CN108269245A - A kind of eyes image restorative procedure based on novel generation confrontation network - Google Patents

A kind of eyes image restorative procedure based on novel generation confrontation network Download PDF

Info

Publication number
CN108269245A
CN108269245A CN201810078940.0A CN201810078940A CN108269245A CN 108269245 A CN108269245 A CN 108269245A CN 201810078940 A CN201810078940 A CN 201810078940A CN 108269245 A CN108269245 A CN 108269245A
Authority
CN
China
Prior art keywords
image
reference picture
code
eyes
generator
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Withdrawn
Application number
CN201810078940.0A
Other languages
Chinese (zh)
Inventor
夏春秋
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.)
Shenzhen Vision Technology Co Ltd
Original Assignee
Shenzhen Vision Technology 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 Shenzhen Vision Technology Co Ltd filed Critical Shenzhen Vision Technology Co Ltd
Priority to CN201810078940.0A priority Critical patent/CN108269245A/en
Publication of CN108269245A publication Critical patent/CN108269245A/en
Withdrawn legal-status Critical Current

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00Image enhancement or restoration
    • G06T5/77Retouching; Inpainting; Scratch removal
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/18Eye characteristics, e.g. of the iris
    • G06V40/193Preprocessing; Feature extraction

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Ophthalmology & Optometry (AREA)
  • Human Computer Interaction (AREA)
  • Multimedia (AREA)
  • Image Analysis (AREA)

Abstract

The present invention proposes a kind of eyes image restorative procedure based on novel generation confrontation network, and main contents include:Eyes image reparation, novel generation confrontation network, model framework, its process fights network (ExGAN) to introduce novel generation, generator is instructed using second photographed image-related information created by generator, when repairing face-image, it is used as using same person in the second image of different time or different gestures with reference to image, ExGAN is while original photo feature is retained, using reference data, the method for proposing to propose to be repaired based on reference picture and code respectively in ExGAN.The present invention is repaired using the sample information in reference picture region, code, which is perceived, using one describes the object, the inserting extra information on multiple points in fighting network, increases its descriptive power, is conducive to generate high quality, lifelike individualized therapy effect as original image.

Description

A kind of eyes image restorative procedure based on novel generation confrontation network
Technical field
The present invention relates to image processing field, more particularly, to a kind of eyes image based on novel generation confrontation network Restorative procedure.
Background technology
Eyes image reparation is an important content in image procossing, the purpose is to using the existing information of image come extensive The information lost again is usually used in the fields such as public security, traffic, medicine, military investigation.Specifically, in police field, available for public security The recovery of face detection and imperfect picture in picture.In field of traffic, image repair can be utilized to analyze traffic accident, The noise of removal accident picture.In medical domain, X-ray lung images is contributed to increase clear, ultrasonography processing etc..And in military affairs Field, the key message that eyes image is read in criminal evidence obtaining are most important.But in public security, medicine, military surveillance and day Usually due to the distortion of picture pick-up device optical system, focus inaccurate, relative motion or personage's blink in the In vivo detection often lived Eye closing etc. causes the fuzzy of image so that and the extraction of information becomes difficult, and the influence of noise eliminated in image is also more and more important, For example identify that suspect, extraction proves or carry out technical appraisement from public security criminal image data from the crowd quickly moved Etc., these important applications are required for removing distortion as much as possible by image repair technology, and it is extensive to carry out personage's eye closing image It is multiple etc., therefore have important practical significance for the technical research of eyes image reparation.
The present invention proposes a kind of eyes image restorative procedure based on novel generation confrontation network, introduces novel generation pair Anti- network (ExGAN) instructs generator using second photographed image-related information created by generator, with more and more numbers It is developed according to collection, it is assumed that the second image of special object rationally exists, when repairing face-image, using same person when different Between or the second image of different gestures be used as with reference to image, ExGAN utilizes reference number while original photo feature is retained According to the method for proposing to be repaired based on reference picture and code.The present invention using the sample information in reference picture region into Row is repaired, and perceiving code using one describes the object, and the inserting extra information on multiple points in fighting network increases it and retouches Ability is stated, is conducive to generate high quality, lifelike individualized therapy effect as original image.
Invention content
For eyes image reparation, the present invention proposes a kind of eyes image reparation side based on novel generation confrontation network Method, introduces novel generation confrontation network (ExGAN), and generation is instructed using second photographed image-related information created by generator Device when repairing face-image, is used as in the second image of different time or different gestures with reference to image using same person, ExGAN, using reference data, proposes to be repaired based on reference picture and code respectively while original photo feature is retained Method.
To solve the above problems, propose a kind of eyes image restorative procedure based on novel generation confrontation network, master Content is wanted to include:
(1) eyes image reparation;
(2) novel generation confrontation network;
(3) model framework.
Wherein, the eyes image reparation, is repaired using the sample information in reference picture region, uses one It perceives code and describes the object, the inserting extra information on multiple points in fighting network increases its descriptive power, generates trueer Real repairing effect.
Wherein, novel generation confrontation network, introduces novel generation confrontation network (ExGAN), and generator creates figure As after, second image is obtained, generator is instructed using the relevant information of second image, with more and more data set quilts Exploitation, it is assumed that the second image of special object rationally exists, when repairing face-image, using same person in different time or not The second image with posture is used as with reference to image, these information are merged into a semantic guide, generation is just by network by study True reparation is as a result, ExGAN while original photo feature is retained, using reference data, proposes two kinds of independences in ExGAN Method repair image:It is the method repaired based on reference picture first, the reference picture r in generator GiAs to It leads, reference picture r is used in discriminator DiCome whether determining generated image is true as additional information;Secondly it is based on The restorative procedure of code creates in information area and perceives code ci, the compressed version of eye image is stored in vector In, wherein each target is with additional information riAnd ciFor condition, additional content loss item is added for target.
Further, the eye image of the discriminator, the entire facial image of discriminator processing and amplification, overall situation confrontation Property loss strengthen whole semantic consistency, local antagonism loss ensures generate the details and clarity of output, the overall situation The output of convolution branch and local convolution branch forms a sigmoid function by connection, and reference picture is inputted in discriminator D, Additional global convolution branch is added in discriminator, the output of three branches is connected.
Further, it is described to be repaired based on reference picture, training set xiIn each image it is right there are one The reference picture r answeredi, training set X is defined as a tuple X={ (x1,r1), in eyes image reparation, riIt is xiIn it is same The image of one people's difference posture, in xiMiddle removal Hotfix generation new images zi, learning objective is defined as:
In order to preferably generalize, in training set xiIn accordingly give one group of reference picture set Ri, training set is extended To a tuple-set:X={ x1×R1,…,xn×Rn, the set by each needs reparation image and its reference picture it Between cartesian product composition.
Further, the reparation based on code, the pixel number of each image is in data set | I |, it is assumed that there are One compression functionWherein N < < | I |, for each image z for needing to repairiAnd its corresponding reference Image ri, use riGenerate code ci=C (ri), in view of encoded sample information, is by confrontation object definition:
Wherein, compression function is a general depth network projected to example in some manifold, in equation (2) most Latter is the generation image G (z in sensing regioni,ci) and original reference image riThe optional loss of distance, with measuring low-dimensional It is corresponding that the distance between image and reference picture are generated in manifold, if generator G is complete convolution, as input ciWhen, It needs to change its framework to handle any number of vector.
Further, the compression function perceives code c to generatei, one is trained individually for compression function C Autocoder, during training C, encoder chooses single eyes as input, and the decoder of autocoder is divided into left and right Branch corresponds to the different target of right and left eyes respectively, not repeated when ensuring the common trait of encoder study eyes, passes through volume Code distinguishes feature, and each eye is all encoded with the floating point vector of 128 dimensions, is formed by combining these codings The eye pattern coding of 256 dimensions.
Wherein, the model framework, has used a Standard convolution generator, and neck region includes the volume of expansion Product, since generation eye image is more confined from than general repair content repair text, so the port number of network internal layer is less, in generator One RGB image of middle input, the mobile part for needing to repair, stacks a single channel binary mask, indicates the area to be filled Domain, generator show eyes by using the rgb value and another single channel mask of four additional channels and reference picture Position, before training, all eye positions being detected store together with data set.
Further, the generator, the generator use the structure of encoder and decoder, share 4 down-samplings With up-sampling layer, there is the full connection bottleneck layer of 256 dimensions, bottleneck layer is connect with eye code, and the overall dimensions of output are 512, eyes code loses item by the perception of equation (2), is attached in the fixed-size penultimate output of discriminator, by Be more than two of original discriminator outputs, therefore before the last one sigmoid function in 256 dimensions of code, by one compared with Small bilayer is fully connected network and global and local output code is tested, with this automatic study code and convolution discriminator Between optimal weight.
Further, the data set, ExGAN need a data set for including each image pair, but due to this The data set of type is not common, in order to evade the limitation of available data collection, develops about 2,000,000 2D alignment images and makees For internal trainer collection, data set ensures everyone at least 3 images, and each image in training set includes an eye opening eyeball Image, the high-definition picture shot under various environment and lighting condition enable ExGAN to it is various input photos into Row internal repair, and noise-free picture and non-extreme posture image improve the eyes quality and clarity of generation.
Description of the drawings
Fig. 1 is a kind of system framework figure of the eyes image restorative procedure based on novel generation confrontation network of the present invention.
Fig. 2 is a kind of system flow chart of the eyes image restorative procedure based on novel generation confrontation network of the present invention.
Fig. 3 is that figure is compared in a kind of reconstruction loss of the eyes image restorative procedure based on novel generation confrontation network of the present invention
Fig. 4 is that a kind of sensing results of the eyes image restorative procedure based on novel generation confrontation network of the present invention compare Figure.
Specific embodiment
It should be noted that in the absence of conflict, the feature in embodiment and embodiment in the application can phase It mutually combines, the present invention is described in further detail in the following with reference to the drawings and specific embodiments.
Fig. 1 is a kind of system framework figure of the eyes image restorative procedure based on novel generation confrontation network of the present invention.It is main To include eyes image reparation, novel generation confrontation network, model framework.
Wherein, the eyes image reparation, is repaired using the sample information in reference picture region, uses one It perceives code and describes the object, the inserting extra information on multiple points in fighting network increases its descriptive power, generates trueer Real repairing effect.
Wherein, novel generation confrontation network, introduces novel generation confrontation network (ExGAN), and generator creates figure As after, second image is obtained, generator is instructed using the relevant information of second image, with more and more data set quilts Exploitation, it is assumed that the second image of special object rationally exists, when repairing face-image, using same person in different time or not The second image with posture is used as with reference to image, these information are merged into a semantic guide, generation is just by network by study True reparation is as a result, ExGAN while original photo feature is retained, using reference data, proposes two kinds of independences in ExGAN Method repair image:It is the method repaired based on reference picture first, the reference picture r in generator GiAs to It leads, reference picture r is used in discriminator DiCome whether determining generated image is true as additional information;Secondly it is based on The restorative procedure of code creates in information area and perceives code ci, the compressed version of eye image is stored in vector In, wherein each target is with additional information riAnd ciFor condition, additional content loss item is added for target.
Further, the eye image of the discriminator, the entire facial image of discriminator processing and amplification, overall situation confrontation Property loss strengthen whole semantic consistency, local antagonism loss ensures generate the details and clarity of output, the overall situation The output of convolution branch and local convolution branch forms a sigmoid function by connection, and reference picture is inputted in discriminator D, Additional global convolution branch is added in discriminator, the output of three branches is connected.
Further, it is described to be repaired based on reference picture, training set xiIn each image it is right there are one The reference picture r answeredi, training set X is defined as a tuple X={ (x1,r1), in eyes image reparation, riIt is xiIn it is same The image of one people's difference posture, in xiMiddle removal Hotfix generation new images zi, learning objective is defined as:
In order to preferably generalize, in training set xiIn accordingly give one group of reference picture set Ri, training set is extended To a tuple-set:X={ x1×R1,…,xn×Rn, the set by each needs reparation image and its reference picture it Between cartesian product composition.
Further, the reparation based on code, the pixel number of each image is in data set | I |, it is assumed that there are One compression functionWherein N < < | I |, for each image z for needing to repairiAnd its corresponding ginseng Examine image ri, use riGenerate code ci=C (ri), in view of encoded sample information, is by confrontation object definition:
Wherein, compression function is a general depth network projected to example in some manifold, in equation (2) most Latter is the generation image G (z in sensing regioni,ci) and original reference image riThe optional loss of distance, with measuring low-dimensional It is corresponding that the distance between image and reference picture are generated in manifold, if generator G is complete convolution, as input ciWhen, It needs to change its framework to handle any number of vector.
Further, the compression function perceives code c to generatei, one is trained individually for compression function C Autocoder, during training C, encoder chooses single eyes as input, and the decoder of autocoder is divided into left and right Branch corresponds to the different target of right and left eyes respectively, not repeated when ensuring the common trait of encoder study eyes, passes through volume Code distinguishes feature, and each eye is all encoded with the floating point vector of 128 dimensions, is formed by combining these codings The eye pattern coding of 256 dimensions.
Wherein, the model framework, has used a Standard convolution generator, and neck region includes the volume of expansion Product, since generation eye image is more confined from than general repair content repair text, so the port number of network internal layer is less, in generator One RGB image of middle input, the mobile part for needing to repair, stacks a single channel binary mask, indicates the area to be filled Domain, generator show eyes by using the rgb value and another single channel mask of four additional channels and reference picture Position, before training, all eye positions being detected store together with data set.
Further, the generator, the generator use the structure of encoder and decoder, share 4 down-samplings With up-sampling layer, there is the full connection bottleneck layer of 256 dimensions, bottleneck layer is connect with eye code, and the overall dimensions of output are 512, eyes code loses item by the perception of equation (2), is attached in the fixed-size penultimate output of discriminator, by Be more than two of original discriminator outputs, therefore before the last one sigmoid function in 256 dimensions of code, by one compared with Small bilayer is fully connected network and global and local output code is tested, with this automatic study code and convolution discriminator Between optimal weight.
Further, the data set, ExGAN need a data set for including each image pair, but due to this The data set of type is not common, in order to evade the limitation of available data collection, develops about 2,000,000 2D alignment images and makees For internal trainer collection, data set ensures everyone at least 3 images, and each image in training set includes an eye opening eyeball Image, the high-definition picture shot under various environment and lighting condition enable ExGAN to it is various input photos into Row internal repair, and noise-free picture and non-extreme posture image improve the eyes quality and clarity of generation.
Fig. 2 is a kind of system flow chart of the eyes image restorative procedure based on novel generation confrontation network of the present invention.It is whole Body training flow can be summarized as (1) and eyes marked from input picture;(2) using reference picture or code as guidance to image into Row is repaired;(3) gradient to calculate generator parameter is lost by input picture and the reconstruct being repaired between image;(4) lead to Reference picture is crossed, the image and uncalibrated image being repaired calculate the gradient of discriminator parameter;(5) it can be reversed by generator The mistake of discriminator is propagated, updates the parameter of generator using loss is perceived, in the ExGAN based on reference picture, compression Function is an identity function.
Fig. 3 is that a kind of reconstruction loss of the eyes image restorative procedure based on novel generation confrontation network of the present invention is compared Figure.The figure shows the influences that ExGAN loses whole body reconstruction.With the increase of eye code, it can be clearly seen that based on generation Code and the loss reduction amplitude based on reference picture are significantly greater than the content loss of non-GAN.
Fig. 4 is that a kind of sensing results of the eyes image restorative procedure based on novel generation confrontation network of the present invention compare Figure.Compare the sensing results that generation confrontation network (GAN) and ExGAN are generated.(a) it is uncalibrated image, (b) is not use ExGAN As a result, (c) is based on reference picture as a result, (d) is the result based on code.It can be seen from the figure that each ExGAN Excellent sensing results are produced, wherein the result that the example model based on code generates is the most true to nature.
For those skilled in the art, the present invention is not limited to the details of above-described embodiment, in the essence without departing substantially from the present invention In the case of refreshing and range, 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)

1. a kind of eyes image restorative procedure based on novel generation confrontation network, which is characterized in that mainly including eyes image It repairs (one);Novel generation confrontation network (two);Model framework (three).
2. based on the eyes image reparation (one) described in claims 1, which is characterized in that use the sample in reference picture region This information is repaired, and perceiving code using one describes the object, the inserting extra information on multiple points in fighting network, Increase its descriptive power, generate more true repairing effect.
3. the novel generation confrontation network (two) described in based on claims 1, which is characterized in that introduce novel generation confrontation net Network (ExGAN) after generator creates image, obtains second image, generator is instructed using the relevant information of second image, As more and more data sets are developed, it is assumed that the second image of special object rationally exists, and when repairing face-image, uses Same person is used as in the second image of different time or different gestures with reference to image, and network is closed these information by study And be a semantic guide, it generates and correctly repairs as a result, ExGAN utilizes reference number while original photo feature is retained According to proposing that two kinds of independent methods repair image in ExGAN:It is the method repaired based on reference picture first, Reference picture r in generator GiAs guide, reference picture r is used in discriminator DiIt determines to be generated as additional information Image whether be true;Secondly the restorative procedure based on code, creates in information area and perceives code ci, by eye image Compressed version is stored in vectorIn, wherein each target is with additional information riAnd ciFor condition, volume is added for target Outer content loss item.
4. based on the discriminator described in claims 3, which is characterized in that the entire facial image of discriminator processing and the eye of amplification Whole semantic consistency is strengthened in eyeball image, global antagonism loss, and local antagonism loss ensures to generate the details of output And clarity, the output of global convolution branch and local convolution branch forms a sigmoid function by connection, in discriminator D Additional global convolution branch is added in discriminator, the output of three branches is connected by middle input reference picture.
5. based on being repaired described in claim 3 based on reference picture, which is characterized in that training set xiIn each image All there are a corresponding reference picture ri, training set X is defined as a tuple X={ (x1,r1), in eyes image reparation In, riIt is xiThe image of middle same person difference posture, in xiMiddle removal Hotfix generation new images zi, learning objective is determined Justice is:
In order to preferably generalize, in training set xiIn accordingly give one group of reference picture set Ri, training set is expanded to one A tuple-set:X={ x1×R1,…,xn×Rn, which is repaired by each needs between image and its reference picture Cartesian product forms.
6. based on the reparation based on code described in claim 3, which is characterized in that the pixel number of each image is in data set | I |, it is assumed that there are a compression functionWherein N < < | I |, for each image z for needing to repairi And its corresponding reference picture ri, use riGenerate code ci=C (ri), in view of encoded sample information, confrontation target is determined Justice is:
Wherein, compression function is a general depth network projected to example in some manifold, last in equation (2) Item is the generation image G (z in sensing regioni,ci) and original reference image riThe optional loss of distance, with measuring low dimensional manifold Middle generation the distance between image and reference picture are corresponding, if generator G is complete convolution, as input ciWhen, it needs Its framework is changed to handle any number of vector.
7. based on the compression function described in claims 6, which is characterized in that perceive code c to generatei, it is compression function C One individual autocoder of training, during training C, encoder chooses single eyes and is used as input, autocoder Decoder is divided into left and right branch, corresponds to the different target of right and left eyes respectively, with ensure encoder study eyes common trait when It does not repeat, feature is distinguished by coding, each eye is all encoded with the floating point vector of 128 dimensions, passes through group Close the eye pattern coding that these codings form 256 dimensions.
8. based on the model framework (five) described in claims 1, which is characterized in that a Standard convolution generator has been used, Its neck region includes the convolution of expansion, since generation eye image is more confined from than general repair content repair text, so in network The port number of layer is less, and a RGB image is inputted in generator, and the mobile part for needing to repair stacks a single channel two System mask, indicates the region to be filled, generator by using four additional channels and reference picture rgb value and Another single channel mask shows the position of eyes, and before training, all eye positions being detected are deposited together with data set Storage.
9. based on the generator described in claims 8, which is characterized in that the generator uses the knot of encoder and decoder Structure shares 4 down-samplings and up-sampling layer, has the full connection bottleneck layer of 256 dimensions, bottleneck layer is connect with eye code, defeated The overall dimensions gone out are 512, and eyes code loses item by the perception of equation (2), is attached to the fixed-size inverse of discriminator In second output, since 256 dimensions of code are more than two outputs of original discriminator, in the last one S-shaped letter Before number, network is fully connected by a smaller bilayer, global and local output code is tested, with this automatic study Optimal weight between code and convolution discriminator.
10. based on the data set described in claims 8, which is characterized in that ExGAN needs a number for including each image pair According to collection, but since such data set is not common, in order to evade the limitation of available data collection, exploitation about 200 Ten thousand 2D are aligned images as internal trainer collection, and data set ensures everyone at least 3 images, each image in training set All images comprising eye opening eyeball, the high-definition picture shot under various environment and lighting condition enable ExGAN Internal repair is carried out, and noise-free picture and non-extreme posture image improve the eyes quality of generation to various input photos And clarity.
CN201810078940.0A 2018-01-26 2018-01-26 A kind of eyes image restorative procedure based on novel generation confrontation network Withdrawn CN108269245A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810078940.0A CN108269245A (en) 2018-01-26 2018-01-26 A kind of eyes image restorative procedure based on novel generation confrontation network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810078940.0A CN108269245A (en) 2018-01-26 2018-01-26 A kind of eyes image restorative procedure based on novel generation confrontation network

Publications (1)

Publication Number Publication Date
CN108269245A true CN108269245A (en) 2018-07-10

Family

ID=62776716

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810078940.0A Withdrawn CN108269245A (en) 2018-01-26 2018-01-26 A kind of eyes image restorative procedure based on novel generation confrontation network

Country Status (1)

Country Link
CN (1) CN108269245A (en)

Cited By (26)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109145745A (en) * 2018-07-20 2019-01-04 上海工程技术大学 A kind of face identification method under circumstance of occlusion
CN109308689A (en) * 2018-10-15 2019-02-05 聚时科技(上海)有限公司 The unsupervised image repair method of confrontation network migration study is generated based on mask
CN109685724A (en) * 2018-11-13 2019-04-26 天津大学 A kind of symmetrical perception facial image complementing method based on deep learning
CN109712092A (en) * 2018-12-18 2019-05-03 上海中信信息发展股份有限公司 Archives scan image repair method, device and electronic equipment
CN109919018A (en) * 2019-01-28 2019-06-21 浙江英索人工智能科技有限公司 Image eyes based on reference picture automatically open method and device
CN109919830A (en) * 2019-01-23 2019-06-21 复旦大学 It is a kind of based on aesthetic evaluation band refer to human eye image repair method
CN110020721A (en) * 2019-04-09 2019-07-16 武汉大学 A kind of target detection deep learning network optimized approach based on compression of parameters
CN110135336A (en) * 2019-05-14 2019-08-16 腾讯科技(深圳)有限公司 Training method, device and the storage medium of pedestrian's generation model
CN110189278A (en) * 2019-06-06 2019-08-30 上海大学 A kind of binocular scene image repair method based on generation confrontation network
CN110210514A (en) * 2019-04-24 2019-09-06 北京林业大学 Production fights network training method, image completion method, equipment and storage medium
CN110289927A (en) * 2019-07-01 2019-09-27 上海大学 The channel simulation implementation method of confrontation network is generated based on condition
CN110310247A (en) * 2019-07-05 2019-10-08 Oppo广东移动通信有限公司 Image processing method, device, terminal and computer readable storage medium
CN110517248A (en) * 2019-08-27 2019-11-29 北京百度网讯科技有限公司 Processing, training method, device and its equipment of eye fundus image
CN111062899A (en) * 2019-10-30 2020-04-24 湖北工业大学 Guidance-based blink video generation method for generating confrontation network
CN111080543A (en) * 2019-12-09 2020-04-28 Oppo广东移动通信有限公司 Image processing method and device, electronic equipment and computer readable storage medium
CN111105377A (en) * 2019-12-19 2020-05-05 西南石油大学 Method for repairing micro-resistivity imaging image
CN111353514A (en) * 2018-12-20 2020-06-30 马上消费金融股份有限公司 Model training method, image recognition method, device and terminal equipment
CN111861945A (en) * 2020-09-21 2020-10-30 浙江大学 Text-guided image restoration method and system
CN112674709A (en) * 2020-12-22 2021-04-20 泉州装备制造研究所 Amblyopia detection method based on anti-noise
CN112861578A (en) * 2019-11-27 2021-05-28 四川大学 Method for generating human face from human eyes based on self-attention mechanism
CN112992304A (en) * 2020-08-24 2021-06-18 湖南数定智能科技有限公司 High-resolution pinkeye case data generation method, equipment and storage medium
CN113344784A (en) * 2019-04-30 2021-09-03 达音网络科技(上海)有限公司 Optimizing supervised generation countermeasure networks through latent spatial regularization
CN113935928A (en) * 2020-07-13 2022-01-14 四川大学 Rock core image super-resolution reconstruction based on Raw format
CN115908205A (en) * 2023-02-21 2023-04-04 成都信息工程大学 Image restoration method and device, electronic equipment and storage medium
WO2023207515A1 (en) * 2022-04-29 2023-11-02 北京字跳网络技术有限公司 Image generation method and device, and storage medium and program product
CN117994173A (en) * 2024-04-07 2024-05-07 腾讯科技(深圳)有限公司 Repair network training method, image processing method, device and electronic equipment

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20080019571A1 (en) * 2006-07-20 2008-01-24 Harris Corporation Geospatial Modeling System Providing Non-Linear In painting for Voids in Geospatial Model Frequency Domain Data and Related Methods
CN107133934A (en) * 2017-05-18 2017-09-05 北京小米移动软件有限公司 Image completion method and device

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20080019571A1 (en) * 2006-07-20 2008-01-24 Harris Corporation Geospatial Modeling System Providing Non-Linear In painting for Voids in Geospatial Model Frequency Domain Data and Related Methods
CN107133934A (en) * 2017-05-18 2017-09-05 北京小米移动软件有限公司 Image completion method and device

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
BRIAN DOLHANSKY ET AL: ""Eye In-Painting with Exemplar Generative Adversarial Networks"", 《HTTPS://ARXIV.ORG/PDF/1712.03999.PDF》 *

Cited By (41)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109145745A (en) * 2018-07-20 2019-01-04 上海工程技术大学 A kind of face identification method under circumstance of occlusion
CN109145745B (en) * 2018-07-20 2022-02-11 上海工程技术大学 Face recognition method under shielding condition
CN109308689A (en) * 2018-10-15 2019-02-05 聚时科技(上海)有限公司 The unsupervised image repair method of confrontation network migration study is generated based on mask
CN109685724A (en) * 2018-11-13 2019-04-26 天津大学 A kind of symmetrical perception facial image complementing method based on deep learning
CN109712092A (en) * 2018-12-18 2019-05-03 上海中信信息发展股份有限公司 Archives scan image repair method, device and electronic equipment
CN111353514A (en) * 2018-12-20 2020-06-30 马上消费金融股份有限公司 Model training method, image recognition method, device and terminal equipment
CN109919830B (en) * 2019-01-23 2023-02-10 复旦大学 Method for restoring image with reference eye based on aesthetic evaluation
CN109919830A (en) * 2019-01-23 2019-06-21 复旦大学 It is a kind of based on aesthetic evaluation band refer to human eye image repair method
CN109919018A (en) * 2019-01-28 2019-06-21 浙江英索人工智能科技有限公司 Image eyes based on reference picture automatically open method and device
CN110020721A (en) * 2019-04-09 2019-07-16 武汉大学 A kind of target detection deep learning network optimized approach based on compression of parameters
CN110210514A (en) * 2019-04-24 2019-09-06 北京林业大学 Production fights network training method, image completion method, equipment and storage medium
CN113344784B (en) * 2019-04-30 2023-09-22 达音网络科技(上海)有限公司 Optimizing a supervisory generated countermeasure network through latent spatial regularization
CN113344784A (en) * 2019-04-30 2021-09-03 达音网络科技(上海)有限公司 Optimizing supervised generation countermeasure networks through latent spatial regularization
CN110135336B (en) * 2019-05-14 2023-08-25 腾讯科技(深圳)有限公司 Training method, device and storage medium for pedestrian generation model
CN110135336A (en) * 2019-05-14 2019-08-16 腾讯科技(深圳)有限公司 Training method, device and the storage medium of pedestrian's generation model
CN110189278A (en) * 2019-06-06 2019-08-30 上海大学 A kind of binocular scene image repair method based on generation confrontation network
CN110289927A (en) * 2019-07-01 2019-09-27 上海大学 The channel simulation implementation method of confrontation network is generated based on condition
CN110289927B (en) * 2019-07-01 2021-06-15 上海大学 Channel simulation realization method for generating countermeasure network based on condition
CN110310247A (en) * 2019-07-05 2019-10-08 Oppo广东移动通信有限公司 Image processing method, device, terminal and computer readable storage medium
CN110310247B (en) * 2019-07-05 2021-06-01 Oppo广东移动通信有限公司 Image processing method, device, terminal and computer readable storage medium
CN110517248A (en) * 2019-08-27 2019-11-29 北京百度网讯科技有限公司 Processing, training method, device and its equipment of eye fundus image
CN111062899A (en) * 2019-10-30 2020-04-24 湖北工业大学 Guidance-based blink video generation method for generating confrontation network
CN111062899B (en) * 2019-10-30 2023-02-17 湖北工业大学 Guidance-based blink video generation method for generating confrontation network
CN112861578A (en) * 2019-11-27 2021-05-28 四川大学 Method for generating human face from human eyes based on self-attention mechanism
CN112861578B (en) * 2019-11-27 2023-07-04 四川大学 Method for generating human face from human eyes based on self-attention mechanism
CN111080543B (en) * 2019-12-09 2024-03-22 Oppo广东移动通信有限公司 Image processing method and device, electronic equipment and computer readable storage medium
CN111080543A (en) * 2019-12-09 2020-04-28 Oppo广东移动通信有限公司 Image processing method and device, electronic equipment and computer readable storage medium
CN111105377B (en) * 2019-12-19 2022-05-06 西南石油大学 Method for repairing micro-resistivity imaging image
CN111105377A (en) * 2019-12-19 2020-05-05 西南石油大学 Method for repairing micro-resistivity imaging image
CN113935928A (en) * 2020-07-13 2022-01-14 四川大学 Rock core image super-resolution reconstruction based on Raw format
CN113935928B (en) * 2020-07-13 2023-04-11 四川大学 Rock core image super-resolution reconstruction based on Raw format
CN112992304A (en) * 2020-08-24 2021-06-18 湖南数定智能科技有限公司 High-resolution pinkeye case data generation method, equipment and storage medium
CN112992304B (en) * 2020-08-24 2023-10-13 湖南数定智能科技有限公司 High-resolution red eye case data generation method, device and storage medium
CN111861945B (en) * 2020-09-21 2020-12-18 浙江大学 Text-guided image restoration method and system
CN111861945A (en) * 2020-09-21 2020-10-30 浙江大学 Text-guided image restoration method and system
CN112674709A (en) * 2020-12-22 2021-04-20 泉州装备制造研究所 Amblyopia detection method based on anti-noise
WO2023207515A1 (en) * 2022-04-29 2023-11-02 北京字跳网络技术有限公司 Image generation method and device, and storage medium and program product
CN115908205B (en) * 2023-02-21 2023-05-30 成都信息工程大学 Image restoration method, device, electronic equipment and storage medium
CN115908205A (en) * 2023-02-21 2023-04-04 成都信息工程大学 Image restoration method and device, electronic equipment and storage medium
CN117994173A (en) * 2024-04-07 2024-05-07 腾讯科技(深圳)有限公司 Repair network training method, image processing method, device and electronic equipment
CN117994173B (en) * 2024-04-07 2024-06-11 腾讯科技(深圳)有限公司 Repair network training method, image processing method, device and electronic equipment

Similar Documents

Publication Publication Date Title
CN108269245A (en) A kind of eyes image restorative procedure based on novel generation confrontation network
CN109815893B (en) Color face image illumination domain normalization method based on cyclic generation countermeasure network
CN109815928B (en) Face image synthesis method and device based on counterstudy
Chen et al. Fsrnet: End-to-end learning face super-resolution with facial priors
CN108495110B (en) Virtual viewpoint image generation method based on generation type countermeasure network
JP7013077B2 (en) Biological detection methods and devices, equipment and storage media
CN105917353B (en) Feature extraction and matching for biological identification and template renewal
CN107944379B (en) Eye white image super-resolution reconstruction and image enhancement method based on deep learning
CN105069400B (en) Facial image gender identifying system based on the sparse own coding of stack
CN109886881B (en) Face makeup removal method
CN110555434A (en) method for detecting visual saliency of three-dimensional image through local contrast and global guidance
CN107886089A (en) A kind of method of the 3 D human body Attitude estimation returned based on skeleton drawing
CN112668519A (en) Abnormal face recognition living body detection method and system based on MCCAE network and Deep SVDD network
Rahman et al. A qualitative survey on deep learning based deep fake video creation and detection method
KR20180103835A (en) A method for authenticating and / or verifying the integrity of a subject
CN117095128A (en) Priori-free multi-view human body clothes editing method
CN114926892A (en) Fundus image matching method and system based on deep learning and readable medium
CN113033305A (en) Living body detection method, living body detection device, terminal equipment and storage medium
CN117975519A (en) Model training and image generating method and device, electronic equipment and storage medium
CN113326531B (en) Robust efficient distributed face image steganography method
CN114036553A (en) K-anonymity-combined pedestrian identity privacy protection method
CN109711286B (en) Control method and device based on artificial retina space perception
CN113538254A (en) Image restoration method and device, electronic equipment and computer readable storage medium
CN116110109A (en) Structure self-adaptive face identity information protection method based on identity deactivation
CN109523478A (en) Image removes grid method, 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
WW01 Invention patent application withdrawn after publication
WW01 Invention patent application withdrawn after publication

Application publication date: 20180710