcates 194 key points and 11 analysis diagrams.
Further, the training image, it is rough to training image execution to cut out, operated in no any prealignment
In the case of, according to the face area of image, image size is adjusted to 128 × 128, the image cut using human-face detector
Inputted as test, the low-resolution image of input makes picture size and the high-resolution of output by the interpolation amplification of bicubic
Rate image is identical, and model is trained using a gradient optimization algorithm, and initial learning rate is 2.5 × 10-4, most in small batches
Measure as 14, λ=1, γ are set in two datasetsC=10-3,γP=10-1。
Brief description of the drawings
Fig. 1 is a kind of system framework of the image alignment method based on new end-to-end human face super-resolution network of the present invention
Figure.
Fig. 2 is a kind of network structure of the image alignment method based on new end-to-end human face super-resolution network of the present invention
Figure.
Fig. 3 is a kind of data collected explanations or commentaries of the image alignment method based on new end-to-end human face super-resolution network of the present invention
Analysis figure.
Embodiment
It should be noted that in the case where there is no conflict, the feature in embodiment and embodiment in the application can phase
Mutually combine, the present invention is described in further detail with specific embodiment below in conjunction with the accompanying drawings.
Fig. 1 is a kind of system framework of the image alignment method based on new end-to-end human face super-resolution network of the present invention
Figure.Mainly include human face super-resolution network, human face super-resolution generation confrontation network.
Wherein, the human face super-resolution network, human face super-resolution network (FSRNet) is by coarseness super-resolution
Network and fine granularity super-resolution network composition, it includes fine granularity SR encoders, prior estimate network and fine granularity SR solutions
Code device, low resolution input picture is represented with x, represents high-resolution (HR) image by FSRNet recoveries with y, p represents logical
The prior information of FSRNet estimations is crossed, since the low-resolution image clarity inputted in prior estimate is relatively low, therefore constructs one
A coarseness SR networks recover coarseness SR images, are given by below equation：
Wherein, x is low resolution (LR) image, ycIt is coarseness SR images,It is coarseness super-resolution network from image
X to image ycMapping, ycIt is respectively sent to prior estimate networkWith fine granularity SR encodersOn, obtain below equation：
Wherein, f is by encoderThe feature of extraction, after coding is completed, utilizes the decoder of SRConnection figure picture is special
F and prior information p is levied to recover SR images, is expressed by the following equation：
Give a training setWherein N is the quantity of institute's training image, x(i)It is super positioned at calibration
Low-resolution image in image in different resolution,It is the prior information of calibration, the loss function of FSRNet is given by：
Wherein, Θ represent parameter set, λ be priori loss weight, y(i)And p(i)Be recover HR images, p(i)It is i-th
The prior information of a Image estimation.
Further, the coarseness super-resolution network, probably recovers coarseness HR figures using coarseness SR networks
Picture, mitigates the difficulty of estimation prior information with this, coarseness SR network architecture is by 3 × 3 convolutional layer, followed by 3
A residual error module, finally rebuilds coarseness HR images using another 3 × 3 convolutional layer.
Further, the fine granularity super-resolution network, in fine granularity SR networks, coarseness HR images are sent
Into prior estimate network and fine granular encoded device network, it is respectively intended to carry out facial prior information estimation and feature extraction, so
Decoder recovers fine granularity HR images using the two results afterwards.
Further, the prior estimate network, estimates facial markers using hourglass configuration in prior estimate network
The crucial figure and analysis diagram of point, since the two prior informations represent the face shape of two dimension, therefore in prior estimate network, are removed
Outside last layer, whole features between the two tasks are shared, in order to merge feature effectively across scale and preserve different rulers
The spatial information of degree, hourglass module, using jump connection mechanism, then handle institute between symmetrical layers with 1 × 1 convolutional layer
The feature of acquisition, the feature being shared connect two single 1 × 1 convolutional layers, generate key point hotspot graph and analysis diagram.
Further, the fine granular encoded device, for fine granularity SR encoders, carries out feature using residual error module and carries
Take, the sizes of priori features is downsampled as 64 × 64, and in order to keep the uniformity of characteristic size, fine granularity SR encoders are from step
A length of 23 × 3 convolutional layers start, and make characteristic pattern be down sampled to 64 × 64, then utilize residual error structure extraction characteristics of image.
Further, the decoder, fine granularity SR decoders recover final using feature and prior information at the same time
Fine granularity HR images, first, the input that priori features p and characteristics of image f are connected as decoder, then volume 3 × 3
The quantity of Feature Mapping is reduced to 64 by lamination, is upsampled to the size of Feature Mapping using the uncoiling lamination of one 4 × 4
128 × 128, then feature is decoded using 3 residual error modules, finally recovers fine granularity HR figures using 3 × 3 convolutional layers
Picture.
Wherein, human face super-resolution generation confrontation network, in order to generate high-resolution human face true to nature, existing
Network (GAN) is resisted using a generation in model, obtains human face super-resolution generation confrontation network (FSRGAN), resists network
The object function of C is expressed as：
Wherein, C output probability attribute be it is true,It is the desired value of probability distribution, except antagonism is lostOutside, make
The advanced features figure trained in advance with network introduces one and perceives loss, for assessing the correlation of perception, passes through below equation
It is given：
Wherein, φ represents the network model in fixing point pre-training, and by image y orIt is mapped to feature space, FSRGAN
Final goal function be：
Wherein, γCIt is to generate the weight for resisting network, γPIt is the perception loss of generation confrontation network.
Further, the training, uses an amount of images to be tested for 2300 data set, in data set most
50 pictures are used to test afterwards, other pictures are used to train, and data enhancing is performed on training image, original image is revolved respectively
Turn 90 °, 180 °, 270 ° and to carry out flip horizontal, make each 7 enhancing images of original image affix, every width in data set
Image all demarcates 194 key points and 11 analysis diagrams.
Further, the training image, it is rough to training image execution to cut out, operated in no any prealignment
In the case of, according to the face area of image, image size is adjusted to 128 × 128, the image cut using human-face detector
Inputted as test, the low-resolution image of input makes picture size and the high-resolution of output by the interpolation amplification of bicubic
Rate image is identical, and model is trained using a gradient optimization algorithm, and initial learning rate is 2.5 × 10-4, most in small batches
Measure as 14, λ=1, γ are set in two datasetsC=10-3,γP=10-1。
Fig. 2 is a kind of network structure of the image alignment method based on new end-to-end human face super-resolution network of the present invention
Figure.It is made of coarseness SR networks and fine granularity SR networks, since the convolutional layer of one 3 × 3, followed by 3 residual error modules,
Finally coarseness HR images are rebuild using another 3 × 3 convolutional layer.
Fig. 3 is a kind of data collected explanations or commentaries of the image alignment method based on new end-to-end human face super-resolution network of the present invention
Analysis figure.(a) it is original image, (b) is that (c) is uncalibrated image by the visualization cromogram of the alignment image generation of 11 calibration
Global parsing figure, (d) is the part analysis figure of uncalibrated image, respectively comprising left eyebrow, right eyebrow, left eye, right eye, nose,
Upper lip, oral cavity and lower lip.
For those skilled in the art, the present invention is not restricted to the details of above-described embodiment, in the essence without departing substantially from the present invention
In the case of refreshing and scope, the present invention can be realized in other specific forms.In addition, those skilled in the art can be to this hair
Bright to carry out various modification and variations without departing from the spirit and scope of the present invention, these improvements and modifications also should be regarded as the present invention's
Protection domain.Therefore, appended claims are intended to be construed to include preferred embodiment and fall into all changes of the scope of the invention
More and change.
Claims (10)
- A kind of 1. image alignment method based on new end-to-end human face super-resolution network, it is characterised in that mainly including people Face super-resolution network (one)；Human face super-resolution generation confrontation network (two).
- 2. based on the human face super-resolution network (one) described in claim 1, it is characterised in that human face super-resolution network (FSRNet) it is made of coarseness super-resolution network and fine granularity super-resolution network, it includes fine granularity SR encoders, elder generation Estimation network and fine granularity SR decoders are tested, low resolution input picture is represented with x, represents what is recovered by FSRNet with y High-resolution (HR) image, p represents the prior information estimated by FSRNet, due to the low resolution inputted in prior estimate Image definition is relatively low, therefore constructs a coarseness SR network to recover coarseness SR images, is given by below equation：Wherein, x is low resolution (LR) image, ycIt is coarseness SR images,It is coarseness super-resolution network from image x to figure As ycMapping, ycIt is respectively sent to prior estimate networkWith fine granularity SR encodersOn, obtain below equation：Wherein, f is by encoderThe feature of extraction, after coding is completed, utilizes the decoder of SRConnect characteristics of image f and Prior information p recovers SR images, is expressed by the following equation：Give a training setWherein N is the quantity of institute's training image, x(i)It is positioned at calibration super-resolution Low-resolution image in rate image,It is the prior information of calibration, the loss function of FSRNet is given by：Wherein, Θ represent parameter set, λ be priori loss weight, y(i)And p(i)Be recover HR images, p(i)It is in i-th of figure As the prior information of estimation.
- 3. based on the coarseness super-resolution network described in claim 2, it is characterised in that probably extensive using coarseness SR networks Multiple coarseness HR images, mitigate the difficulty of estimation prior information with this, coarseness SR network architecture by 3 × 3 convolutional layer Start, followed by 3 residual error modules, finally coarseness HR images are rebuild using another 3 × 3 convolutional layer.
- 4. based on the fine granularity super-resolution network described in claim 2, it is characterised in that in fine granularity SR networks, coarse grain Degree HR images are sent in prior estimate network and fine granular encoded device network, are respectively intended to carry out facial prior information estimation And feature extraction, then decoder recover fine granularity HR images using the two results.
- 5. based on the prior estimate network described in claim 4, it is characterised in that using hourglass configuration in prior estimate network The crucial figure and analysis diagram of facial markers point are estimated, since the two prior informations represent the face shape of two dimension, therefore in priori Estimate in network, in addition to last layer, share whole features between the two tasks, in order to merge feature effectively across scale And the spatial information of different scale is preserved, hourglass module uses jump connection mechanism between symmetrical layers, then with one 1 × 1 The obtained feature of convolutional layer processing, the feature being shared connect two single 1 × 1 convolutional layers, generate key point hotspot graph And analysis diagram.
- 6. based on the fine granular encoded device described in claim 4, it is characterised in that for fine granularity SR encoders, utilize residual error Module carries out feature extraction, and the sizes of priori features is downsampled as 64 × 64, in order to keep the uniformity of characteristic size, particulate SR encoders are spent since step-length is 23 × 3 convolutional layers, are made characteristic pattern be down sampled to 64 × 64, are then carried using residual error structure Take characteristics of image.
- 7. based on the decoder described in claim 4, it is characterised in that fine granularity SR decoders are believed using feature and priori at the same time Cease to recover final fine granularity HR images, first, priori features p and characteristics of image f are connected as the defeated of decoder Enter, then the quantity of Feature Mapping is reduced to 64 by 3 × 3 convolutional layers, using the uncoiling lamination of one 4 × 4 by Feature Mapping Size be upsampled to 128 × 128, then feature is decoded using 3 residual error modules, finally using 3 × 3 convolutional layers come Recover fine granularity HR images.
- 8. based on the human face super-resolution generation confrontation network (two) described in claim 1, it is characterised in that true to nature in order to generate High-resolution human face, in existing model using one generation resist network (GAN), obtain human face super-resolution generation confrontation Network (FSRGAN), the object function for resisting network C are expressed as：Wherein, C output probability attribute be it is true,It is the desired value of probability distribution, except antagonism is lostOutside, use net The advanced features figure that network is trained in advance introduces one and perceives loss, for assessing the correlation of perception, is given by below equation：Wherein, φ represents the network model in fixing point pre-training, and by image y orFeature space is mapped to, FSRGAN is most Whole object function is：Wherein, γCIt is to generate the weight for resisting network, γPIt is the perception loss of generation confrontation network.
- 9. based on the training described in claim 8, it is characterised in that use an amount of images to be surveyed for 2300 data set Trying, last 50 pictures are used to test in data set, other pictures are used to train, and data enhancing is performed on training image, by Original image is rotated by 90 ° respectively, 180 °, 270 ° and carry out flip horizontal, make each 7 enhancing images of original image affix, Each image in data set all demarcates 194 key points and 11 analysis diagrams.
- 10. based on the training image described in claim 9, it is characterised in that it is rough to training image execution to cut out, do not having In the case of any prealignment operation, according to the face area of image, image size is adjusted to 128 × 128, is examined using face Survey the image that device is cut to input as test, the low-resolution image of input makes picture size by the interpolation amplification of bicubic It is identical with the high-definition picture of output, model is trained using a gradient optimization algorithm, initial learning rate is 2.5×10-4, minimum lot size 14, sets λ=1, γ in two datasetsC=10-3,γP=10-1。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810045095.7A CN107958246A (en) | 2018-01-17 | 2018-01-17 | A kind of image alignment method based on new end-to-end human face super-resolution network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810045095.7A CN107958246A (en) | 2018-01-17 | 2018-01-17 | A kind of image alignment method based on new end-to-end human face super-resolution network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107958246A true CN107958246A (en) | 2018-04-24 |
Family
ID=61956259
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810045095.7A Withdrawn CN107958246A (en) | 2018-01-17 | 2018-01-17 | A kind of image alignment method based on new end-to-end human face super-resolution network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107958246A (en) |
Cited By (29)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108764085A (en) * | 2018-05-17 | 2018-11-06 | 上海交通大学 | Based on the people counting method for generating confrontation network |
| CN108921782A (en) * | 2018-05-17 | 2018-11-30 | 腾讯科技（深圳）有限公司 | A kind of image processing method, device and storage medium |
| CN109101919A (en) * | 2018-08-03 | 2018-12-28 | 北京字节跳动网络技术有限公司 | Method and apparatus for generating information |
| CN109191378A (en) * | 2018-07-25 | 2019-01-11 | 成都信息工程大学 | Based on ERGAN network single image super-resolution reconstruction method |
| CN109190520A (en) * | 2018-08-16 | 2019-01-11 | 广州视源电子科技股份有限公司 | A kind of super-resolution rebuilding facial image method and device |
| CN109544450A (en) * | 2018-11-09 | 2019-03-29 | 武汉烽火众智数字技术有限责任公司 | A kind of confrontation generates network establishing method and device, image reconstructing method and device |
| CN109636721A (en) * | 2018-11-29 | 2019-04-16 | 武汉大学 | Video super-resolution method based on confrontation study and attention mechanism |
| CN109753946A (en) * | 2019-01-23 | 2019-05-14 | 哈尔滨工业大学 | A kind of real scene pedestrian's small target deteection network and detection method based on the supervision of body key point |
| CN109993782A (en) * | 2019-04-02 | 2019-07-09 | 中国矿业大学 | A kind of annular generates the heterologous remote sensing image registration method and device of confrontation network |
| CN110111251A (en) * | 2019-04-22 | 2019-08-09 | 电子科技大学 | A kind of combination depth supervision encodes certainly and perceives the image super-resolution rebuilding method of iterative backprojection |
| CN110263756A (en) * | 2019-06-28 | 2019-09-20 | 东北大学 | A kind of human face super-resolution reconstructing system based on joint multi-task learning |
| CN110443782A (en) * | 2019-07-03 | 2019-11-12 | 杭州深睿博联科技有限公司 | Chest x-ray piece model alignment schemes and device, storage medium |
| CN110490158A (en) * | 2019-08-23 | 2019-11-22 | 安徽大学 | A kind of robust human face alignment schemes based on multistage model |
| CN110782398A (en) * | 2018-12-13 | 2020-02-11 | 北京嘀嘀无限科技发展有限公司 | Image processing method, generation type countermeasure network, electronic equipment and storage medium |
| JP2020047270A (en) * | 2018-09-17 | 2020-03-26 | 株式会社ストラドビジョン | Learning method and learning device applying multi-feeding, and test method and test device utilizing the same |
| CN110956575A (en) * | 2018-09-26 | 2020-04-03 | 京东方科技集团股份有限公司 | Method and device for converting image style and convolution neural network processor |
| CN111080511A (en) * | 2019-11-18 | 2020-04-28 | 杭州时光坐标影视传媒股份有限公司 | End-to-end face exchange method for high-resolution multi-feature extraction |
| CN111199522A (en) * | 2019-12-24 | 2020-05-26 | 重庆邮电大学 | Single-image blind motion blur removing method for generating countermeasure network based on multi-scale residual errors |
| CN111221118A (en) * | 2020-02-26 | 2020-06-02 | 南京理工大学 | Microscopic imaging method based on phase coding single lens |
| CN111507914A (en) * | 2020-04-10 | 2020-08-07 | 北京百度网讯科技有限公司 | Training method, repairing method, device, equipment and medium of face repairing model |
| CN111523553A (en) * | 2020-04-03 | 2020-08-11 | 中国计量大学 | Central point network multi-target detection method based on similarity matrix |
| CN112348103A (en) * | 2020-11-16 | 2021-02-09 | 南开大学 | Image block classification method and device and super-resolution reconstruction method and device thereof |
| CN113094422A (en) * | 2021-03-12 | 2021-07-09 | 中山大学 | Urban road traffic flow chart generation method, system and equipment |
| CN113108792A (en) * | 2021-03-16 | 2021-07-13 | 中山大学 | Wi-Fi fingerprint map reconstruction method and device, terminal equipment and medium |
| CN113159171A (en) * | 2021-04-20 | 2021-07-23 | 复旦大学 | Plant leaf image fine classification method based on counterstudy |
| CN113160044A (en) * | 2020-01-23 | 2021-07-23 | 百度在线网络技术（北京）有限公司 | Depth image super-resolution method, training device, equipment and medium |
| CN113379606A (en) * | 2021-08-16 | 2021-09-10 | 之江实验室 | Face super-resolution method based on pre-training generation model |
| CN113658040A (en) * | 2021-07-14 | 2021-11-16 | 西安理工大学 | Face super-resolution method based on prior information and attention fusion mechanism |
| CN115994966A (en) * | 2023-03-22 | 2023-04-21 | 北京红棉小冰科技有限公司 | Multi-view image generation method and device, readable storage medium and electronic equipment |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107154023A (en) * | 2017-05-17 | 2017-09-12 | 电子科技大学 | Face super-resolution reconstruction method based on generation confrontation network and sub-pix convolution |
| CN107451619A (en) * | 2017-08-11 | 2017-12-08 | 深圳市唯特视科技有限公司 | A kind of small target detecting method that confrontation network is generated based on perception |
| CN107481188A (en) * | 2017-06-23 | 2017-12-15 | 珠海经济特区远宏科技有限公司 | A kind of image super-resolution reconstructing method |
-
2018
- 2018-01-17 CN CN201810045095.7A patent/CN107958246A/en not_active Withdrawn
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107154023A (en) * | 2017-05-17 | 2017-09-12 | 电子科技大学 | Face super-resolution reconstruction method based on generation confrontation network and sub-pix convolution |
| CN107481188A (en) * | 2017-06-23 | 2017-12-15 | 珠海经济特区远宏科技有限公司 | A kind of image super-resolution reconstructing method |
| CN107451619A (en) * | 2017-08-11 | 2017-12-08 | 深圳市唯特视科技有限公司 | A kind of small target detecting method that confrontation network is generated based on perception |
Non-Patent Citations (1)
| Title |
|---|
| YU CHEN ET AL.: "FSRNet: End-to-End Learning Face Super-Resolution with Facial Priors", 《ARXIV》 * |
Cited By (44)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108764085A (en) * | 2018-05-17 | 2018-11-06 | 上海交通大学 | Based on the people counting method for generating confrontation network |
| CN108921782A (en) * | 2018-05-17 | 2018-11-30 | 腾讯科技（深圳）有限公司 | A kind of image processing method, device and storage medium |
| CN108764085B (en) * | 2018-05-17 | 2022-02-25 | 上海交通大学 | Crowd counting method based on generation of confrontation network |
| CN108921782B (en) * | 2018-05-17 | 2023-04-14 | 腾讯科技（深圳）有限公司 | Image processing method, device and storage medium |
| CN109191378A (en) * | 2018-07-25 | 2019-01-11 | 成都信息工程大学 | Based on ERGAN network single image super-resolution reconstruction method |
| CN109101919A (en) * | 2018-08-03 | 2018-12-28 | 北京字节跳动网络技术有限公司 | Method and apparatus for generating information |
| CN109190520A (en) * | 2018-08-16 | 2019-01-11 | 广州视源电子科技股份有限公司 | A kind of super-resolution rebuilding facial image method and device |
| JP2020047270A (en) * | 2018-09-17 | 2020-03-26 | 株式会社ストラドビジョン | Learning method and learning device applying multi-feeding, and test method and test device utilizing the same |
| CN110956575B (en) * | 2018-09-26 | 2022-04-12 | 京东方科技集团股份有限公司 | Method and device for converting image style and convolution neural network processor |
| CN110956575A (en) * | 2018-09-26 | 2020-04-03 | 京东方科技集团股份有限公司 | Method and device for converting image style and convolution neural network processor |
| US11842267B2 (en) | 2018-09-26 | 2023-12-12 | Boe Technology Group Co., Ltd. | Computer-implemented method using a neural network comprising down-sampling segment, densely connected segment, and up-sampling segment, apparatus, and computer-program product |
| CN109544450A (en) * | 2018-11-09 | 2019-03-29 | 武汉烽火众智数字技术有限责任公司 | A kind of confrontation generates network establishing method and device, image reconstructing method and device |
| CN109544450B (en) * | 2018-11-09 | 2022-08-19 | 武汉烽火众智数字技术有限责任公司 | Method and device for constructing confrontation generation network and method and device for reconstructing image |
| CN109636721B (en) * | 2018-11-29 | 2023-06-23 | 武汉大学 | Video super-resolution method based on countermeasure learning and attention mechanism |
| CN109636721A (en) * | 2018-11-29 | 2019-04-16 | 武汉大学 | Video super-resolution method based on confrontation study and attention mechanism |
| CN110782398A (en) * | 2018-12-13 | 2020-02-11 | 北京嘀嘀无限科技发展有限公司 | Image processing method, generation type countermeasure network, electronic equipment and storage medium |
| CN109753946A (en) * | 2019-01-23 | 2019-05-14 | 哈尔滨工业大学 | A kind of real scene pedestrian's small target deteection network and detection method based on the supervision of body key point |
| CN109993782A (en) * | 2019-04-02 | 2019-07-09 | 中国矿业大学 | A kind of annular generates the heterologous remote sensing image registration method and device of confrontation network |
| CN109993782B (en) * | 2019-04-02 | 2020-12-04 | 中国矿业大学 | Heterogeneous remote sensing image registration method and device for ring-shaped generation countermeasure network |
| CN110111251A (en) * | 2019-04-22 | 2019-08-09 | 电子科技大学 | A kind of combination depth supervision encodes certainly and perceives the image super-resolution rebuilding method of iterative backprojection |
| CN110111251B (en) * | 2019-04-22 | 2023-04-28 | 电子科技大学 | Image super-resolution reconstruction method combining depth supervision self-coding and perception iterative back projection |
| CN110263756A (en) * | 2019-06-28 | 2019-09-20 | 东北大学 | A kind of human face super-resolution reconstructing system based on joint multi-task learning |
| CN110443782A (en) * | 2019-07-03 | 2019-11-12 | 杭州深睿博联科技有限公司 | Chest x-ray piece model alignment schemes and device, storage medium |
| CN110490158A (en) * | 2019-08-23 | 2019-11-22 | 安徽大学 | A kind of robust human face alignment schemes based on multistage model |
| CN111080511A (en) * | 2019-11-18 | 2020-04-28 | 杭州时光坐标影视传媒股份有限公司 | End-to-end face exchange method for high-resolution multi-feature extraction |
| CN111080511B (en) * | 2019-11-18 | 2020-12-25 | 杭州时光坐标影视传媒股份有限公司 | End-to-end face exchange method for high-resolution multi-feature extraction |
| CN111199522B (en) * | 2019-12-24 | 2024-02-09 | 芽米科技(广州)有限公司 | Single-image blind removal motion blurring method for generating countermeasure network based on multi-scale residual error |
| CN111199522A (en) * | 2019-12-24 | 2020-05-26 | 重庆邮电大学 | Single-image blind motion blur removing method for generating countermeasure network based on multi-scale residual errors |
| CN113160044B (en) * | 2020-01-23 | 2023-12-26 | 百度在线网络技术（北京）有限公司 | Depth image super-resolution method, training method and device, equipment and medium |
| CN113160044A (en) * | 2020-01-23 | 2021-07-23 | 百度在线网络技术（北京）有限公司 | Depth image super-resolution method, training device, equipment and medium |
| CN111221118A (en) * | 2020-02-26 | 2020-06-02 | 南京理工大学 | Microscopic imaging method based on phase coding single lens |
| CN111523553A (en) * | 2020-04-03 | 2020-08-11 | 中国计量大学 | Central point network multi-target detection method based on similarity matrix |
| CN111507914A (en) * | 2020-04-10 | 2020-08-07 | 北京百度网讯科技有限公司 | Training method, repairing method, device, equipment and medium of face repairing model |
| CN111507914B (en) * | 2020-04-10 | 2023-08-08 | 北京百度网讯科技有限公司 | Training method, repairing method, device, equipment and medium for face repairing model |
| CN112348103A (en) * | 2020-11-16 | 2021-02-09 | 南开大学 | Image block classification method and device and super-resolution reconstruction method and device thereof |
| CN112348103B (en) * | 2020-11-16 | 2022-11-11 | 南开大学 | Image block classification method and device and super-resolution reconstruction method and device thereof |
| CN113094422B (en) * | 2021-03-12 | 2023-07-07 | 中山大学 | Urban road traffic flow map generation method, system and equipment |
| CN113094422A (en) * | 2021-03-12 | 2021-07-09 | 中山大学 | Urban road traffic flow chart generation method, system and equipment |
| CN113108792A (en) * | 2021-03-16 | 2021-07-13 | 中山大学 | Wi-Fi fingerprint map reconstruction method and device, terminal equipment and medium |
| CN113159171B (en) * | 2021-04-20 | 2022-07-22 | 复旦大学 | Plant leaf image fine classification method based on counterstudy |
| CN113159171A (en) * | 2021-04-20 | 2021-07-23 | 复旦大学 | Plant leaf image fine classification method based on counterstudy |
| CN113658040A (en) * | 2021-07-14 | 2021-11-16 | 西安理工大学 | Face super-resolution method based on prior information and attention fusion mechanism |
| CN113379606A (en) * | 2021-08-16 | 2021-09-10 | 之江实验室 | Face super-resolution method based on pre-training generation model |
| CN115994966A (en) * | 2023-03-22 | 2023-04-21 | 北京红棉小冰科技有限公司 | Multi-view image generation method and device, readable storage medium and electronic equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107958246A (en) | A kind of image alignment method based on new end-to-end human face super-resolution network | |
| Xiong et al. | Hscnn: Cnn-based hyperspectral image recovery from spectrally undersampled projections | |
| Shao et al. | Remote sensing image fusion with deep convolutional neural network | |
| Yu et al. | Hallucinating very low-resolution unaligned and noisy face images by transformative discriminative autoencoders | |
| Su et al. | Object detection and instance segmentation in remote sensing imagery based on precise mask R-CNN | |
| Zhang et al. | Sparse representation based multi-sensor image fusion for multi-focus and multi-modality images: A review | |
| Ling et al. | A saliency prediction model on 360 degree images using color dictionary based sparse representation | |
| US20220044375A1 (en) | Saliency Map Enhancement-Based Infrared and Visible Light Fusion Method | |
| Xu et al. | High quality remote sensing image super-resolution using deep memory connected network | |
| Chen et al. | SIRF: Simultaneous satellite image registration and fusion in a unified framework | |
| Li et al. | From beginner to master: A survey for deep learning-based single-image super-resolution | |
| Abiantun et al. | SSR2: Sparse signal recovery for single-image super-resolution on faces with extreme low resolutions | |
| CN108596108B (en) | Aerial remote sensing image change detection method based on triple semantic relation learning | |
| An et al. | TR-MISR: Multiimage super-resolution based on feature fusion with transformers | |
| CN108765280A (en) | A kind of high spectrum image spatial resolution enhancement method | |
| JP2018072938A (en) | Number-of-targets estimation device, number-of-targets estimation method, and program | |
| Küchhold et al. | Scale-adaptive real-time crowd detection and counting for drone images | |
| Pashaei et al. | Deep learning-based single image super-resolution: an investigation for dense scene reconstruction with UAS photogrammetry | |
| CN110647909A (en) | Remote sensing image classification method based on three-dimensional dense convolution neural network | |
| CN113139902A (en) | Hyperspectral image super-resolution reconstruction method and device and electronic equipment | |
| Chen et al. | MICU: Image super-resolution via multi-level information compensation and U-net | |
| Xia et al. | Cloud removal of optical remote sensing imagery with multitemporal SAR-optical data using X-Mtgan | |
| Zhu et al. | Super resolution reconstruction method for infrared images based on pseudo transferred features | |
| Wang et al. | A Review of GAN-Based Super-Resolution Reconstruction for Optical Remote Sensing Images | |
| CN104182931B (en) | Super resolution method and device |
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 | ||
| WW01 | Invention patent application withdrawn after publication | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20180424 |