CN108986041A - A kind of image recovery method, device, electronic equipment and readable storage medium storing program for executing - Google Patents
A kind of image recovery method, device, electronic equipment and readable storage medium storing program for executing Download PDFInfo
- Publication number
- CN108986041A CN108986041A CN201810608192.2A CN201810608192A CN108986041A CN 108986041 A CN108986041 A CN 108986041A CN 201810608192 A CN201810608192 A CN 201810608192A CN 108986041 A CN108986041 A CN 108986041A
- Authority
- CN
- China
- Prior art keywords
- image
- input
- probability
- input picture
- reconstructed image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 54
- 238000011084 recovery Methods 0.000 title claims abstract description 50
- 238000003860 storage Methods 0.000 title claims abstract description 17
- 230000001815 facial effect Effects 0.000 claims abstract description 132
- 238000004519 manufacturing process Methods 0.000 claims abstract description 65
- 238000003709 image segmentation Methods 0.000 claims abstract description 32
- 238000012549 training Methods 0.000 claims description 46
- 230000008447 perception Effects 0.000 claims description 28
- 238000004891 communication Methods 0.000 claims description 19
- 238000005516 engineering process Methods 0.000 claims description 16
- 238000004590 computer program Methods 0.000 claims description 11
- 230000008569 process Effects 0.000 claims description 11
- 230000004927 fusion Effects 0.000 claims description 8
- 239000000203 mixture Substances 0.000 claims description 5
- 230000004069 differentiation Effects 0.000 claims description 2
- 230000000694 effects Effects 0.000 abstract description 16
- 238000010586 diagram Methods 0.000 description 15
- 238000000513 principal component analysis Methods 0.000 description 8
- 238000012545 processing Methods 0.000 description 8
- 239000003086 colorant Substances 0.000 description 5
- 238000012986 modification Methods 0.000 description 5
- 230000004048 modification Effects 0.000 description 5
- 238000004422 calculation algorithm Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 230000002093 peripheral effect Effects 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 230000011218 segmentation Effects 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 238000000802 evaporation-induced self-assembly Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 238000004321 preservation Methods 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- 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
- G06T2207/10004—Still image; Photographic image
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20212—Image combination
- G06T2207/20221—Image fusion; Image merging
Abstract
The invention discloses a kind of image recovery method, device, electronic equipment and readable storage medium storing program for executing, this method comprises: facial image to be restored is input in Image Segmentation Model, determine the first area being blocked in the facial image；The first area is marked in the facial image determines the first input picture；First input picture is input in the generation model of production confrontation network, determines the first reconstructed image；First input picture and the first reconstructed image are input in the discrimination model of the production confrontation network, determine that first reconstructed image is the first probability of complete facial image；Judge first probability whether more than the first probability threshold value；If so, being restored according to first reconstructed image to the facial image；If not, being input to first reconstructed image as the first input picture in the generation model, until restoring to the facial image, the present invention can reach more preferably face recovery effects.
Description
Technical field
The present invention relates to technical field of face recognition more particularly to a kind of image recovery method, device, electronic equipments and can
Read storage medium.
Background technique
An important biomolecule feature of the face as human body, in recent years in image procossing, vision technique, the neck such as information security
There is increasingly important role in domain.Face recognition technology obtains huge development and progress in recent years, however part hides
Gear cause face information missing, seriously affected the extraction and identification of feature, be in recognition of face one it is extremely challenging
Problem.It is increasing for the demand of partial occlusion recognition of face with the fast development of face recognition technology, much it is used for people
Face image occlusion area recovery algorithms are suggested.
The method that common facial image occlusion area restores includes being based on PCA (Principal Component
Analysis, principal component analysis) face blocked area method for reconstructing, and to the stragetic innovation method of PCA method, such as FW-
PCA utilizes quick weighted principal component analyzing, removes algorithm and automatic more based on improved Gabor-PCA reconstruct face occluder
It is worth exposure mask PCA face reconstruction model etc..In face blocked area method for reconstructing based on PCA, first to block face original image into
Row singular value decomposition rebuilds covariance matrix and projects to eigenface space, the main component of face in original image is extracted, such as face
Face obtain a reconstruction face by the linear combination of eigenface, calculate the difference for rebuilding face and original image, later by fast
Speed weighting, Gabor filter and normalizes after as the probability being blocked, be that weight by original image and reconstruction is schemed to synthesize newly using this probability
Face.
But above-mentioned face blocked area method for reconstructing and its stragetic innovation method based on PCA, need to extract face
Five official rank main components, only block effectively small area, block discrimination decline to the large area as mask, and in face
What is extracted in feature extraction is five official rank main components of face, does not account for face manifold geometry, obtained feature is not
Enough robusts, therefore for the effect of face recovery, still there are a certain distance with real image in the prior art.
Summary of the invention
The present invention provides a kind of image recovery method, device, electronic equipment and readable storage medium storing program for executing, existing to solve
The problems in technology.
The present invention provides a kind of image recovery method, this method comprises:
By facial image to be restored, it is input in the Image Segmentation Model that training is completed in advance, determines the face figure
The first area being blocked as in；The first area is marked in the facial image, determines the first input picture；
By first input picture, it is input in the generation model for the production confrontation network that training is completed in advance, really
First reconstructed image of fixed first input picture；First input picture and first reconstructed image are input to institute
In the discrimination model for stating production confrontation network, determine that first reconstructed image is the first probability of complete facial image；
Judge whether first probability is more than preset first probability threshold value；
If so, being restored according to first reconstructed image to the facial image；
If not, being input in the generation model using first reconstructed image as the first input picture, until right
The facial image is restored.
Further, described that the first area is marked in the facial image, determine that the first input picture includes:
In the facial image, the pixel value of pixel in the first area is updated to presetted pixel value；
By the updated facial image of pixel value, it is determined as the first input picture.
Further, the generation model of training production confrontation network and the process of discrimination model include: in advance
For each sample image, the arbitrary region in the sample image is determined as being blocked in the sample image
Second area；The second area is marked in the sample image, determines the second input picture；By the second input figure
Picture is input in the generation model of production confrontation network, determines the second reconstructed image of second input picture；It will be described
Second input picture and second reconstructed image are input in the discrimination model of production confrontation network, determine described the
Two reconstructed images are the second probability of complete facial image；
It is when second probability is less than preset second probability threshold value, second reconstructed image is defeated as second
Enter image, be input in the generation model, the generation model and the discrimination model are alternately trained.
Further, described by the input picture, it is input in the generation model of production confrontation network, described in determination
The reconstructed image of input picture includes:
By the input picture, it is input in the generation model of production confrontation network, determines the puppet of the input picture
Image；
Third corresponding with region region is determined in the pseudo- image, and the area is replaced using the third region
Domain generates reconstructed image.
Further, it is determined that the probability of reconstructed image includes:
Formula is lost according to the input picture, the reconstructed image and the context pre-saved, determines that context loss is general
Rate；
Formula is lost according to the reconstructed image and the perception pre-saved, determines perception loss probability；
According to the context loss probability, the perception loss probability and the new probability formula pre-saved are determined described heavy
The probability of composition picture.
Further, described according to first reconstructed image, carrying out recovery to the facial image includes:
First reconstructed image is fitted by Poisson image fusion technology, the facial image is carried out extensive
It is multiple.
The present invention provides a kind of image recovery device, which includes:
First determining module, for by facial image to be restored, being input to the Image Segmentation Model that training is completed in advance
In, determine the first area being blocked in the facial image；The first area is marked in the facial image, determines
One input picture；
Second determining module, for by first input picture, being input to the production confrontation net that training is completed in advance
In the generation model of network, the first reconstructed image of first input picture is determined；By first input picture and described
One reconstructed image is input in the discrimination model of the production confrontation network, determines that first reconstructed image is complete face
First probability of image；
Judgment module, for judging whether first probability is more than preset first probability threshold value；
Recovery module, for when the judging result of the judgment module be when, according to first reconstructed image, to institute
Facial image is stated to be restored；When the judging result of the judgment module is no, using first reconstructed image as first
Input picture is input in the generation model, until restoring to the facial image.
Further, first determining module is specifically used in the facial image, by picture in the first area
The pixel value of vegetarian refreshments is updated to presetted pixel value；By the updated facial image of pixel value, it is determined as the first input picture.
Further, second determining module was also used to for each sample image, by appointing in the sample image
Meaning region is determined as the second area being blocked in the sample image；The second area is marked in the sample image,
Determine the second input picture；It is input in the generation model of production confrontation network, second input picture described in determination
Second reconstructed image of the second input picture；Second input picture and second reconstructed image are input to the generation
Formula is fought in the discrimination model of network, determines that second reconstructed image is the second probability of complete facial image；When described
When two probability are less than preset second probability threshold value, using second reconstructed image as the second input picture, it is input to institute
It states and generates in model, the generation model and the discrimination model are alternately trained.
Further, second determining module is specifically used for the input picture being input to production confrontation network
Generation model in, determine the pseudo- image of the input picture；Third corresponding with the region is determined in the pseudo- image
The Area generation reconstructed image is replaced using the third region in region.
Further, second determining module, be specifically used for according to the input picture, the reconstructed image and in advance
The context of preservation loses formula, determines context loss probability；Formula is lost according to the reconstructed image and the perception pre-saved,
Determine perception loss probability；According to the context loss probability, the perception loss probability and the new probability formula pre-saved, really
The probability of the fixed reconstructed image.
Further, the recovery module is specifically used for through Poisson image fusion technology to first reconstructed image
It is fitted, the facial image is restored.
The present invention provides a kind of electronic equipment, comprising: processor, communication interface, memory and communication bus, wherein place
Device, communication interface are managed, memory completes mutual communication by communication bus；
It is stored with computer program in the memory, when described program is executed by the processor, so that the place
Manage the step of device executes any of the above-described the method.
The present invention provides a kind of computer readable storage medium, is stored with the computer journey that can be executed by electronic equipment
Sequence, when described program is run on the electronic equipment, so that the electronic equipment executes any of the above-described the method
Step.
The present invention provides a kind of image recovery method, device, electronic equipment and readable storage medium storing program for executing, this method comprises:
By facial image to be restored, it is input in the Image Segmentation Model that training is completed in advance, determines and hidden in the facial image
The first area of gear；The first area is marked in the facial image, determines the first input picture；Described first is inputted
Image is input in the generation model for the production confrontation network that training is completed in advance, determines the of first input picture
One reconstructed image；First input picture and first reconstructed image are input to the differentiation of the production confrontation network
In model, determine that first reconstructed image is the first probability of complete facial image；Judge first probability whether be more than
Preset first probability threshold value；If so, being restored according to first reconstructed image to the facial image；If not,
Using first reconstructed image as the first input picture, be input in the generation model, until to the facial image into
Row restores.Image Segmentation Model identification facial image can identify large area occlusion area, and image segmentation mould in the present invention
It is usually the pixel for considering each position in image in type, the feature of available more robust is conducive to production and fights
Identification and recovery of the network to facial image, therefore more preferably face recovery effects can be reached.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
It obtains other drawings based on these drawings.
Fig. 1 is a kind of schematic diagram for image recovery process that present example 1 provides；
Fig. 2 is the facial image schematic diagram to be restored that the embodiment of the present invention 1 provides；
Fig. 3 is the facial image schematic diagram after the mark that the embodiment of the present invention 1 provides；
Fig. 4 be the embodiment of the present invention 4 provide restored using image recovery method after facial image；
Fig. 5 is a kind of schematic diagram for image recovery process that the embodiment of the present invention 5 provides；
Fig. 6 is the structural schematic diagram for a kind of electronic equipment that the embodiment of the present invention 6 provides；
Fig. 7 is a kind of image recovery device schematic diagram provided in an embodiment of the present invention.
Specific embodiment
In order to reach more preferably face recovery effects, the embodiment of the invention provides a kind of image recovery method, device, electricity
Sub- equipment and readable storage medium storing program for executing.
To make the objectives, technical solutions, and advantages of the present invention clearer, make below in conjunction with the attached drawing present invention into one
Step ground detailed description, it is clear that described embodiment is only a part of the embodiments of the present invention, rather than whole implementation
Example.Based on the embodiments of the present invention, obtained by those of ordinary skill in the art without making creative efforts
Every other embodiment, shall fall within the protection scope of the present invention.
Embodiment 1:
Fig. 1 be a kind of schematic diagram of image recovery process provided in an embodiment of the present invention, the process the following steps are included:
S101: it by facial image to be restored, is input in the Image Segmentation Model that training is completed in advance, determines the people
The first area being blocked in face image；The first area is marked in the facial image, determines the first input picture.
Image recovery method provided in an embodiment of the present invention is applied to electronic equipment, and electronic equipment can be Desktop Computing
Machine, portable computer, smart phone, tablet computer, personal digital assistant (Personal Digital Assistant,
PDA), the electronic equipments such as server.
Electronic equipment is available to arrive facial image to be restored, there is first to be blocked in facial image to be restored
Region is also possible to user or other equipment wherein facial image to be restored can be and be pre-reserved in electronic equipment
It is input in the electronic equipment.
The Image Segmentation Model that training is completed in advance is preserved in electronic equipment, electronic equipment is by face figure to be restored
Picture is input in Image Segmentation Model, can determine the first area being blocked in facial image.It is illustrated in figure 2 to be restored
Facial image, the corresponding region of mask is the first area being blocked in facial image in Fig. 2.
Image Segmentation Model is using the Image Segmentation Model provided in the prior art.Due to true by deep learning network
Fixed Image Segmentation Model when performing image segmentation, can be flattened original two-dimensional matrix by last several layers of full articulamentum
At one-dimensional, to lose spatial information, therefore preferably the Image Segmentation Model in the embodiment of the present invention can be using based on complete
The Image Segmentation Model of the face occlusion area of convolutional network will applied to the Image Segmentation Model based on full convolutional network
Full articulamentum in deep learning network replaces with convolutional layer, so as to recover belonging to each pixel from abstract feature
Classification, after multiple convolution, facial image is smaller and smaller, and resolution ratio is lower and lower, is conducive to image segmentation, by anti-
Convolution is available and the big segmentation figure such as original image, so as to not lose image while guaranteeing preferable segmentation effect
Spatial information.
After electronic equipment determines the first area being blocked in facial image, first facial image can be labeled,
First area is marked in facial image, area can be subject to the region not being blocked and the first area being blocked with not isolabeling
Point, for example, by using different Fill Colors, or use different filling symbols.Electronic equipment marks in facial image
Behind one region, the region not being blocked and the first area being blocked can be determined according to different labels, so that it is determined that face
The information for needing to retain in image, and need to fill the information of completion.
After marking first area in facial image, determines the first input picture, i.e., determine the facial image after mark
For the first input picture.
Fig. 3 is the facial image schematic diagram after mark, i.e. the first input picture, black region is background, and gray area is
The region not being blocked in facial image, white area are the first area being blocked in facial image.
S102: by first input picture, it is input to the generation model for the production confrontation network that training is completed in advance
In, determine the first reconstructed image of first input picture；First input picture and first reconstructed image is defeated
Enter into the discrimination model of production confrontation network, determines that first reconstructed image is complete facial image first is general
Rate.
The production confrontation network that training is completed in advance is preserved in electronic equipment, it includes generating in network that production, which is fought,
Model and discrimination model.It can be DCGAN (Deep Convolutional Generative that production, which fights network,
Adversarial Networks, depth convolution production fight network).
After electronic equipment determines the first input picture, the first input picture is input to and is generated in model, by generating mould
Type generates the first reconstructed image of the first input picture.
Since electronic equipment can determine the information for needing to retain in facial image according to different labels, and need to fill
The information of completion, therefore the first reconstructed image can be understood as carrying out the information for needing to fill completion in the first input picture
Image after filling completion.
After electronic equipment gets the first reconstructed image for generating model generation, by the first input picture and the first reconstruct image
As being input in discrimination model, determine that the first reconstructed image is the first probability of complete facial image by discrimination model.
S103: judge whether first probability is more than preset first probability threshold value；If so, carrying out step S104；
If not, carrying out step S105.
Electronic equipment judges whether the first probability is more than preset first probability threshold value, is executed according to different judging results
Corresponding step.
First probability threshold value pre-saves in the electronic device, and the first probability threshold value is the number not less than 0 and no more than 1
Value, such as the first probability threshold value can be 0.5,0.7 etc..
S104: according to first reconstructed image, the facial image is restored.
If the first probability is more than the first probability threshold value, then it is assumed that current first reconstructed image is true complete face figure
The probability of picture is higher, can be using the first current reconstructed image as the recovery image of facial image, then can be according to the first weight
Composition picture, restores facial image.
According to the first reconstructed image, facial image is restored the first reconstructed image after filling completion can be made to become
Smoothly, with closer to a true facial image.
S105: using first reconstructed image as the first input picture, S102 is returned, the generation model is input to
In, until restoring to the facial image.
If the first probability is no more than the first probability threshold value, then it is assumed that current first reconstructed image and true complete face
The difference of image is larger, it is believed that the first current reconstructed image is not the recovery image of facial image, then by the first reconstruct
Image returns to S102 as the first input picture, is input to and generates in model, continues to generate the first new reconstructed image, until
First probability of the first new reconstructed image is more than the first probability threshold value, according to the first new reconstructed image, to facial image into
Row restores.
Since Image Segmentation Model identification facial image can identify large area occlusion area in the embodiment of the present invention, and
It is usually the pixel for considering each position in image in Image Segmentation Model, the feature of available more robust is conducive to
Production fights identification and recovery of the network to facial image, therefore can reach more preferably face recovery effects.
Embodiment 2:
On the basis of the above embodiments, described that described first is marked in the facial image in the embodiment of the present invention
Region determines that the first input picture includes:
In the facial image, the pixel value of pixel in the first area is updated to presetted pixel value；
By the updated facial image of pixel value, it is determined as the first input picture.
The region not being blocked and the first area being blocked are distinguish as label using different Fill Colors,
The first area being blocked clearly intuitively more can be embodied, need to fill benefit so that electronic equipment be made more accurately to determine
Full information, to reach more face recovery effects.
In electronic equipment in facial image, the pixel value of pixel in first area is updated to presetted pixel value, it will
The updated facial image of pixel value, is determined as the first input picture, to reach using different colors to the area not being blocked
The effect that domain and the first area being blocked are distinguished.
Presetted pixel value can be the specific pixel value in selection, be also possible to the pixel determining by certain calculating
Value.
If presetted pixel value is the specific pixel value chosen, specific pixel value can be chosen between 0 to 255 and is made
For presetted pixel value.
If presetted pixel value is the pixel value determining by certain calculating, image segmentation information can be obtained according to
Two-value mask M is taken, 1 indicates the information for needing to retain in M, i.e., the region not being blocked, 0 indicates the information for needing to fill, i.e.,
The mask value of first area is assigned 0 by the first area being blocked, and the mask value in remaining region assigns 1, completes to picture in first area
The pixel value of the update of the pixel value of vegetarian refreshments, the pixel in region not being blocked in updated facial image keeps preimage
Element value is constant, and the pixel value for the pixel in first area being blocked is updated to 0.
Due to the region not being blocked and being blocked using different Fill Colors as label in the embodiment of the present invention
First area be distinguish, more intuitively the first area being blocked clearly can be embodied, to make electronic equipment more
The information for needing to fill completion is accurately determined, to reach more face recovery effects.
Embodiment 3:
On the basis of the various embodiments described above, in the embodiment of the present invention, training production fights the generation mould of network in advance
The process of type and discrimination model includes:
For each sample image, the arbitrary region in the sample image is determined as being blocked in the sample image
Second area；The second area is marked in the sample image, determines the second input picture；By the second input figure
Picture is input in the generation model of production confrontation network, determines the second reconstructed image of second input picture；It will be described
Second input picture and second reconstructed image are input in the discrimination model of production confrontation network, determine described the
Two reconstructed images are the second probability of complete facial image；
It is when second probability is less than preset second probability threshold value, second reconstructed image is defeated as second
Enter image, be input in the generation model, the generation model and the discrimination model are alternately trained.
It is in the generation model and discrimination model training for fighting network to production, region any one in sample image is true
It is set to the second area being blocked in sample image, avoids limitation caused by the second area being blocked in sample image,
So that the generation model for the production confrontation network that training is completed and the application range of discrimination model are more extensive, to reach more
Good face recovery effects.
Since data volume is larger during model training, the electronic equipment for being accordingly used in model training, which can also use, to be met
The stronger electronic equipment of the computing capability of big data deep learning.
In the training process for generating model and discrimination model of production confrontation network, for each sample image to life
It is identical with the training process of discrimination model at model, therefore only said by taking a sample image as an example in the embodiment of the present invention
It is bright.
Electronic equipment can be direct chosen area when choosing arbitrary region in sample image, the area that will directly choose
Domain is determined as the second area being blocked in sample image, is also possible to choose any pixel in sample image, will be any
The arbitrary region is determined as the secondth area being blocked in sample image as the arbitrary region chosen by the region that pixel is constituted
Domain etc..
Second area is marked in sample image, determines the second input picture i.e. image missing sample, electronic equipment can be with
The region not being blocked and the region being blocked are distinguish with not isolabeling, for example, by using different Fill Colors, or
It is using different filling symbols, to determine the region not being blocked in image missing sample according to different labels and be hidden
The second area of gear, to carry out completion to second area.Second area is marked in sample image determines the second input picture
Method, from facial image mark first area determine the first input picture method can with it is identical can be different, usual feelings
Under condition, the second area method that determines the second input picture is marked in sample image, and the firstth area is marked in facial image
Domain determines that the method for the first input picture is identical.
It mainly includes two parts that production, which fights network (GAN, Generative Adversarial Networks): raw
Grow up to be a useful person (generator) generate model and arbiter (discriminator) i.e. discrimination model.Generator is mainly according to figure
As missing sample carries out image completion, it can be understood as the sample of same distribution is generated from training data, and arbiter is then
The image for differentiating input is that real human face image still generates image, i.e., judgement input is that truthful data or generator generate
Data, when arbiter can not differentiate whether input is generator picture generated, then it is assumed that the generator that training is completed
The image of approaching to reality picture can be generated.
After electronic equipment determines the second input picture, the second input picture is input to after generating model, by generating mould
Type generates the second reconstructed image of the second input picture.After electronic equipment gets the second reconstructed image for generating model generation,
Second input picture and the second reconstructed image are input in discrimination model, determine that the second reconstructed image has been by discrimination model
Second probability of whole facial image.
When the second probability is more than preset second probability threshold value, determine that the training for the second input picture is completed, after
Continuous remaining second input picture generated for sample image is alternately trained to model and discrimination model is generated.
When the second probability is no more than preset second probability threshold value, using the second reconstructed image as the second input picture,
It is input to and generates in model, continue to continue to generate the second new reconstructed image, directly to the alternately training of model and discrimination model is generated
The second probability to the second new reconstructed image is more than the second probability threshold value, determines that the training for the second input picture is completed,
Continuing with remaining second input picture that sample image generates, alternately trained to model and discrimination model is generated.
Second probability threshold value pre-saves in the electronic device, and the second probability threshold value is the number not less than 0 and no more than 1
Value, such as the second probability threshold value can be 0.5,0.7 etc., the second probability threshold value and the first probability threshold value can with it is identical can not
Together.
The embodiment of the present invention is on the basis of existing GAN, can also be using following behaviour in order to reach the training effect more having
Make: all pooling (pond) layers are replaced using stride convolution sum micro-stepping width convolution, in generating model and discrimination model
Using batch regularization, LeakyRelu activation primitive is used in all layers of discrimination model, passes through depth convolution production pair
Anti- network DCGAN is trained LFW (Labeled Faces in the Wild, recognition of face public data collection) data set,
Obtain the model that approaching to reality face can be generated.
Depth convolution production, which fights network, to be realized by minimax strategy shown in following formula.
X indicates true picture, and true picture can be sample image or the second reconstructed image, x~p at this timedata(x) it indicates
Image interval to be trained, z indicate noise i.e. the second input picture of input G network, z~pz(z) it indicates and sample image pair
The second input picture answered, and G (z) indicates that G network is directed to the second reconstructed image that the second input picture generates, E indicates mathematics
It is expected that D (x) indicates that D network judges the whether true probability of true picture, it is the input of sample image for true picture, makes
D (x) be the bigger the better, ideally input true picture be sample image when D (x) be 1.D (G (z)) indicates that D network is sentenced
Disconnected G generates the whether true probability of picture, this probability is bigger indicate picture that generator generates closer to true picture,
Exactly minimize (1-D (G (z)).First step training arbiter, V (G, D) are the bigger the better first, so be to add stochastic gradient,
Then second step training generator, V (G, D) is the smaller the better, so be to subtract stochastic gradient, to the entire of arbiter and generator
Training process is alternately.
To the training termination condition of the generation model and discrimination model of production confrontation network, with training in the prior art
Termination condition is identical, in embodiments of the present invention without limitation.
Face occluder minimizing technology and its stragetic innovation side based on sparse expression can also be commonly used in the prior art
Method carries out face recovery, and the face based on sparse expression blocks in minimizing technology, according to the principle of lack sampling, a small amount of limited
The thought of original signal is recovered under information state, sampling has the image blocked, carries out restoration and reconstruction to obtained information, obtains
The facial image not blocked.But needing dictionary to construct different types of block, application scenarios more limit to, and part hides
The sparse expression algorithm of gear assumes to indicate mutually indepedent between coefficient, and the face of recovery is not fitted original image well, due to
In the embodiment of the present invention in the generation model and discrimination model training for fighting network to production, any one region conduct is chosen
The second area being blocked in sample image, and it is not involved with hypothesis expression coefficient in training process, therefore can be fine
Ground avoids limitation.
Due in the embodiment of the present invention to production confrontation network generation model and discrimination model training when, by sample
Any one region is determined as the second area being blocked in sample image in image, avoids second to be blocked in sample image
Limitation caused by region, so that the application range of the generation model for the production confrontation network that training is completed and discrimination model is more
Add extensively, to reach more preferably face recovery effects.
Embodiment 4:
It is described by the input picture in the embodiment of the present invention on the basis of the various embodiments described above, it is input to production
In the generation model for fighting network, determine that the reconstructed image of the input picture includes:
By the input picture, it is input in the generation model of production confrontation network, determines the puppet of the input picture
Image；
Third corresponding with region region is determined in the pseudo- image, and the area is replaced using the third region
Domain generates reconstructed image.
It is described according to first reconstructed image, the facial image restore include:
First reconstructed image is fitted by Poisson image fusion technology, the facial image is carried out extensive
It is multiple.
After input picture is input to generation model by electronic equipment, it may be implemented by generation model true according to input picture
Determine the corresponding pseudo- image of input picture.Since electronic equipment can be with not isolabeling to the region not being blocked and the area being blocked
Domain is distinguish, therefore generation model can determine third region corresponding with the region being blocked in pseudo- image, using third
Region is replaced the region being blocked in input picture, generates reconstructed image.
When electronic equipment is replaced the region being blocked in input picture using third region, it can be in pseudo- image
Middle to extract the corresponding subgraph in third region, the corresponding subgraph in region that will be blocked in the input image extracts
Come, the corresponding subgraph in third region is filled into input picture and is blocked in region, completion input picture generates reconstruct image
Picture.
The image after completion can be made to become more smooth by Poisson image fusion technology, close to normal face
Image.
Fig. 4 be using it is provided in an embodiment of the present invention restored using image recovery method after facial image, as shown in Figure 4
From left to right (left and right as shown in Figure 4) be followed successively by facial image to be restored, the reconstructed image that obtains for the first time, second
Facial image after the reconstructed image and recovery that arrive.
Due in the embodiment of the present invention, reconstructed image is determined according to the pseudo- image that model generates is generated, to meeting condition
Reconstructed image is fitted by Poisson image fusion technology, so as to obtain more preferably image recovery effects.
Embodiment 5:
On the basis of the various embodiments described above, in the embodiment of the present invention, determine that the probability of reconstructed image includes:
Formula is lost according to the input picture, the reconstructed image and the context pre-saved, determines that context loss is general
Rate；
Formula is lost according to the reconstructed image and the perception pre-saved, determines perception loss probability；
According to the context loss probability, the perception loss probability and the new probability formula pre-saved are determined described heavy
The probability of composition picture.
Context is lost and perceived loss portfolio, can determine the probability of reconstructed image, so that it is determined that suitable reconstruct image
Picture.
Electronic equipment is got generate the reconstructed image that model is generated according to input picture after, by input picture and reconstruct image
As being input in discrimination model, context loss is determined when determining that reconstructed image is the probability of facial image by discrimination model first
Probability and perception loss probability, then by context loss probability and perception loss probability combination, determine the probability of reconstructed image.
Electronic equipment loses formula according to input picture, reconstructed image and the context pre-saved, determines that context loss is general
Rate specifically includes: according to input picture, reconstructed image and formula ηcontextual(z)=| | M × G (z)-M × y | |, determine context
Loss probability, ηcontextualIt (z) is context loss probability, z is the noise for being input to the discrimination model of production confrontation network, G
It (z) is reconstructed image, y is input picture, and M is corresponding two-value mask.
Electronic equipment loses formula according to reconstructed image and the perception pre-saved, determines that perception loss probability specifically wraps
It includes: according to the pseudo- image and formula ηperceptual(z)=log (1-D (G (z))) determines perception loss probability, ηperceptual
It (z) is perception loss probability, D (G (z)) is that discrimination model judges the whether true probability of reconstructed image.
For electronic equipment according to context loss probability, the new probability formula for perceiving loss probability and pre-saving determines reconstruct image
The probability of picture specifically includes: according to context loss probability, perceiving loss probability and formula
Z*=arg min (ηcontextual(z)+ηperceptual(z)), z* is matching probability.
The various embodiments described above are illustrated with a specific embodiment below, as shown in Figure 5:
S501: will be blocked face test image, be input in depth convolved image parted pattern.
Facial image that will be to be replied is input in the depth convolved image parted pattern that training is completed in advance.
Depth convolved image parted pattern can be trained by the image segmentation sample with markup information, it is described logical
It crosses the process that the image segmentation sample with markup information is trained depth convolved image parted pattern and belongs to the prior art,
It is not repeated them here in the embodiment of the present invention.
S502: the region that is blocked is determined, and pixel is gone in the region that will be blocked.
The first area being blocked in facial image is determined by depth convolved image parted pattern, in facial image
First area is marked by the way of two-value mask, determines the first input picture.
S503: fighting network using production, generates pseudo- image, region determines reconstructed image for being blocked.
First input picture is input to the generation model for the production confrontation network that training is completed in advance, generates model
The pseudo- image for generating the first input picture, using third corresponding with first area region in pseudo- image, replacement the first input figure
First area as in generates the first reconstructed image.
S504: optimal reconstructed image is chosen based on perception loss and context loss.
The first reconstructed image is input in the discrimination model in production confrontation network, determines the by discrimination model
The perception loss probability and context loss probability of one reconstructed image, according to perception loss probability and context loss probability combination,
The first probability of the first reconstructed image is determined, if the first probability is more than the first probability threshold value, it is determined that the first reconstructed image is
Optimal reconstructed image；If the first probability is no more than the first probability threshold value, using the first reconstructed image as the first input picture,
It is input to and generates in model, continue to generate the first new reconstructed image, until the first of the first new reconstructed image generated is general
Rate is more than the first probability threshold value, determines that the first new reconstructed image of the generation is optimal reconstructed image.
S505: by Poisson image fusion technology, optimal reconstructed image is fitted.
S506: the restored image for the face test image that is blocked is generated.
Due to that, by context loss and perception loss portfolio, can determine the probability of reconstructed image in the embodiment of the present invention, from
And determine suitable reconstructed image.
Embodiment 6:
On the basis of the various embodiments described above, the embodiment of the invention also provides a kind of electronic equipment, as shown in fig. 6, packet
It includes: processor 601, communication interface 602, memory 603 and communication bus 604, wherein processor 601, communication interface 602 are deposited
Reservoir 603 completes mutual communication by communication bus 604；
It is stored with computer program in the memory 603, when described program is executed by the processor 601, so that
The processor 601 executes following steps:
By facial image to be restored, it is input in the Image Segmentation Model that training is completed in advance, determines the face figure
The first area being blocked as in；The first area is marked in the facial image, determines the first input picture；
By first input picture, it is input in the generation model for the production confrontation network that training is completed in advance, really
First reconstructed image of fixed first input picture；First input picture and first reconstructed image are input to institute
In the discrimination model for stating production confrontation network, determine that first reconstructed image is the first probability of complete facial image；
Judge whether first probability is more than preset first probability threshold value；
If so, being restored according to first reconstructed image to the facial image；
If not, being input in the generation model using first reconstructed image as the first input picture, until right
The facial image is restored.
Electronic equipment provided in an embodiment of the present invention is specifically as follows desktop computer, server, network side equipment etc..
The communication bus that above-mentioned electronic equipment is mentioned can be Peripheral Component Interconnect standard (Peripheral Component
Interconnect, PCI) bus or expanding the industrial standard structure (Extended Industry Standard
Architecture, EISA) bus etc..The communication bus can be divided into address bus, data/address bus, control bus etc..For just
It is only indicated with a thick line in expression, figure, it is not intended that an only bus or a type of bus.
Communication interface 602 is for the communication between above-mentioned electronic equipment and other equipment.
Memory may include random access memory (Random Access Memory, RAM), also may include non-easy
The property lost memory (Non-Volatile Memory, NVM), for example, at least a magnetic disk storage.Optionally, memory may be used also
To be storage device that at least one is located remotely from aforementioned processor.
Above-mentioned processor can be general processor, including central processing unit, network processing unit (Network
Processor, NP) etc.；It can also be digital command processor (Digital Signal Processing, DSP), dedicated collection
At circuit, field programmable gate array or other programmable logic device, discrete gate or transistor logic, discrete hard
Part component etc..
In embodiments of the present invention, Image Segmentation Model identification facial image can identify large area occlusion area, and
It is usually the pixel for considering each position in image in Image Segmentation Model, the feature of available more robust is conducive to
Production fights identification and recovery of the network to facial image, therefore can reach more preferably face recovery effects.
Embodiment 7:
On the basis of the various embodiments described above, the embodiment of the invention also provides a kind of computers to store readable storage medium
Matter is stored with the computer program that can be executed by electronic equipment in the computer readable storage medium, when described program is in institute
It states when being run on electronic equipment, so that the electronic equipment realizes following steps when executing:
By facial image to be restored, it is input in the Image Segmentation Model that training is completed in advance, determines the face figure
The first area being blocked as in；The first area is marked in the facial image, determines the first input picture；
By first input picture, it is input in the generation model for the production confrontation network that training is completed in advance, really
First reconstructed image of fixed first input picture；First input picture and first reconstructed image are input to institute
In the discrimination model for stating production confrontation network, determine that first reconstructed image is the first probability of complete facial image；
Judge whether first probability is more than preset first probability threshold value；
If so, being restored according to first reconstructed image to the facial image；
If not, being input in the generation model using first reconstructed image as the first input picture, until right
The facial image is restored.
Above-mentioned computer readable storage medium can be any usable medium that the processor in electronic equipment can access
Or data storage device, including but not limited to magnetic storage such as floppy disk, hard disk, tape, magneto-optic disk (MO) etc., optical memory
Such as CD, DVD, BD, HVD and semiconductor memory such as ROM, EPROM, EEPROM, nonvolatile memory (NAND
FLASH), solid state hard disk (SSD) etc..
Image Segmentation Model identification facial image can identify large area occlusion area in embodiments of the present invention, and scheme
As being usually the pixel for considering each position in image, the feature of available more robust, conducive to life in parted pattern
An accepted way of doing sth fights identification and recovery of the network to facial image, therefore can reach more preferably face recovery effects.
Fig. 7 is a kind of image recovery device schematic diagram provided in an embodiment of the present invention, which is applied to electronic equipment, should
Device includes:
First determining module 71, for by facial image to be restored, being input to the image segmentation mould that training is completed in advance
In type, the first area being blocked in the facial image is determined；The first area is marked in the facial image, is determined
First input picture；
Second determining module 72, for by first input picture, being input to the production confrontation that training is completed in advance
In the generation model of network, the first reconstructed image of first input picture is determined；By first input picture and described
First reconstructed image is input in the discrimination model of the production confrontation network, determines that first reconstructed image is whole person
First probability of face image；
Judgment module 73, for judging whether first probability is more than preset first probability threshold value；
Recovery module 74, for when the judging result of the judgment module 73, which is, is, according to first reconstructed image,
The facial image is restored；When the judging result of the judgment module 73 is no, first reconstructed image is made
It for the first input picture, is input in the generation model, until restoring to the facial image.
First determining module 71 is specifically used in the facial image, by pixel in the first area
Pixel value is updated to presetted pixel value；By the updated facial image of pixel value, it is determined as the first input picture.
Second determining module 72 is also used to for each sample image, by the arbitrary region in the sample image
It is determined as the second area being blocked in the sample image；The second area is marked in the sample image, determines
Two input pictures；By second input picture, it is input in the generation model of production confrontation network, determines that described second is defeated
Enter the second reconstructed image of image；Second input picture and second reconstructed image are input to the production confrontation
In the discrimination model of network, determine that second reconstructed image is the second probability of complete facial image；When second probability
When being less than preset second probability threshold value, using second reconstructed image as the second input picture, it is input to the generation
In model, the generation model and the discrimination model are alternately trained.
Second determining module 72 is specifically used for for the input picture being input to the generation of production confrontation network
In model, the pseudo- image of the input picture is determined；Third corresponding with region region is determined in the pseudo- image, is adopted
The Area generation reconstructed image is replaced with the third region.
Second determining module, specifically for according to the input picture, the reconstructed image and the language pre-saved
Formula is lost in border, determines context loss probability；Formula is lost according to the reconstructed image and the perception pre-saved, determines perception
Loss probability；According to the context loss probability, the perception loss probability and the new probability formula pre-saved are determined described heavy
The probability of composition picture.
The recovery module 74, specifically for being intended by Poisson image fusion technology first reconstructed image
It closes, the facial image is restored.
Since Image Segmentation Model identification facial image can identify large area occlusion area in the embodiment of the present invention, and
It is usually the pixel for considering each position in image in Image Segmentation Model, the feature of available more robust is conducive to
Production fights identification and recovery of the network to facial image, therefore can reach more preferably face recovery effects.
For systems/devices embodiment, since it is substantially similar to the method embodiment, so the comparison of description is simple
Single, the relevent part can refer to the partial explaination of embodiments of method.
It should be understood by those skilled in the art that, embodiments herein can provide as method, system or computer program
Product.Therefore, complete hardware embodiment, complete software embodiment or reality combining software and hardware aspects can be used in the application
Apply the form of example.Moreover, it wherein includes the computer of computer usable program code that the application, which can be used in one or more,
The computer program implemented in usable storage medium (including but not limited to magnetic disk storage, CD-ROM, optical memory etc.) produces
The form of product.
The application is referring to method, the process of equipment (system) and computer program product according to the embodiment of the present application
Figure and/or block diagram describe.It should be understood that every one stream in flowchart and/or the block diagram can be realized by computer program instructions
The combination of process and/or box in journey and/or box and flowchart and/or the block diagram.It can provide these computer programs
Instruct the processor of general purpose computer, special purpose computer, Embedded Processor or other programmable data processing devices to produce
A raw machine, so that being generated by the instruction that computer or the processor of other programmable data processing devices execute for real
The device for the function of being specified in present one or more flows of the flowchart and/or one or more blocks of the block diagram.
These computer program instructions, which may also be stored in, is able to guide computer or other programmable data processing devices with spy
Determine in the computer-readable memory that mode works, so that it includes referring to that instruction stored in the computer readable memory, which generates,
Enable the manufacture of device, the command device realize in one box of one or more flows of the flowchart and/or block diagram or
The function of being specified in multiple boxes.
These computer program instructions also can be loaded onto a computer or other programmable data processing device, so that counting
Series of operation steps are executed on calculation machine or other programmable devices to generate computer implemented processing, thus in computer or
The instruction executed on other programmable devices is provided for realizing in one or more flows of the flowchart and/or block diagram one
The step of function of being specified in a box or multiple boxes.
Although the preferred embodiment of the application has been described, it is created once a person skilled in the art knows basic
Property concept, then additional changes and modifications can be made to these embodiments.So it includes excellent that the following claims are intended to be interpreted as
It selects embodiment and falls into all change and modification of the application range.
Obviously, various changes and modifications can be made to the invention without departing from essence of the invention by those skilled in the art
Mind and range.In this way, if these modifications and changes of the present invention belongs to the range of the claims in the present invention and its equivalent technologies
Within, then the present invention is also intended to include these modifications and variations.
Claims (14)
1. a kind of image recovery method, which is characterized in that this method comprises:
By facial image to be restored, it is input in the Image Segmentation Model that training is completed in advance, determines in the facial image
The first area being blocked；The first area is marked in the facial image, determines the first input picture；
By first input picture, it is input in the generation model for the production confrontation network that training is completed in advance, determines institute
State the first reconstructed image of the first input picture；First input picture and first reconstructed image are input to the life
An accepted way of doing sth is fought in the discrimination model of network, determines that first reconstructed image is the first probability of complete facial image；
Judge whether first probability is more than preset first probability threshold value；
If so, being restored according to first reconstructed image to the facial image；
If not, being input in the generation model using first reconstructed image as the first input picture, until to described
Facial image is restored.
2. the method as described in claim 1, which is characterized in that it is described that the first area is marked in the facial image,
Determine that the first input picture includes:
In the facial image, the pixel value of pixel in the first area is updated to presetted pixel value；
By the updated facial image of pixel value, it is determined as the first input picture.
3. the method as described in claim 1, which is characterized in that the generation model of training production confrontation network and differentiation in advance
The process of model includes:
For each sample image, the arbitrary region in the sample image is determined as be blocked in the sample image
Two regions；The second area is marked in the sample image, determines the second input picture；By second input picture,
It is input in the generation model of production confrontation network, determines the second reconstructed image of second input picture；By described
Two input pictures and second reconstructed image are input in the discrimination model of the production confrontation network, determine described second
Reconstructed image is the second probability of complete facial image；
When second probability is less than preset second probability threshold value, scheme second reconstructed image as the second input
Picture is input in the generation model, is alternately trained to the generation model and the discrimination model.
4. method as claimed in claim 1 or 3, which is characterized in that it is described by the input picture, it is input to production confrontation
In the generation model of network, determine that the reconstructed image of the input picture includes:
By the input picture, it is input in the generation model of production confrontation network, determines the pseudo- image of the input picture；
Third corresponding with region region is determined in the pseudo- image, and it is raw to replace the region using the third region
At reconstructed image.
5. method as claimed in claim 1 or 3, which is characterized in that the probability for determining reconstructed image includes:
Formula is lost according to the input picture, the reconstructed image and the context pre-saved, determines context loss probability；
Formula is lost according to the reconstructed image and the perception pre-saved, determines perception loss probability；
According to the context loss probability, the perception loss probability and the new probability formula pre-saved determine the reconstruct image
The probability of picture.
6. the method as described in claim 1, which is characterized in that it is described according to first reconstructed image, to the face figure
Include: as restore
First reconstructed image is fitted by Poisson image fusion technology, the facial image is restored.
7. a kind of image recovery device, which is characterized in that the device includes:
First determining module, for by facial image to be restored, being input in the Image Segmentation Model that training is completed in advance, really
The first area being blocked in the fixed facial image；The first area is marked in the facial image, determines that first is defeated
Enter image；
Second determining module fights network for by first input picture, being input to the production that training is completed in advance
It generates in model, determines the first reconstructed image of first input picture；By first input picture and first weight
Composition picture is input in the discrimination model of the production confrontation network, determines that first reconstructed image is complete facial image
The first probability；
Judgment module, for judging whether first probability is more than preset first probability threshold value；
Recovery module, for when the judging result of the judgment module be when, according to first reconstructed image, to the people
Face image is restored；When the judging result of the judgment module is no, using first reconstructed image as the first input
Image is input in the generation model, until restoring to the facial image.
8. device as claimed in claim 7, which is characterized in that first determining module is specifically used in the face figure
As in, the pixel value of pixel in the first area is updated to presetted pixel value；By the updated facial image of pixel value,
It is determined as the first input picture.
9. device as claimed in claim 7, which is characterized in that second determining module is also used to for each sample graph
Arbitrary region in the sample image is determined as the second area being blocked in the sample image by picture；In the sample
The second area is marked in image, determines the second input picture；By second input picture, it is input to production confrontation net
In the generation model of network, the second reconstructed image of second input picture is determined；By second input picture and described
Two reconstructed images are input in the discrimination model of the production confrontation network, determine that second reconstructed image is complete face
Second probability of image；When second probability is less than preset second probability threshold value, second reconstructed image is made
It for the second input picture, is input in the generation model, the generation model and the discrimination model is alternately trained.
10. the device as described in claim 7 or 9, which is characterized in that second determining module, being specifically used for will be described defeated
Enter image, is input in the generation model of production confrontation network, determines the pseudo- image of the input picture；In the pseudo- image
The Area generation reconstructed image is replaced using the third region in middle determination third corresponding with region region.
11. the device as described in claim 7 or 9, which is characterized in that second determining module is specifically used for according to
Input picture, the reconstructed image and the context loss formula pre-saved, determine context loss probability；According to the reconstruct image
Picture and the perception pre-saved loss formula, determine perception loss probability；According to the context loss probability, the perception loss
Probability and the new probability formula pre-saved, determine the probability of the reconstructed image.
12. device as claimed in claim 7, which is characterized in that the recovery module is specifically used for passing through Poisson image co-registration
Technology is fitted first reconstructed image, restores to the facial image.
13. a kind of electronic equipment characterized by comprising processor, communication interface, memory and communication bus, wherein place
Device, communication interface are managed, memory completes mutual communication by communication bus；
It is stored with computer program in the memory, when described program is executed by the processor, so that the processor
Perform claim requires the step of any one of 1~6 the method.
14. a kind of computer readable storage medium, which is characterized in that it is stored with the computer journey that can be executed by electronic equipment
Sequence, when described program is run on the electronic equipment, so that the electronic equipment perform claim requires any one of 1~6 institute
The step of stating method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810608192.2A CN108986041A (en) | 2018-06-13 | 2018-06-13 | A kind of image recovery method, device, electronic equipment and readable storage medium storing program for executing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810608192.2A CN108986041A (en) | 2018-06-13 | 2018-06-13 | A kind of image recovery method, device, electronic equipment and readable storage medium storing program for executing |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108986041A true CN108986041A (en) | 2018-12-11 |
Family
ID=64540325
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810608192.2A Pending CN108986041A (en) | 2018-06-13 | 2018-06-13 | A kind of image recovery method, device, electronic equipment and readable storage medium storing program for executing |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108986041A (en) |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109785258A (en) * | 2019-01-10 | 2019-05-21 | 华南理工大学 | A kind of facial image restorative procedure generating confrontation network based on more arbiters |
| CN109978805A (en) * | 2019-03-18 | 2019-07-05 | Oppo广东移动通信有限公司 | It takes pictures processing method, device, mobile terminal and storage medium |
| CN110189336A (en) * | 2019-05-30 | 2019-08-30 | 上海极链网络科技有限公司 | Image generating method, system, server and storage medium |
| CN110348331A (en) * | 2019-06-24 | 2019-10-18 | 深圳和而泰家居在线网络科技有限公司 | Face identification method and electronic equipment |
| CN110956097A (en) * | 2019-11-13 | 2020-04-03 | 北京影谱科技股份有限公司 | Method and module for extracting occluded human body and method and device for scene conversion |
| CN110991299A (en) * | 2019-11-27 | 2020-04-10 | 中新国际联合研究院 | Confrontation sample generation method aiming at face recognition system in physical domain |
| CN111340030A (en) * | 2020-02-14 | 2020-06-26 | 北京市商汤科技开发有限公司 | Image processing method and device, electronic equipment and computer readable storage medium |
| 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 |
| CN111353943A (en) * | 2018-12-20 | 2020-06-30 | 杭州海康威视数字技术股份有限公司 | Face image recovery method and device and readable storage medium |
| CN111753882A (en) * | 2020-06-01 | 2020-10-09 | Oppo广东移动通信有限公司 | Training method and device of image recognition network and electronic equipment |
| CN111861894A (en) * | 2019-04-25 | 2020-10-30 | 上海理工大学 | Image motion blur removing method based on generating type countermeasure network |
| CN112308104A (en) * | 2019-08-02 | 2021-02-02 | 杭州海康威视数字技术股份有限公司 | Abnormity identification method and device and computer storage medium |
| CN112614066A (en) * | 2020-12-23 | 2021-04-06 | 文思海辉智科科技有限公司 | Image restoration method and device and electronic equipment |
| CN112837228A (en) * | 2019-11-22 | 2021-05-25 | 北京金山云网络技术有限公司 | Image processing method and device and terminal equipment |
| CN113761997A (en) * | 2020-08-27 | 2021-12-07 | 北京沃东天骏信息技术有限公司 | Method and device for generating semi-occlusion face recognizer |
| CN114641980A (en) * | 2020-02-28 | 2022-06-17 | 联合技术旧金山公司 | Reconstruction of occluded views of a captured image using arbitrarily captured input |
| CN114862699A (en) * | 2022-04-14 | 2022-08-05 | 中国科学院自动化研究所 | Face repairing method, device and storage medium based on generation countermeasure network |
| CN112597867B (en) * | 2020-12-17 | 2024-04-26 | 佛山科学技术学院 | Face recognition method and system for wearing mask, computer equipment and storage medium |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101770649A (en) * | 2008-12-30 | 2010-07-07 | 中国科学院自动化研究所 | Automatic synthesis method for facial image |
| CN107590774A (en) * | 2017-09-18 | 2018-01-16 | 北京邮电大学 | A kind of car plate clarification method and device based on generation confrontation network |
| CN107729819A (en) * | 2017-09-22 | 2018-02-23 | 华中科技大学 | A kind of face mask method based on sparse full convolutional neural networks |
| CN107730458A (en) * | 2017-09-05 | 2018-02-23 | 北京飞搜科技有限公司 | A kind of fuzzy facial reconstruction method and system based on production confrontation network |
| CN107945118A (en) * | 2017-10-30 | 2018-04-20 | 南京邮电大学 | A kind of facial image restorative procedure based on production confrontation network |
| US20180124415A1 (en) * | 2016-05-06 | 2018-05-03 | Magic Pony Technology Limited | Encoder pre-analyser |
| CN108492281A (en) * | 2018-03-06 | 2018-09-04 | 陕西师范大学 | A method of fighting Bridge Crack image detection of obstacles and the removal of network based on production |
| CN109816593A (en) * | 2019-01-18 | 2019-05-28 | 大连海事大学 | A kind of super-resolution image reconstruction method of the generation confrontation network based on attention mechanism |
-
2018
- 2018-06-13 CN CN201810608192.2A patent/CN108986041A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101770649A (en) * | 2008-12-30 | 2010-07-07 | 中国科学院自动化研究所 | Automatic synthesis method for facial image |
| US20180124415A1 (en) * | 2016-05-06 | 2018-05-03 | Magic Pony Technology Limited | Encoder pre-analyser |
| CN107730458A (en) * | 2017-09-05 | 2018-02-23 | 北京飞搜科技有限公司 | A kind of fuzzy facial reconstruction method and system based on production confrontation network |
| CN107590774A (en) * | 2017-09-18 | 2018-01-16 | 北京邮电大学 | A kind of car plate clarification method and device based on generation confrontation network |
| CN107729819A (en) * | 2017-09-22 | 2018-02-23 | 华中科技大学 | A kind of face mask method based on sparse full convolutional neural networks |
| CN107945118A (en) * | 2017-10-30 | 2018-04-20 | 南京邮电大学 | A kind of facial image restorative procedure based on production confrontation network |
| CN108492281A (en) * | 2018-03-06 | 2018-09-04 | 陕西师范大学 | A method of fighting Bridge Crack image detection of obstacles and the removal of network based on production |
| CN109816593A (en) * | 2019-01-18 | 2019-05-28 | 大连海事大学 | A kind of super-resolution image reconstruction method of the generation confrontation network based on attention mechanism |
Non-Patent Citations (1)
| Title |
|---|
| RAYMOND A. YEH 等: ""Semantic Image Inpainting with Deep Generative Models"", 《ARXIV.ORG》 * |
Cited By (25)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111353943B (en) * | 2018-12-20 | 2023-12-26 | 杭州海康威视数字技术股份有限公司 | Face image recovery method and device and readable storage medium |
| 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 |
| CN111353943A (en) * | 2018-12-20 | 2020-06-30 | 杭州海康威视数字技术股份有限公司 | Face image recovery method and device and readable storage medium |
| CN109785258A (en) * | 2019-01-10 | 2019-05-21 | 华南理工大学 | A kind of facial image restorative procedure generating confrontation network based on more arbiters |
| CN109978805A (en) * | 2019-03-18 | 2019-07-05 | Oppo广东移动通信有限公司 | It takes pictures processing method, device, mobile terminal and storage medium |
| CN111861894A (en) * | 2019-04-25 | 2020-10-30 | 上海理工大学 | Image motion blur removing method based on generating type countermeasure network |
| CN111861894B (en) * | 2019-04-25 | 2023-06-20 | 上海理工大学 | Image motion blur removing method based on generation type countermeasure network |
| CN110189336A (en) * | 2019-05-30 | 2019-08-30 | 上海极链网络科技有限公司 | Image generating method, system, server and storage medium |
| CN110189336B (en) * | 2019-05-30 | 2020-05-01 | 上海极链网络科技有限公司 | Image generation method, system, server and storage medium |
| CN110348331A (en) * | 2019-06-24 | 2019-10-18 | 深圳和而泰家居在线网络科技有限公司 | Face identification method and electronic equipment |
| CN110348331B (en) * | 2019-06-24 | 2022-01-14 | 深圳数联天下智能科技有限公司 | Face recognition method and electronic equipment |
| CN112308104A (en) * | 2019-08-02 | 2021-02-02 | 杭州海康威视数字技术股份有限公司 | Abnormity identification method and device and computer storage medium |
| CN110956097A (en) * | 2019-11-13 | 2020-04-03 | 北京影谱科技股份有限公司 | Method and module for extracting occluded human body and method and device for scene conversion |
| CN110956097B (en) * | 2019-11-13 | 2023-07-21 | 北京影谱科技股份有限公司 | Method and module for extracting occlusion human body, and scene conversion method and device |
| CN112837228A (en) * | 2019-11-22 | 2021-05-25 | 北京金山云网络技术有限公司 | Image processing method and device and terminal equipment |
| CN110991299A (en) * | 2019-11-27 | 2020-04-10 | 中新国际联合研究院 | Confrontation sample generation method aiming at face recognition system in physical domain |
| CN110991299B (en) * | 2019-11-27 | 2023-03-14 | 中新国际联合研究院 | Confrontation sample generation method aiming at face recognition system in physical domain |
| CN111340030A (en) * | 2020-02-14 | 2020-06-26 | 北京市商汤科技开发有限公司 | Image processing method and device, electronic equipment and computer readable storage medium |
| CN114641980A (en) * | 2020-02-28 | 2022-06-17 | 联合技术旧金山公司 | Reconstruction of occluded views of a captured image using arbitrarily captured input |
| CN111753882A (en) * | 2020-06-01 | 2020-10-09 | Oppo广东移动通信有限公司 | Training method and device of image recognition network and electronic equipment |
| CN113761997A (en) * | 2020-08-27 | 2021-12-07 | 北京沃东天骏信息技术有限公司 | Method and device for generating semi-occlusion face recognizer |
| CN113761997B (en) * | 2020-08-27 | 2024-04-09 | 北京沃东天骏信息技术有限公司 | Method and device for generating semi-occlusion face recognition device |
| CN112597867B (en) * | 2020-12-17 | 2024-04-26 | 佛山科学技术学院 | Face recognition method and system for wearing mask, computer equipment and storage medium |
| CN112614066A (en) * | 2020-12-23 | 2021-04-06 | 文思海辉智科科技有限公司 | Image restoration method and device and electronic equipment |
| CN114862699A (en) * | 2022-04-14 | 2022-08-05 | 中国科学院自动化研究所 | Face repairing method, device and storage medium based on generation countermeasure network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108986041A (en) | A kind of image recovery method, device, electronic equipment and readable storage medium storing program for executing | |
| Liu et al. | Attribute-aware face aging with wavelet-based generative adversarial networks | |
| Sabir et al. | Recurrent convolutional strategies for face manipulation detection in videos | |
| Xiang et al. | Deep learning for image inpainting: A survey | |
| Zhang et al. | Cosaliency detection based on intrasaliency prior transfer and deep intersaliency mining | |
| Li et al. | Generative face completion | |
| CN111047548B (en) | Attitude transformation data processing method and device, computer equipment and storage medium | |
| CN106897658B (en) | Method and device for identifying human face living body | |
| Zhang et al. | Cross-view cross-scene multi-view crowd counting | |
| US20210158023A1 (en) | System and Method for Generating Image Landmarks | |
| CN109657583B (en) | Face key point detection method and device, computer equipment and storage medium | |
| Cai et al. | Semi-supervised natural face de-occlusion | |
| CN110866509B (en) | Action recognition method, device, computer storage medium and computer equipment | |
| CN106068514A (en) | For identifying the system and method for face in free media | |
| CN109754403A (en) | Tumour automatic division method and system in a kind of CT image | |
| CN106326857A (en) | Gender identification method and gender identification device based on face image | |
| CN111080670B (en) | Image extraction method, device, equipment and storage medium | |
| CN107886062B (en) | Image processing method, system and server | |
| CN111754396A (en) | Face image processing method and device, computer equipment and storage medium | |
| JP2024501986A (en) | 3D face reconstruction method, 3D face reconstruction apparatus, device, and storage medium | |
| CN111881731A (en) | Behavior recognition method, system, device and medium based on human skeleton | |
| CN109063678A (en) | The method, apparatus and storage medium of face image identification | |
| KR20230085931A (en) | Method and system for extracting color from face images | |
| CN112200236B (en) | Training method of face parameter identification model and face parameter identification method | |
| CN113705301A (en) | Image processing 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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20181211 |
|
| RJ01 | Rejection of invention patent application after publication |