CN108596833A - Super-resolution image reconstruction method, device, equipment and readable storage medium storing program for executing - Google Patents
Super-resolution image reconstruction method, device, equipment and readable storage medium storing program for executing Download PDFInfo
- Publication number
- CN108596833A CN108596833A CN201810388108.0A CN201810388108A CN108596833A CN 108596833 A CN108596833 A CN 108596833A CN 201810388108 A CN201810388108 A CN 201810388108A CN 108596833 A CN108596833 A CN 108596833A
- Authority
- CN
- China
- Prior art keywords
- resolution
- image
- super
- convolutional neural
- neural networks
- 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 50
- 238000003860 storage Methods 0.000 title claims abstract description 13
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 58
- 230000001815 facial effect Effects 0.000 claims abstract description 43
- 230000004913 activation Effects 0.000 claims abstract description 39
- 238000012549 training Methods 0.000 claims abstract description 30
- 238000003062 neural network model Methods 0.000 claims abstract description 17
- 238000012545 processing Methods 0.000 claims abstract description 17
- 230000006870 function Effects 0.000 claims description 15
- 230000001537 neural effect Effects 0.000 claims description 5
- 238000004590 computer program Methods 0.000 claims description 3
- 230000006866 deterioration Effects 0.000 abstract description 5
- 230000007812 deficiency Effects 0.000 abstract description 4
- 238000013528 artificial neural network Methods 0.000 description 11
- 230000008569 process Effects 0.000 description 9
- 210000002569 neuron Anatomy 0.000 description 6
- 238000010168 coupling process Methods 0.000 description 4
- 238000005859 coupling reaction Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 230000009471 action Effects 0.000 description 3
- 230000008901 benefit Effects 0.000 description 3
- 230000008878 coupling Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000013507 mapping Methods 0.000 description 3
- 238000005452 bending Methods 0.000 description 2
- 230000001133 acceleration Effects 0.000 description 1
- 238000012098 association analyses Methods 0.000 description 1
- 230000008827 biological function Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000003475 lamination Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000035479 physiological effects, processes and functions Effects 0.000 description 1
- 238000000513 principal component analysis Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4053—Scaling of whole images or parts thereof, e.g. expanding or contracting based on super-resolution, i.e. the output image resolution being higher than the sensor resolution
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
This application provides a kind of super-resolution image reconstruction method, device, equipment and readable storage medium storing program for executing, specifically super-resolution image reconstruction method is by obtaining preset high-definition picture, and is handled preset high-definition picture to obtain corresponding low-resolution image；Preset high-definition picture and low-resolution image and the convolutional neural networks for being written with activation primitive are subjected to neural network model training, obtain convolutional neural networks model；Using human face region image to be reconstructed as dictionary to be reconstructed, the convolutional neural networks model parameter obtained with the convolutional neural networks model for having input preset external trainer sample dictionary carries out the reconstruction processing of super-resolution facial image, solve the super-resolution algorithms currently based on facial image, under image deterioration serious situation, quality reconstruction is unsatisfactory, the technical issues of leading to the technical problem of image definition deficiency.
Description
Technical field
The invention relates to image processing techniques and computer vision field more particularly to a kind of super-resolution images
Reconstructing method, device, equipment and readable storage medium storing program for executing.
Background technology
Super-resolution image reconstruction refers to being handled low quality, low point of variability image (Low-resolution, LR),
The technology for recovering high-definition picture (High-resolution, HR), in military affairs, medicine, public safety, computer vision
Etc. all there is important application prospects.Past decades, image super-resolution have been widely studied, and are produced
Many process usefuls.These methods are substantially segmented into three classes:Method based on interpolation, method based on reconstruct and based on learning
The method of habit.
In recent years, it has obtained the extensive concern of researchers specifically for the super-resolution reconstruction algorithm of facial image and has ground
Study carefully.Facial image has extremely complex physiology, compared to it as a kind of special, multidimensional non-rigid pattern
His image format, challenge of the reconstruct with bigger based on the super-resolution image reconstruction of facial image relative to other images
Property.In face image super-resolution restructing algorithm, Baker et al. proposes " illusory face " algorithm for the first time, by introducing image
Gradient prior information be trained and obtain effective detailed information.Zhu Huasheng et al. combines sparse representation theory, it is proposed that one
The face image super-resolution restructing algorithm based on local restriction is planted to constrain image.It is equal super based on pixel space in tradition to open ground
The basis of resolution reconstruction algorithm, it is proposed that a kind of human face super-resolution image reconstruction algorithm in feature based space, it can
Obtain more excellent feature.An etc. proposes one kind and being based on the typical association analysis facial image super-resolution reconstruction algorithm of two dimension,
Have found the relevance of facial image.
Super-resolution algorithms based on facial image above, under image deterioration serious situation, quality reconstruction is simultaneously paid no attention to
Think, such as in certain outline portions, there are broken edges, lead to the technical problem of image definition deficiency.
Invention content
The embodiment of the present application provides a kind of super-resolution image reconstruction method, device, equipment and readable storage medium storing program for executing solution
The super-resolution algorithms determined currently based on facial image, under image deterioration serious situation, quality reconstruction is unsatisfactory, leads
The technical issues of causing the technical problem of image definition deficiency.
The embodiment of the present application provides a kind of super-resolution image reconstruction method, including：
Preset high-definition picture is obtained, and the preset high-definition picture is handled to obtain corresponding low resolution
Rate image；
By the preset high-definition picture and the low-resolution image and the convolutional Neural net for being written with activation primitive
Network carries out neural network model training, obtains convolutional neural networks model；
Using human face region image to be reconstructed as dictionary to be reconstructed, with the institute for having input preset external trainer sample dictionary
State the reconstruction processing that the convolutional neural networks model parameter that convolutional neural networks model obtains carries out super-resolution facial image.
Optionally, preset high-definition picture is obtained, and the preset high-definition picture is handled and is corresponded to
Low-resolution image specifically include：
The preset high-definition picture in external data base is obtained, and is contracted to the preset high-definition picture
Small processing obtains corresponding low-resolution image；
The preset high-definition picture and the low resolution are handled.
Optionally, by the preset high-definition picture and the low-resolution image and the convolution for being written with activation primitive
Neural network carries out neural network model training, obtains convolutional neural networks model and specifically includes：
The preset high-definition picture after being divided into row block is with the low-resolution image as the input of training library
Neural network model training is carried out by the convolutional neural networks of activation primitive activated, obtains including training parameter sample
This convolutional neural networks model.
Optionally, using human face region image to be reconstructed as dictionary to be reconstructed, and preset external trainer sample is had input
The convolutional neural networks model parameter that the convolutional neural networks model of dictionary obtains carries out the weight of super-resolution facial image
Structure processing specifically includes：
Using bicubic interpolation by the human face region image magnification to be reconstructed of input to and the preset high resolution graphics
As onesize, and as initial reconstructed image；
Dictionary to be reconstructed is built according to the initial reconstructed image；
Using the training parameter sample as input, convolutional Neural net is obtained by the convolutional neural networks model learning
Network model parameter；
Super-resolution face image is carried out using the dictionary to be reconstructed and the convolutional neural networks model parameter
Processing.
Optionally, the human face region image acquisition mode to be reconstructed is to calculate the coupling obtained song by energy function
Line determines.
Optionally, the activation primitive is Noisy Softplus activation primitives.
The embodiment of the present application provides a kind of device of super-resolution facial image reconstruct, including：
Acquiring unit for obtaining preset high-definition picture, and to the preset high-definition picture handle
To corresponding low-resolution image；
Neural metwork training unit, for by the preset high-definition picture and the low-resolution image be written with
The convolutional neural networks of activation primitive carry out neural network model training, obtain convolutional neural networks model；
Reconfiguration unit, for using human face region image to be reconstructed as dictionary to be reconstructed, being instructed with preset outside is had input
Practice the convolutional neural networks model parameter progress super-resolution face figure that the convolutional neural networks model of sample dictionary obtains
The reconstruction processing of picture.
The embodiment of the present application provides a kind of super-resolution facial image reconstructing arrangement, the equipment include processor and
Memory：
Said program code is transferred to the processor by the memory for storing program code；
Any institute that the processor is used to be referred to according to instruction execution the embodiment of the present application in said program code
The super-resolution facial image reconstructing method stated.
The embodiment of the present application provides a kind of readable storage medium storing program for executing, and the readable storage medium storing program for executing is used to store program code,
Said program code is used to execute any super-resolution facial image reconstructing method that the embodiment of the present application refers to.
It includes the computer program product instructed that the embodiment of the present application, which provides a kind of, when run on a computer,
So that the computer executes any super-resolution facial image reconstructing method that the embodiment of the present application refers to.
As can be seen from the above technical solutions, the embodiment of the present application has the following advantages：
The embodiment of the present application provides a kind of super-resolution image reconstruction method, by obtaining preset high-definition picture,
And preset high-definition picture is handled to obtain corresponding low-resolution image；By preset high-definition picture and low resolution
Rate image and the convolutional neural networks for being written with activation primitive carry out neural network model training, obtain convolutional neural networks mould
Type；Using human face region image to be reconstructed as dictionary to be reconstructed, with the convolution god for having input preset external trainer sample dictionary
The convolutional neural networks model parameter obtained through network model carries out the reconstruction processing of super-resolution facial image, solves at present
Super-resolution algorithms based on facial image, under image deterioration serious situation, quality reconstruction is unsatisfactory, causes image clear
The technical issues of technical problem that clear degree is insufficient.
Description of the drawings
Fig. 1 is a kind of flow signal of one embodiment of super-resolution facial image reconstructing method in the embodiment of the present application
Figure；
Fig. 2 is a kind of structural representation of one embodiment of super-resolution facial image reconstruct device in the embodiment of the present application
Figure；
Fig. 3 to Fig. 6 is the corresponding application examples schematic diagram of Fig. 1 embodiments.
Specific implementation mode
The embodiment of the present application provides a kind of super-resolution image reconstruction method, device, equipment and readable storage medium storing program for executing solution
The super-resolution algorithms determined currently based on facial image, under image deterioration serious situation, quality reconstruction is unsatisfactory, leads
The technical issues of causing the technical problem of image definition deficiency.
To enable the goal of the invention, feature, advantage of the embodiment of the present application more apparent and understandable, below in conjunction with
Attached drawing in the embodiment of the present application, technical solutions in the embodiments of the present application are clearly and completely described, it is clear that below
Described embodiment is only the embodiment of the present application a part of the embodiment, and not all embodiment.Implemented based on the application
Embodiment in example, all other implementation obtained by those of ordinary skill in the art without making creative efforts
Example belongs to the range of the embodiment of the present application protection.
Referring to Fig. 1, a kind of one embodiment of super-resolution facial image reconstructing method provided by the embodiments of the present application
Including：
101, preset high-definition picture is obtained, and preset high-definition picture is handled to obtain corresponding low resolution
Rate image；
In the present embodiment, when needing the reconstruction of human face super-resolution, the preset high score in external data base is obtained first
Resolution image, and diminution is carried out to preset high-definition picture and handles to obtain corresponding low-resolution image, by preset high-resolution
Rate image and low resolution processing.
Specifically, the high-definition picture in incoming external data library is reduced three times, obtains low-resolution image.
102, by preset high-definition picture and low-resolution image and be written with the convolutional neural networks of activation primitive into
Row neural network model is trained, and convolutional neural networks model is obtained；
Preset high-definition picture after being divided into row block passes through activation with low-resolution image as the input of training library
The convolutional neural networks that have activated of function carry out neural network model training, obtain include training parameter sample convolution
Neural network model.
Specifically, by external data base high-definition picture and low-resolution image be divided into 3 X, 3 fritter.It will tool
There are external data base LR and the HR image of structure self-similarity to there is the didactic activation model of biology as the input of training library
Convolutional neural networks.Utilize the didactic activation convolutional neural networks of biology.The mapping function for learning LR to HR images, is instructed
Practice parameter set={ W1, W2, W3, B1, B2, B3 }.
103, using human face region image to be reconstructed as dictionary to be reconstructed, and preset external trainer sample dictionary is had input
The obtained convolutional neural networks model parameter of convolutional neural networks model carry out the reconstruction processing of super-resolution facial image.
Using bicubic interpolation by the human face region image magnification to be reconstructed of input to and preset high-definition picture it is same
Sample size, and as initial reconstructed image；Dictionary to be reconstructed is built according to initial reconstructed image；Using training parameter sample as defeated
Enter, convolutional neural networks model parameter is obtained by convolutional neural networks model learning；Utilize dictionary to be reconstructed and convolutional Neural
Network model parameter carries out super-resolution face image processing.
Specifically, using bicubic interpolation by the LR image magnifications to be reconstructed of input to as HR image sizes, as
Initial reconstructed image X0.Initial reconstructed image X0 is built into dictionary to be reconstructed.Using external trainer sample dictionary as input, pass through
Learn to obtain convolutional neural networks model parameter with the didactic activation model convolutional neural networks of biology.Utilize word to be reconstructed
Allusion quotation and convolutional neural networks model parameter carry out reconstruction target image.
It should be noted that human face region image acquisition mode to be reconstructed above-mentioned is to calculate to obtain by energy function
Connecting curve determine.Activation primitive above-mentioned is Noisy Softplus activation primitives.
Fig. 3 is the Face datection figure of S-Shaped Algorithm, and Fig. 4 emulates for Noisey softplus activation primitives, and Fig. 5 is convolution
Neural network model, Fig. 6 are that the concrete application of Fig. 1 embodiments is described in detail, and application examples includes：
1, Face datection algorithm
Face datection is mainly the following method, one is judging whether it is skin of face color according to color,
So that it is determined that human face region, is primarily adapted for use in coloured image；One is carried out with parameterized template according to the shape of face contour
Detection；One is the methods with principal component analysis or neural network, first learn to face characteristic, then in image
Each piece is split；One is determine face contour with S-Shaped Algorithm.But it is to rely on the selection of initial position, here
A kind of improved S-Shaped Algorithm is used, initial position is not strict with.
The present invention has used a kind of parametrization S-Shaped Algorithm.Parametrization S-Shaped Algorithm is a parameter curve in internal force and outer
The minimum position of some energy function is moved under the action of power, so that it is determined that the method for contour of object.If r (s)=(x (s), y
(s)), s ∈ [0,1] are a parameter curve, then energy function is：
Wherein, EintIndicate that the internal energy determined by the internal force such as the bending force of curve itself and stretching force, Eext indicate
External energy determined by the external force determined by image.
α, β are parameter,Effect be ensure curve length stays constant under the action of stretching force.
By the deduction and calculating of formula, the estimation of parameter, finally finding out connecting curve is：
Wherein so is parametrization step-length,Effect be that curvature is small as possible under the action of allowing bending force.
2, the heuristic activation model of biology
2.1 Noisy Softplus activation primitives
Activation primitive, be not deactivate what, and refer to how " feature of the neuron of activation " by function handle
Feature, which retains and maps out, comes (keeping characteristics remove the redundancy for being in some data), and it is non-linear that this is that neural network can solve
Key to the issue.
Neural network uses ReLU as activation primitive under normal conditions.
ReLU activation primitives are substituted for a Noisy Softplus activation primitive with biological authenticity by the present invention,
Its expression formula is：
Constant k is scale factor, controls its curve shape, and σ is used for controlling its noise power, expression formula：
After introducing Noisy Softplus activation primitives in neural network, network is just provided with biological authenticity,
Certainly the due characteristic of a lot of other activation primitives is also included.
Biology inspires activation function, and Noisy Softplus adapt to the noise level of input current automatically, are to attempt for the first time
The transmitting response that unit will be activated to be mapped exactly to LIF neurons.Noisy Softplus are not only that activation functional bands are come more
Biological function, and also confirm can be showed in the ConvNet identification missions of acceleration it is good.With use Siegert mono-
The result of position is compared, and the spike version of Noisy Softplus surpasses the accuracy of S-shaped neuron.Since it is more accurately reflected
It penetrates, Noisy Softplus are better than Softplus.
The future work of SNN is included within limitation of the limitation function in effective range during training, this is equivalent to constraint
The emission maximum rate of LIF neurons.Therefore, process need not be scaled after training.For more accurate mapping, it should (digitally)
K is to avoid calibration for the derived proportions factor.In artificial neural network, swash noise as additional information to learn Softplus
It may be useful to live to reinforce classification
2.2 neural network model
Image super-resolution method (Super-resolution convolution based on convolutional neural networks
Neural network, SRCNN), using LR and HR image blocks pair pairs of in external data base, pass through convolutional neural networks mould
Type obtains corresponding prior information, to realize super-resolution rebuilding.Super-resolution convolutional neural networks are made of three-layer coil lamination,
Respectively feature extraction, Nonlinear Mapping and high-definition picture reconstruct.Three convolution nets in the deep learning network of SRCNN
Network formula is expressed as follows:
Y1=max (0, W1*X+B1) (7)
Y2=max (0, W2*Y1+B2) (8)
Y3=max (0, W3*Y2+B3) (9)
In above-mentioned formula, matrix X represents LR subgraphs, and Yi (i=1,2,3) indicates the output of each convolutional layer, Wi (i
=1,2,3) and Bi (i=1,2,3) respectively represents neuron convolution kernel and neuron bias vector, and * indicates convolution algorithm, convolution
Obtained characteristic pattern will be handled through ReLU activation primitives max (0, x) again.We are made that it corresponding improvement herein,
Refer to it is a kind of there is the didactic activation primitive noisey softplus of biology, it for entire neural network study and
Training process is more efficient.
In training process, which needs learning parameter Θ={ W1, W2, W3, B1, B2, B3 }, the parameter to pass through most
Smallization neural network exports the error loss between HR images and is trained estimation error.Given high-definition picture set Yi
Its corresponding low-resolution image set Xi uses its mean square error L (Θ) as driving loss function：
Formula (4) can be trained solution by stochastic gradient descent and back-propagation algorithm.
The embodiment of the present application is in order to which the dictionary learning process for solving traditional rarefaction representation algorithm structure is complicated and neural network
Activate the problem of models fitting effect difference, the reality of human face super-resolution, it is proposed that a kind of new based on biological heuristic activation
The face super-resolution method of model.Theory of algorithm proposed by the present invention is feasible.Meanwhile it taking with the didactic activation of biology
Function carries out the structure of neural network, so that it is had and is fitted perfect and adjustable benefit, so that being extracted in experimentation
Feature it is more efficient, recognition accuracy is also correspondingly improved.The heuristic activation model of biology is applied to super-resolution image,
The use of noisy softplus so that network can be self-introduced into sparse activity, this way is equivalent to unsupervised learning
Pre-training.Noisy softplus reduce the generation gap between unsupervised learning and supervised learning.Certainly, it also instructs faster
Practice speed and better feature learning rate.
Referring to Fig. 2, an a kind of implementation of the device of super-resolution facial image reconstruct provided by the embodiments of the present application
Example include：
Acquiring unit 201 is handled to obtain for obtaining preset high-definition picture, and to preset high-definition picture
Corresponding low-resolution image；
Neural metwork training unit 202, for by preset high-definition picture and low-resolution image and being written with activation
The convolutional neural networks of function carry out neural network model training, obtain convolutional neural networks model；
Reconfiguration unit 203 and has input preset outside for using human face region image to be reconstructed as dictionary to be reconstructed
The convolutional neural networks model parameter that the convolutional neural networks model of training sample dictionary obtains carries out super-resolution facial image
Reconstruction processing.
The embodiment of the present application also provides another super-resolution facial image reconstructing arrangement, the equipment includes processor
And memory：
Said program code is transferred to the processor by the memory for storing program code；
The processor is used for super according to any one of instruction execution claim 1-6 in said program code
Resolution ratio facial image reconstructing method.
The embodiment of the present application also provides a kind of readable storage medium storing program for executing, and for storing program code, the program code is for holding
Any one embodiment in a kind of super-resolution facial image reconstructing method described in row foregoing individual embodiments.
It includes the computer program product instructed that the embodiment of the present application, which also provides a kind of, when run on a computer,
So that computer executes any one reality in a kind of super-resolution facial image reconstructing method described in foregoing individual embodiments
Apply mode.
It is apparent to those skilled in the art that for convenience and simplicity of description, the system of foregoing description,
The specific work process of device and unit, can refer to corresponding processes in the foregoing method embodiment, and details are not described herein.
Term " first ", " second ", " third ", " the 4th " in the description of the present application and above-mentioned attached drawing etc. are (if deposited
) it is for distinguishing similar object, without being used to describe specific sequence or precedence.It should be appreciated that use in this way
Data can be interchanged in the appropriate case, so that embodiments herein described herein for example can be in addition to illustrating herein
Or the sequence other than those of description is implemented.In addition, term " comprising " and " having " and their any deformation, it is intended that
Cover it is non-exclusive include, for example, containing the process of series of steps or unit, method, system, product or equipment need not limit
In those of clearly listing step or unit, but may include not listing clearly or for these processes, method, production
The intrinsic other steps of product or equipment or unit.
It should be appreciated that in this application, " at least one (item) " refers to one or more, and " multiple " refer to two or two
More than a."and/or", the incidence relation for describing affiliated partner indicate may exist three kinds of relationships, for example, " A and/or B "
It can indicate：A is only existed, B is only existed and exists simultaneously tri- kinds of situations of A and B, wherein A, B can be odd number or plural number.Word
It is a kind of relationship of "or" that symbol "/", which typicallys represent forward-backward correlation object,.At least one of " following (a) " or its similar expression refers to
Arbitrary combination in these, including individual event (a) or the arbitrary combination of complex item (a).At least one of for example, in a, b or c
(a) can indicate：A, b, c, " a and b ", " a and c ", " b and c ", or " a and b and c ", wherein a, b, c can be single, also may be used
To be multiple.
In several embodiments provided herein, it should be understood that disclosed system, device and method can be with
It realizes by another way.For example, the apparatus embodiments described above are merely exemplary, for example, the unit
It divides, only a kind of division of logic function, formula that in actual implementation, there may be another division manner, such as multiple units or component
It can be combined or can be integrated into another system, or some features can be ignored or not executed.Another point, it is shown or
The mutual coupling, direct-coupling or communication connection discussed can be the indirect coupling by some interfaces, device or unit
It closes or communicates to connect, can be electrical, machinery or other forms.
The unit illustrated as separating component may or may not be physically separated, aobvious as unit
The component shown may or may not be physical unit, you can be located at a place, or may be distributed over multiple
In network element.Some or all of unit therein can be selected according to the actual needs to realize the mesh of this embodiment scheme
's.
In addition, each functional unit in each embodiment of the application can be integrated in a processing unit, it can also
It is that each unit physically exists alone, it can also be during two or more units be integrated in one unit.Above-mentioned integrated list
The form that hardware had both may be used in member is realized, can also be realized in the form of SFU software functional unit.
If the integrated unit is realized in the form of SFU software functional unit and sells or use as independent product
When, it can be stored in a computer read/write memory medium.Based on this understanding, the technical solution of the application is substantially
The all or part of the part that contributes to existing technology or the technical solution can be in the form of software products in other words
It embodies, which is stored in a storage medium, including some instructions are used so that a computer
Equipment (can be personal computer, server or the network equipment etc.) executes the complete of each embodiment the method for the application
Portion or part steps.And storage medium above-mentioned includes：USB flash disk, mobile hard disk, read-only memory (full name in English：Read-Only
Memory, english abbreviation：ROM), random access memory (full name in English：Random Access Memory, english abbreviation：
RAM), the various media that can store program code such as magnetic disc or CD.
The above, above example are only to illustrate the technical solution of the application, rather than its limitations；Although with reference to before
Embodiment is stated the application is described in detail, it will be understood by those of ordinary skill in the art that：It still can be to preceding
The technical solution recorded in each embodiment is stated to modify or equivalent replacement of some of the technical features；And these
Modification or replacement, the spirit and scope of each embodiment technical solution of the application that it does not separate the essence of the corresponding technical solution.
Claims (10)
1. a kind of super-resolution facial image reconstructing method, which is characterized in that including：
Preset high-definition picture is obtained, and the preset high-definition picture is handled to obtain corresponding low resolution figure
Picture；
By the preset high-definition picture and the low-resolution image and be written with the convolutional neural networks of activation primitive into
Row neural network model is trained, and convolutional neural networks model is obtained；
Using human face region image to be reconstructed as dictionary to be reconstructed, with the volume for having input preset external trainer sample dictionary
The convolutional neural networks model parameter that product neural network model obtains carries out the reconstruction processing of super-resolution facial image.
2. super-resolution facial image reconstructing method according to claim 1, which is characterized in that obtain preset high-resolution
Image, and the preset high-definition picture is handled to obtain corresponding low-resolution image and is specifically included：
The preset high-definition picture in external data base is obtained, and the preset high-definition picture is carried out at diminution
Reason obtains corresponding low-resolution image；
The preset high-definition picture and the low resolution are handled.
3. super-resolution facial image reconstructing method according to claim 1, which is characterized in that by the preset high-resolution
Rate image carries out neural network model training with the low-resolution image with the convolutional neural networks for being written with activation primitive, obtains
It is specifically included to convolutional neural networks model：
The preset high-definition picture after being divided into row block passes through with the low-resolution image as the input of training library
The convolutional neural networks of activation primitive activated carry out neural network model training, obtain including training parameter sample
Convolutional neural networks model.
4. super-resolution facial image reconstructing method according to claim 1, which is characterized in that by face area to be reconstructed
Area image is obtained as dictionary to be reconstructed with the convolutional neural networks model for having input preset external trainer sample dictionary
The reconstruction processing that convolutional neural networks model parameter carries out super-resolution facial image specifically includes：
Using bicubic interpolation by the human face region image magnification to be reconstructed of input to and the preset high-definition picture it is same
Sample size, and as initial reconstructed image；
Dictionary to be reconstructed is built according to the initial reconstructed image；
Using the training parameter sample as input, convolutional neural networks mould is obtained by the convolutional neural networks model learning
Shape parameter；
Super-resolution face image processing is carried out using the dictionary to be reconstructed and the convolutional neural networks model parameter.
5. super-resolution facial image reconstructing method according to claim 4, which is characterized in that the face to be reconstructed
Area image acquisition modes are to calculate the connecting curve obtained by energy function to determine.
6. super-resolution facial image reconstructing method as claimed in any of claims 1 to 5, which is characterized in that institute
It is Noisy Softplus activation primitives to state activation primitive.
7. a kind of device of super-resolution facial image reconstruct, which is characterized in that including：
Acquiring unit is handled to obtain pair for obtaining preset high-definition picture, and to the preset high-definition picture
The low-resolution image answered；
Neural metwork training unit, for by the preset high-definition picture and the low-resolution image and being written with activation
The convolutional neural networks of function carry out neural network model training, obtain convolutional neural networks model；
Reconfiguration unit and has input preset external trainer sample for using human face region image to be reconstructed as dictionary to be reconstructed
The convolutional neural networks model parameter that the convolutional neural networks model of this dictionary obtains carries out super-resolution facial image
Reconstruction processing.
8. a kind of super-resolution facial image reconstructing arrangement, which is characterized in that the equipment includes processor and memory：
Said program code is transferred to the processor by the memory for storing program code；
The processor is used for the super-resolution according to any one of instruction execution claim 1-6 in said program code
Rate facial image reconstructing method.
9. a kind of readable storage medium storing program for executing, which is characterized in that the readable storage medium storing program for executing is for storing program code, described program generation
Code requires the super-resolution facial image reconstructing method described in any one of 1-6 for perform claim.
10. a kind of includes the computer program product of instruction, which is characterized in that when run on a computer so that described
Computer perform claim requires the super-resolution facial image reconstructing method described in any one of 1-6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810388108.0A CN108596833A (en) | 2018-04-26 | 2018-04-26 | Super-resolution image reconstruction method, device, equipment and readable storage medium storing program for executing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810388108.0A CN108596833A (en) | 2018-04-26 | 2018-04-26 | Super-resolution image reconstruction method, device, equipment and readable storage medium storing program for executing |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108596833A true CN108596833A (en) | 2018-09-28 |
Family
ID=63609780
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810388108.0A Pending CN108596833A (en) | 2018-04-26 | 2018-04-26 | Super-resolution image reconstruction method, device, equipment and readable storage medium storing program for executing |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108596833A (en) |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109360154A (en) * | 2018-10-29 | 2019-02-19 | 厦门美图之家科技有限公司 | A kind of super-resolution method of convolutional neural networks generation method and image |
| CN109559360A (en) * | 2018-09-29 | 2019-04-02 | 深圳先进技术研究院 | A kind of PET image reconstruction method, device and calculate equipment |
| CN109727195A (en) * | 2018-12-25 | 2019-05-07 | 成都元点智库科技有限公司 | A kind of image super-resolution reconstructing method |
| CN109872377A (en) * | 2019-02-28 | 2019-06-11 | 上海交通大学 | Brain tissue fast imaging and image rebuilding method for magnetic resonance navigation |
| CN109903229A (en) * | 2019-03-04 | 2019-06-18 | 科新(杭州)能源环境科技有限公司 | A kind of μ-CT image reconstructing method based on convolutional neural networks |
| CN110288530A (en) * | 2019-06-28 | 2019-09-27 | 北京金山云网络技术有限公司 | A kind of pair of image carries out the processing method and processing device of super-resolution rebuilding |
| CN110310229A (en) * | 2019-06-28 | 2019-10-08 | Oppo广东移动通信有限公司 | Image processing method, image processing apparatus, terminal device and readable storage medium storing program for executing |
| CN111223161A (en) * | 2020-01-02 | 2020-06-02 | 京东数字科技控股有限公司 | Image reconstruction method and device and storage medium |
| CN111260566A (en) * | 2020-01-09 | 2020-06-09 | 复旦大学 | Image restoration method based on sparse prior |
| CN111383172A (en) * | 2018-12-29 | 2020-07-07 | Tcl集团股份有限公司 | Training method and device of neural network model and intelligent terminal |
| CN111489290A (en) * | 2019-04-02 | 2020-08-04 | 同观科技（深圳）有限公司 | Face image super-resolution reconstruction method and device and terminal equipment |
| CN111524072A (en) * | 2020-04-26 | 2020-08-11 | 重庆紫光华山智安科技有限公司 | Super-resolution reconstruction network training method and device, electronic equipment and storage medium |
| CN111681165A (en) * | 2020-06-02 | 2020-09-18 | 上海闻泰信息技术有限公司 | Image processing method, image processing device, computer equipment and computer readable storage medium |
| CN111932462A (en) * | 2020-08-18 | 2020-11-13 | Oppo（重庆）智能科技有限公司 | Training method and device for image degradation model, electronic equipment and storage medium |
| CN111951203A (en) * | 2020-07-01 | 2020-11-17 | 北京大学深圳研究生院 | Viewpoint synthesis method, apparatus, device and computer readable storage medium |
| CN113191955A (en) * | 2021-06-17 | 2021-07-30 | 江苏奥易克斯汽车电子科技股份有限公司 | Method and device for reconstructing image super-resolution |
| CN113222814A (en) * | 2021-04-22 | 2021-08-06 | 平安国际智慧城市科技股份有限公司 | Image resolution processing method, device, equipment and storage medium |
| CN114238904A (en) * | 2021-12-08 | 2022-03-25 | 马上消费金融股份有限公司 | Identity recognition method, and training method and device of two-channel hyper-resolution model |
| CN116485713A (en) * | 2023-03-01 | 2023-07-25 | 北京邮电大学 | Counting method and device for images, electronic equipment and medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106204449A (en) * | 2016-07-06 | 2016-12-07 | 安徽工业大学 | A kind of single image super resolution ratio reconstruction method based on symmetrical degree of depth network |
| CN106600538A (en) * | 2016-12-15 | 2017-04-26 | 武汉工程大学 | Human face super-resolution algorithm based on regional depth convolution neural network |
| CN107292896A (en) * | 2017-08-15 | 2017-10-24 | 电子科技大学 | Contour extraction method based on Snake models |
-
2018
- 2018-04-26 CN CN201810388108.0A patent/CN108596833A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106204449A (en) * | 2016-07-06 | 2016-12-07 | 安徽工业大学 | A kind of single image super resolution ratio reconstruction method based on symmetrical degree of depth network |
| CN106600538A (en) * | 2016-12-15 | 2017-04-26 | 武汉工程大学 | Human face super-resolution algorithm based on regional depth convolution neural network |
| CN107292896A (en) * | 2017-08-15 | 2017-10-24 | 电子科技大学 | Contour extraction method based on Snake models |
Non-Patent Citations (2)
| Title |
|---|
| QIAN LIU，STEVE FURBER: "Noisy Softplus: A Biology Inspired Activation Function", 《INTERNATIONAL CONFERENCE ON NEURAL INFORMATION PROCESSING》 * |
| 向文 等: "结合结构自相似性和卷积网络的单幅图像超分辨率", 《计算机应用》 * |
Cited By (27)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109559360B (en) * | 2018-09-29 | 2020-08-28 | 深圳先进技术研究院 | PET image reconstruction method and device and computing equipment |
| CN109559360A (en) * | 2018-09-29 | 2019-04-02 | 深圳先进技术研究院 | A kind of PET image reconstruction method, device and calculate equipment |
| CN109360154A (en) * | 2018-10-29 | 2019-02-19 | 厦门美图之家科技有限公司 | A kind of super-resolution method of convolutional neural networks generation method and image |
| CN109360154B (en) * | 2018-10-29 | 2022-09-20 | 厦门美图之家科技有限公司 | Convolutional neural network generation method and super-resolution method of image |
| CN109727195A (en) * | 2018-12-25 | 2019-05-07 | 成都元点智库科技有限公司 | A kind of image super-resolution reconstructing method |
| CN111383172A (en) * | 2018-12-29 | 2020-07-07 | Tcl集团股份有限公司 | Training method and device of neural network model and intelligent terminal |
| CN109872377A (en) * | 2019-02-28 | 2019-06-11 | 上海交通大学 | Brain tissue fast imaging and image rebuilding method for magnetic resonance navigation |
| CN109872377B (en) * | 2019-02-28 | 2023-01-20 | 上海交通大学 | Brain tissue rapid imaging and image reconstruction method for magnetic resonance navigation |
| CN109903229A (en) * | 2019-03-04 | 2019-06-18 | 科新(杭州)能源环境科技有限公司 | A kind of μ-CT image reconstructing method based on convolutional neural networks |
| CN111489290B (en) * | 2019-04-02 | 2023-05-16 | 长信智控网络科技有限公司 | Face image super-resolution reconstruction method and device and terminal equipment |
| CN111489290A (en) * | 2019-04-02 | 2020-08-04 | 同观科技（深圳）有限公司 | Face image super-resolution reconstruction method and device and terminal equipment |
| CN110310229A (en) * | 2019-06-28 | 2019-10-08 | Oppo广东移动通信有限公司 | Image processing method, image processing apparatus, terminal device and readable storage medium storing program for executing |
| CN110288530A (en) * | 2019-06-28 | 2019-09-27 | 北京金山云网络技术有限公司 | A kind of pair of image carries out the processing method and processing device of super-resolution rebuilding |
| CN111223161B (en) * | 2020-01-02 | 2024-04-12 | 京东科技控股股份有限公司 | Image reconstruction method, device and storage medium |
| CN111223161A (en) * | 2020-01-02 | 2020-06-02 | 京东数字科技控股有限公司 | Image reconstruction method and device and storage medium |
| CN111260566A (en) * | 2020-01-09 | 2020-06-09 | 复旦大学 | Image restoration method based on sparse prior |
| CN111260566B (en) * | 2020-01-09 | 2022-08-09 | 复旦大学 | Image restoration method based on sparse prior |
| CN111524072A (en) * | 2020-04-26 | 2020-08-11 | 重庆紫光华山智安科技有限公司 | Super-resolution reconstruction network training method and device, electronic equipment and storage medium |
| CN111681165A (en) * | 2020-06-02 | 2020-09-18 | 上海闻泰信息技术有限公司 | Image processing method, image processing device, computer equipment and computer readable storage medium |
| CN111951203A (en) * | 2020-07-01 | 2020-11-17 | 北京大学深圳研究生院 | Viewpoint synthesis method, apparatus, device and computer readable storage medium |
| CN111932462B (en) * | 2020-08-18 | 2023-01-03 | Oppo（重庆）智能科技有限公司 | Training method and device for image degradation model, electronic equipment and storage medium |
| CN111932462A (en) * | 2020-08-18 | 2020-11-13 | Oppo（重庆）智能科技有限公司 | Training method and device for image degradation model, electronic equipment and storage medium |
| CN113222814A (en) * | 2021-04-22 | 2021-08-06 | 平安国际智慧城市科技股份有限公司 | Image resolution processing method, device, equipment and storage medium |
| CN113222814B (en) * | 2021-04-22 | 2023-08-04 | 深圳赛安特技术服务有限公司 | Resolution processing method, device, equipment and storage medium for image |
| CN113191955A (en) * | 2021-06-17 | 2021-07-30 | 江苏奥易克斯汽车电子科技股份有限公司 | Method and device for reconstructing image super-resolution |
| CN114238904A (en) * | 2021-12-08 | 2022-03-25 | 马上消费金融股份有限公司 | Identity recognition method, and training method and device of two-channel hyper-resolution model |
| CN116485713A (en) * | 2023-03-01 | 2023-07-25 | 北京邮电大学 | Counting method and device for images, electronic equipment and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108596833A (en) | Super-resolution image reconstruction method, device, equipment and readable storage medium storing program for executing | |
| CN109685819B (en) | Three-dimensional medical image segmentation method based on feature enhancement | |
| CN109117864B (en) | Coronary heart disease risk prediction method, model and system based on heterogeneous feature fusion | |
| CN107292256B (en) | Auxiliary task-based deep convolution wavelet neural network expression recognition method | |
| CN111429460B (en) | Image segmentation method, image segmentation model training method, device and storage medium | |
| CN110599500B (en) | Tumor region segmentation method and system of liver CT image based on cascaded full convolution network | |
| CN110276745B (en) | Pathological image detection algorithm based on generation countermeasure network | |
| CN110378381A (en) | Object detecting method, device and computer storage medium | |
| CN108615010A (en) | Facial expression recognizing method based on the fusion of parallel convolutional neural networks characteristic pattern | |
| CN108108677A (en) | One kind is based on improved CNN facial expression recognizing methods | |
| CN112288011B (en) | Image matching method based on self-attention deep neural network | |
| Mallouh et al. | Utilizing CNNs and transfer learning of pre-trained models for age range classification from unconstrained face images | |
| CN111696101A (en) | Light-weight solanaceae disease identification method based on SE-Inception | |
| CN108062543A (en) | A kind of face recognition method and device | |
| JP2017157138A (en) | Image recognition device, image recognition method and program | |
| KR20210028185A (en) | Human posture analysis system and method | |
| CN109492668B (en) | MRI (magnetic resonance imaging) different-phase multimode image characterization method based on multi-channel convolutional neural network | |
| CN110309835B (en) | Image local feature extraction method and device | |
| CN113989890A (en) | Face expression recognition method based on multi-channel fusion and lightweight neural network | |
| CN113344864A (en) | Ultrasonic thyroid nodule benign and malignant prediction method based on deep learning | |
| CN112132145B (en) | Image classification method and system based on model extended convolutional neural network | |
| CN113780249B (en) | Expression recognition model processing method, device, equipment, medium and program product | |
| CN115578589B (en) | Unsupervised echocardiography section identification method | |
| CN109948662B (en) | Face image depth clustering method based on K-means and MMD | |
| CN114782979A (en) | Training method and device for pedestrian re-recognition model, storage medium and terminal |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180928 |