CN107087201B - Image processing method and device - Google Patents
Image processing method and device Download PDFInfo
- Publication number
- CN107087201B CN107087201B CN201710326762.4A CN201710326762A CN107087201B CN 107087201 B CN107087201 B CN 107087201B CN 201710326762 A CN201710326762 A CN 201710326762A CN 107087201 B CN107087201 B CN 107087201B
- Authority
- CN
- China
- Prior art keywords
- image
- neural network
- resolution
- wavelet
- transmitting terminal
- 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.)
- Active
Links
- 238000003672 processing method Methods 0.000 title claims abstract description 45
- 238000013528 artificial neural network Methods 0.000 claims abstract description 221
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 67
- 238000000034 method Methods 0.000 claims abstract description 60
- 230000006835 compression Effects 0.000 claims abstract description 43
- 238000007906 compression Methods 0.000 claims abstract description 43
- 238000012549 training Methods 0.000 claims description 48
- 230000005540 biological transmission Effects 0.000 claims description 27
- 238000005457 optimization Methods 0.000 claims description 15
- 238000012545 processing Methods 0.000 claims description 14
- 210000005036 nerve Anatomy 0.000 claims description 5
- 238000005516 engineering process Methods 0.000 abstract description 10
- 230000008569 process Effects 0.000 description 7
- 230000009466 transformation Effects 0.000 description 7
- 238000010586 diagram Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 6
- 210000004218 nerve net Anatomy 0.000 description 5
- 238000013527 convolutional neural network Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 238000001914 filtration Methods 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000032798 delamination Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000002955 isolation Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 230000008520 organization Effects 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
- H04N19/102—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or selection affected or controlled by the adaptive coding
- H04N19/132—Sampling, masking or truncation of coding units, e.g. adaptive resampling, frame skipping, frame interpolation or high-frequency transform coefficient masking
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/60—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using transform coding
- H04N19/63—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using transform coding using sub-band based transform, e.g. wavelets
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
- H04N21/20—Servers specifically adapted for the distribution of content, e.g. VOD servers; Operations thereof
- H04N21/23—Processing of content or additional data; Elementary server operations; Server middleware
- H04N21/234—Processing of video elementary streams, e.g. splicing of video streams or manipulating encoded video stream scene graphs
- H04N21/2343—Processing of video elementary streams, e.g. splicing of video streams or manipulating encoded video stream scene graphs involving reformatting operations of video signals for distribution or compliance with end-user requests or end-user device requirements
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
- H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
- H04N21/43—Processing of content or additional data, e.g. demultiplexing additional data from a digital video stream; Elementary client operations, e.g. monitoring of home network or synchronising decoder's clock; Client middleware
- H04N21/44—Processing of video elementary streams, e.g. splicing a video clip retrieved from local storage with an incoming video stream or rendering scenes according to encoded video stream scene graphs
- H04N21/4402—Processing of video elementary streams, e.g. splicing a video clip retrieved from local storage with an incoming video stream or rendering scenes according to encoded video stream scene graphs involving reformatting operations of video signals for household redistribution, storage or real-time display
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Image Processing (AREA)
- Compression Of Band Width Or Redundancy In Fax (AREA)
Abstract
The invention discloses a kind of image processing method and devices.Wherein, this method comprises: receiving the data that transmitting terminal is sent, and data determine wavelet coefficient based on the received, wherein include the wavelet coefficient obtained after wavelet decomposition to original image in the data that transmitting terminal is sent；Low-resolution image is generated according to wavelet coefficient；Super-resolution rebuilding is carried out to low-resolution image according to neural network, obtains high-definition picture.The present invention solves the technical issues of method for compressing image in the related technology can not meet higher compression efficiency and higher image restoring quality simultaneously.
Description
Technical field
The present invention relates to field of image processings, in particular to a kind of image processing method and device.
Background technique
In image transmitting, the limitation of bandwidth is usually needed so that the size of image and the transmission time of image cannot be taken into account
Image is compressed.Traditional picture compression method is confined to remove redundancy, algorithm to image data from the angle of mathematics
Design framework is substantially stationary, and compression efficiency is difficult to have greatly improved again, and it is solid to generate picture quality without learning functionality for system
It is fixed, when bandwidth limitation is serious, image quality is influenced huge.In the prior art, can by the way of wavelet transformation come pair
Image is compressed, and carries out super-resolution rebuilding to the low-resolution image of compression after compression of images.Wavelet transformation is close
A kind of tool of mathematical analysis being widely used over year, it overcomes defect of the short time discrete Fourier transform in single resolution ratio, In
Time domain and frequency domain have the ability of characterization signal detail information, and the window size of time and frequency is dynamically adapted, and adapt to model
It encloses extensively.In recent years, the compression of images based on wavelet transformation and transmission algorithm are widely used in Image Compression.In
5/3 lossless wavelet transformation and 9/7 wavelet transformation damaged has been used to realize picture pressure in the compression algorithm of JPEG 2000
Contracting, the image compression algorithm based on 5/3 small echo is using relatively broad.
Higher compression efficiency and higher image can not be met simultaneously also for method for compressing image in the related technology
The technical issues of proper mass, currently no effective solution has been proposed.
Summary of the invention
The embodiment of the invention provides a kind of image processing method and devices, at least to solve image pressure in the related technology
Contracting method can not meet the technical issues of higher compression efficiency and higher image restoring quality simultaneously.
According to an aspect of an embodiment of the present invention, a kind of image processing method is provided, is sent this method comprises: receiving
The data sent are held, and data determine wavelet coefficient based on the received, wherein include to original graph in the data that transmitting terminal is sent
As the wavelet coefficient obtained after wavelet decomposition；Low-resolution image is generated according to wavelet coefficient；According to neural network to low
Image in different resolution carries out super-resolution rebuilding, obtains high-definition picture.
Further, super-resolution rebuilding is carried out to low-resolution image according to neural network and comprises determining that current mind
Through network；Receive undated parameter, wherein undated parameter be transmitting terminal send according to original image to current neural network into
The parameter for the neural network that row optimization obtains；Current neural network is updated according to undated parameter, the mind updated
Through network；Super-resolution rebuilding is carried out to low-resolution image according to the neural network of update.
Further, this method further include: wavelet decomposition is carried out to multiple sample images, obtains corresponding multiple low resolutions
Rate sample image；According to multiple sample images and corresponding multiple low resolution sample image training neural networks；After training
Neural network be stored in transmitting terminal and receiving end as initial neural network, wherein according to transmitting terminal for the first time send
Data carry out super-resolution rebuilding when determine current neural network be initial neural network.
Further, carrying out super-resolution rebuilding to low-resolution image according to neural network includes: according to low resolution
The classification of the content selection neural network of image；Super-resolution reconstruction is carried out to low-resolution image according to the neural network of selection
It builds.
According to another aspect of an embodiment of the present invention, a kind of image processing method is additionally provided, this method comprises: transmitting terminal
Wavelet decomposition is carried out to original image, obtains wavelet coefficient；Transmitting terminal sends wavelet coefficient to receiving end；Receiving end is according to small echo
Coefficient generates low-resolution image；Receiving end carries out super-resolution rebuilding to low-resolution image according to neural network, obtains height
Image in different resolution.
Further, wavelet decomposition is carried out to original image in transmitting terminal, after obtaining wavelet coefficient, this method is also wrapped
Include: transmitting terminal generates low-resolution image according to wavelet coefficient；Transmitting terminal is according to original image and low-resolution image to transmission
Current neural network is held to optimize；After transmitting terminal is to optimization of the neural network Jing Guo preset times, transmitting terminal is to connecing
Receiving end sends undated parameter, wherein undated parameter is the parameter of the neural network in transmitting terminal after preset times optimization；
Receiving end is updated to be updated according to neural network of the undated parameter to receiving end after receiving undated parameter
Neural network, and oversubscription is carried out to the subsequently received corresponding low-resolution image of wavelet coefficient based on the neural network of update
Resolution is rebuild.
Further, wavelet decomposition is carried out to original image in transmitting terminal, before obtaining wavelet coefficient, this method is also wrapped
Include: transmitting terminal or receiving end carry out wavelet decomposition to multiple sample images, obtain corresponding multiple low resolution sample images；Hair
Sending end or receiving end are according to multiple sample images and corresponding multiple low resolution sample image training neural networks；After training
Neural network be stored in transmitting terminal and receiving end as initial neural network.
Further, transmitting terminal or receiving end are instructed according to multiple sample images and corresponding multiple low resolution sample images
Practicing neural network includes: that transmitting terminal or receiving end carry out low resolution sample image according to the content of low resolution sample image
Classification；The low resolution sample image of each classification is trained using corresponding neural network, obtains the mind of multiple classifications
Through network.
Further, transmitting terminal to receiving end send wavelet coefficient include: transmitting terminal to wavelet coefficient according to default compression
Algorithm is encoded, and code stream is obtained；Obtained code stream is sent to receiving end by transmitting terminal, wherein receiving end is receiving code stream
It is decoded later according to default compression algorithm, obtains wavelet coefficient.
According to another aspect of an embodiment of the present invention, a kind of image processing apparatus is additionally provided, which includes: to receive list
Member, for receiving the data of transmitting terminal transmission, and data determine wavelet coefficient based on the received, wherein the number that transmitting terminal is sent
It include the wavelet coefficient obtained after wavelet decomposition to original image in；Generation unit, for being generated according to wavelet coefficient
Low-resolution image；Reconstruction unit obtains high score for carrying out super-resolution rebuilding to low-resolution image according to neural network
Resolution image.
Further, reconstruction unit comprises determining that module, for determining current neural network；First receiving module is used
In reception undated parameter, wherein undated parameter is the excellent to current neural network progress according to original image of transmitting terminal transmission
Change the parameter of obtained neural network；Update module is obtained for being updated according to undated parameter to current neural network
The neural network of update；First rebuilds module, for carrying out super-resolution to low-resolution image according to the neural network of update
It rebuilds.
Further, reconstruction unit includes: selecting module, for the content selection neural network according to low-resolution image
Classification；Second rebuilds module, for carrying out super-resolution rebuilding to low-resolution image according to the neural network of selection.
According to another aspect of an embodiment of the present invention, a kind of image processing apparatus is additionally provided, which includes: to decompose list
Member obtains wavelet coefficient for carrying out wavelet decomposition to original image by transmitting terminal；Transmission unit, for passing through transmitting terminal
Wavelet coefficient is sent to receiving end；Generation unit, for generating low-resolution image according to wavelet coefficient by receiving end；It rebuilds
Unit carries out super-resolution rebuilding to low-resolution image according to neural network for receiving end, obtains high-definition picture.
In embodiments of the present invention, the data sent by receiving transmitting terminal, and data determine wavelet systems based on the received
Number, wherein include the wavelet coefficient obtained after wavelet decomposition to original image in the data that transmitting terminal is sent；According to small echo
Coefficient generates low-resolution image；Super-resolution rebuilding is carried out to low-resolution image according to neural network, obtains high-resolution
Image, the method for compressing image solved in the related technology can not meet higher compression efficiency and higher image restoring simultaneously
The technical issues of quality, and then realize the technology that can combine higher compression efficiency and higher image restoring quality
Effect.
Detailed description of the invention
The drawings described herein are used to provide a further understanding of the present invention, constitutes part of this application, this hair
Bright illustrative embodiments and their description are used to explain the present invention, and are not constituted improper limitations of the present invention.In the accompanying drawings:
Fig. 1 is a kind of flow chart of optional image processing method according to an embodiment of the present invention；
Fig. 2 is the flow chart of another optional image processing method according to an embodiment of the present invention；
Fig. 3 is a kind of flow chart of optional wavelet decomposition according to an embodiment of the present invention；
Fig. 4 is a kind of flow chart of optional trained neural network according to an embodiment of the present invention；
Fig. 5 is the flow chart that a kind of optional transmitting terminal according to an embodiment of the present invention carries out image processing method；
Fig. 6 is the flow chart that a kind of optional receiving end according to an embodiment of the present invention carries out image processing method；
Fig. 7 is a kind of schematic diagram of optional image processing apparatus according to an embodiment of the present invention；
Fig. 8 is the schematic diagram of another optional image processing apparatus according to an embodiment of the present invention；
Fig. 9 is the schematic diagram of another optional image processing apparatus according to an embodiment of the present invention.
Specific embodiment
In order to enable those skilled in the art to better understand the solution of the present invention, below in conjunction in the embodiment of the present invention
Attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is only
The embodiment of a part of the invention, instead of all the embodiments.Based on the embodiments of the present invention, ordinary skill people
The model that the present invention protects all should belong in member's every other embodiment obtained without making creative work
It encloses.
It should be noted that description and claims of this specification and term " first " in above-mentioned attached drawing, "
Two " etc. be to be used to distinguish similar objects, without being used to describe a particular order or precedence order.It should be understood that using in this way
Data be interchangeable under appropriate circumstances, so as to the embodiment of the present invention described herein can in addition to illustrating herein or
Sequence other than those of description is implemented.In addition, term " includes " and " having " and their any deformation, it is intended that cover
Cover it is non-exclusive include, for example, the process, method, system, product or equipment for containing a series of steps or units are not necessarily limited to
Step or unit those of is clearly listed, but may include be not clearly listed or for these process, methods, product
Or other step or units that equipment is intrinsic.
This application provides a kind of embodiments of image processing method.
Fig. 1 is a kind of flow chart of optional image processing method according to an embodiment of the present invention, as shown in Figure 1, the party
Method includes the following steps:
Step S101 receives the data that transmitting terminal is sent, and data determine wavelet coefficient based on the received:
Optionally, the embodiment provide image processing method can be receiving end or decoding end execution.It is receiving
After the data that transmitting terminal is sent, wavelet coefficient is determined according to the data of receiving, wherein includes pair in the data that transmitting terminal is sent
The wavelet coefficient that original image obtains after wavelet decomposition, transmitting terminal are passing through small echo to original image (image to be transmitted)
After the size of the mode compressed file of decomposition, wavelet coefficient is obtained, optionally, the 5/3 of level-one can be used to original image
Lossless wavelet-decomposing method obtains after original image is decomposed by 5/3 lossless small echo as the wavelet-decomposing method of wavelet basis
To four subband wavelet coefficients: LL subband wavelet coefficient, LH subband wavelet coefficient, HL subband wavelet coefficient, HH subband wavelet systems
Number, can choose LL subband wavelet coefficient in four subband wavelet coefficients and be transmitted.
Optionally, transmitting terminal before transmitting, can carry out compressed encoding to wavelet coefficient, for example, transmitting terminal passes through in advance
If compression algorithm will compress the wavelet coefficient obtained after wavelet decomposition, code stream is obtained, correspondingly, is sent receiving
After holding transmitted stream, code stream can be decoded according to default compression algorithm, obtain wavelet coefficient.Verifying can obtain, only
When compressing to the LL subband wavelet coefficient that level-one wavelet decomposition obtains, code stream size is only the corresponding code stream of original image
40%-50%, compression efficiency promotes 2 times or more compared with prior art.
Step S102 generates low-resolution image according to wavelet coefficient:
After the data that transmitting terminal based on the received is sent determine wavelet coefficient, low point can be generated according to wavelet coefficient
Resolution image.Wavelet coefficient is the coefficient without dimension, which can construct phase using when carrying out wavelet decomposition with transmitting terminal
Wavelet coefficient is reconstructed in same small echo, obtains low-resolution image corresponding with wavelet coefficient after reconstruct, the low resolution
Rate image is image determined by wavelet coefficient.
Step S103 carries out super-resolution rebuilding to low-resolution image according to neural network, obtains high-definition picture:
After generating low-resolution image according to wavelet coefficient, oversubscription is carried out to low-resolution image according to neural network
Resolution is rebuild, and high-definition picture is obtained.Neural network is preparatory trained neural network.
Optionally, the neural network in the embodiment can have the function of study, and each use can be carried out updating,
Wherein, update method can be transmitting terminal according to the original image of transmission and low-resolution image as sample to continuing to optimize
After training neural network reaches preset times, the undated parameter of updated neural network is sent to receiving end, wherein pre-
If number can be repeatedly, it is also possible to once, in the case where preset times are primary, it is corresponding sends original image every time
The undated parameter of the neural network after original image training is sent when wavelet coefficient.Receiving end is receiving neural network
Undated parameter after, current neural network is updated according to undated parameter, the neural network updated and basis
The neural network of update carries out super-resolution rebuilding to current low-resolution image.
Specifically, carrying out super-resolution rebuilding to low-resolution image according to neural network comprises determining that current nerve
Network, current neural network are the nerve nets stored in device or equipment for executing the image processing method of the embodiment etc.
Network, when first time carrying out image reconstruction, current neural network is initial neural network, and initial neural network is stored in
Transmitting terminal and receiving end, after receiving undated parameter, wherein undated parameter be transmitting terminal send according to original image to working as
The parameter for the neural network that preceding neural network optimizes carries out more current neural network according to undated parameter
Newly, the neural network updated carries out super-resolution rebuilding to low-resolution image according to the neural network of update, each
After receiving the corresponding wavelet coefficient of original image and the undated parameter of neural network that optimizes according to original image,
According to undated parameter current neural network is updated, the neural network after updating is as when being rebuild next time
Current Situation of Neural Network.
Optionally, initial neural network can be in transmitting terminal generation, be also possible in receiving end generation, in life
After initial neural network, initial neural network is stored in transmitting terminal and receiving end.It is in initial neural network
In the case where receiving end generates, this method further include: wavelet decomposition is carried out to multiple sample images, is obtained corresponding multiple low
Resolution ratio sample image, wherein multiple sample images and corresponding multiple low resolution sample image composing training sample databases, instruction
Practice sample database in include multiple samples pair, each sample to include a sample image and corresponding low resolution sample image,
, can be by each sample to neural network is gradually inputted when being trained to neural network, training neural network makes will be low
High-definition picture after resolution ratio sample image input neural network reconstruction is as identical as possible as sample image, and error is less than
Preset condition.After according to multiple sample images and corresponding multiple low resolution sample image training neural networks, it will instruct
Neural network after white silk is stored in transmitting terminal and receiving end as initial neural network, wherein according to transmitting terminal first time
The data of transmission carry out determining that current neural network is initial neural network when super-resolution rebuilding.
Optionally, according to the difference of the content in image, neural network can be divided into multiple classifications, pass through different minds
Low-resolution image is rebuild through network.Specifically, in the initial neural network of determination, it can be according to preset classifier pair
Content in sample image is classified, for example, being divided into text class image and picture category image, further, picture is also
The classifications such as portrait, animation, animal can be classified as, according to different classes of image pattern to the corresponding neural network of training, are made
Preferably reconstruction effect can be reached for the image of the category by obtaining each neural network.It stores and divides in transmitting terminal and receiving end
Not Dui Yingyu multiple classifications image multiple neural networks after, receiving end can basis after determining low-resolution image
Then the classification of the content selection neural network of low-resolution image carries out low-resolution image according to the neural network of selection
Super-resolution rebuilding.
The data that the embodiment is sent by receiving transmitting terminal, and data determine wavelet coefficient based on the received, wherein hair
It include the wavelet coefficient obtained after wavelet decomposition to original image in the data that sending end is sent；It is generated according to wavelet coefficient low
Image in different resolution；Super-resolution rebuilding is carried out to low-resolution image according to neural network, high-definition picture is obtained, solves
Method for compressing image in the related technology can not meet the technology of higher compression efficiency and higher image restoring quality simultaneously
Problem, and then realize the technical effect that can combine higher compression efficiency and higher image restoring quality.
Optionally, it in above-mentioned steps, executes the module for receiving data and execution is handled (example to the data received
Such as, rebuild etc.) executing subject can be identical module, be also possible to different modules, different modules can be set
In identical equipment, also can be set on different devices, as the case may be depending on, the present invention does not limit this specifically
It is fixed.
Present invention also provides a kind of storage medium, which includes the program of storage, wherein in program operation
Equipment where control storage medium executes the image processing method of the above embodiment of the present invention offer.
Present invention also provides a kind of processor, the processor is for running program, wherein program executes this hair when running
Bright image processing method provided by the above embodiment.
Present invention also provides the embodiments of another image processing method.It should be noted that the embodiment provided
Image processing method can be by need to send original image transmitting terminal execute, this method comprises the following steps:
Step 1 carries out wavelet decomposition to original image, obtains wavelet coefficient:
Wavelet decomposition is carried out according to the small echo of default wavelet basis to original image, obtains multiple subband wavelet coefficients, it is optional
, one of wavelet coefficient can be chosen as data to be transmitted.
Step 2 generates low-resolution image according to wavelet coefficient:
The receiving end for the image processing method that the embodiment provides can be by neural network to the wavelet coefficient received
Determining low-resolution image is rebuild.
In addition to wavelet coefficient is transmitted to receiving end, the actuating station for the image processing method which provides can also be incited somebody to action
Receiving end is sent to according to the parameter of the neural network after original image training optimization.
After executing step 1 and determining wavelet coefficient, step 2 is executed, low-resolution image is generated according to wavelet coefficient.
After executing step 2, step 3 is executed.
Step 3 optimizes to obtain neural network according to original image and low-resolution image to current neural network
Undated parameter:
After generating low-resolution image according to wavelet coefficient, using original image and corresponding low-resolution image as
Training sample pair optimizes current neural network, the undated parameter of the neural network after obtaining training optimization.
The definition of current neural network is identical as the definition in above-described embodiment, and details are not described herein.
Correspondingly, this method can also include: to carry out wavelet decomposition to multiple sample images, obtain corresponding multiple low points
Resolution sample image；According to multiple sample images and corresponding multiple low resolution sample image training neural networks；It will train
Neural network afterwards is stored in transmitting terminal and receiving end as initial neural network, wherein to first original to be sent
Beginning image determines that current neural network is initial neural network when optimizing.
Optionally, include: according to multiple sample images and corresponding multiple low resolution sample image training neural networks
Classified according to the content of low resolution sample image to low resolution sample image；To the low resolution sample of each classification
Image is trained using different neural networks, obtains the neural network of multiple classifications.
Wavelet coefficient and undated parameter are sent to receiving end by step 4:
After the undated parameter and wavelet coefficient for determining neural network, wavelet coefficient and undated parameter are sent to and connect
Receiving end.
It should be noted that the sending time of wavelet coefficient and undated parameter can be it is nonsynchronous, specifically, can be with
Wavelet coefficient is first sent to receiving end after determining wavelet coefficient, at the same according to wavelet coefficient generate low-resolution image and
Then undated parameter is sent to receiving end after determining undated parameter by the step of training sample, wherein undated parameter can
To be to be sent to receiving end after training in every suboptimization, it is also possible to update after optimization training reaches preset times
Parameter is sent to receiving end.
It optionally, can be according to default compression algorithm pair before wavelet coefficient and undated parameter are sent to receiving end
The data to be sent are compressed, specifically, it includes: to wavelet coefficient that wavelet coefficient and undated parameter, which are sent to receiving end,
It is encoded with undated parameter according to default compression algorithm, obtains code stream；Obtained code stream is sent to receiving end.
The image processing method that the embodiment provides is improved from the frame structure of image coding and decoding, incorporates nerve net
Network system keeps entire algorithm frame more flexible, and more adaptable to network bandwidth, the nerve net in the image processing method
Network has unsupervised self-learning function, and with using, the effect that image is generated under the premise of not influencing performance can become better and better,
Neural network is introduced in decoding end, image can be carried out to thinner classification, improve the effect for generating image.
Optionally, it in above-mentioned steps, executes the module for receiving data and execution is handled (example to the data received
Such as, rebuild etc.) executing subject can be identical module, be also possible to different modules, different modules can be set
In identical equipment, also can be set on different devices, as the case may be depending on, the present invention does not limit this specifically
It is fixed.
Present invention also provides a kind of storage medium, which includes the program of storage, wherein in program operation
Equipment where control storage medium executes the image processing method of the above embodiment of the present invention offer.
Present invention also provides a kind of processor, the processor is for running program, wherein program executes this hair when running
Bright image processing method provided by the above embodiment.
Fig. 2 is the flow chart of another optional image processing method according to an embodiment of the present invention, which provides
Image processing method can be and executed by a kind of image processing system, which includes that the embodiment of the present invention mentions
The transmitting terminal of confession and receiving end, as shown in Fig. 2, this method comprises the following steps:
Step S201, transmitting terminal carry out wavelet decomposition to original image, obtain wavelet coefficient:
Transmitting terminal carries out wavelet decomposition according to the small echo of default wavelet basis to sent original image, obtains multiple subbands
Wavelet coefficient optionally can choose one of wavelet coefficient as data to be transmitted.
Step S202, transmitting terminal send wavelet coefficient to receiving end:
After transmitting terminal obtains wavelet coefficient, wavelet coefficient is sent to receiving end, optionally, transmitting terminal can sent
First data are compressed before data, the data after compression are sent to receiving end.
Step S203, receiving end generate low-resolution image according to wavelet coefficient:
After obtaining wavelet coefficient, receiving end generates low-resolution image according to wavelet coefficient for receiving end.Wavelet coefficient
It is the coefficient without dimension, which can construct identical small echo to wavelet coefficient using when carrying out wavelet decomposition with transmitting terminal
It is reconstructed, obtains low-resolution image corresponding with wavelet coefficient after reconstruct, which is wavelet coefficient institute
Determining image.
Step S204, receiving end carry out super-resolution rebuilding to low-resolution image according to neural network, obtain high-resolution
Rate image:
After generating low-resolution image according to wavelet coefficient, oversubscription is carried out to low-resolution image according to neural network
Resolution is rebuild, and high-definition picture is obtained.
Optionally, the neural network of transmitting terminal and receiving end can optimize update, and transmitting terminal can be according to sending every time
Original image and the corresponding low-resolution image of original image are trained the neural network continued to transmitting terminal as sample,
The parameter of the optimization neural network obtained after retraining preset times is sent to receiving end, receiving end can be according to acquisition
Parameter is updated the neural network of receiving end.
Specifically, carrying out wavelet decomposition to original image in transmitting terminal, after obtaining wavelet coefficient, this method can be with
It include: transmitting terminal according to wavelet coefficient generation low-resolution image；Transmitting terminal is according to original image and low-resolution image to hair
The current neural network of sending end optimizes；After transmitting terminal is to optimization of the neural network Jing Guo preset times, transmitting terminal to
Receiving end sends undated parameter, wherein undated parameter is the ginseng of the neural network in transmitting terminal after preset times optimization
Number；Receiving end is updated after receiving undated parameter according to neural network of the undated parameter to receiving end to obtain more
New neural network, and the subsequently received corresponding low-resolution image of wavelet coefficient is carried out based on the neural network of update
Super-resolution rebuilding.
Optionally, wavelet decomposition is carried out to original image in transmitting terminal, before obtaining wavelet coefficient, this method can also be wrapped
Include: transmitting terminal or receiving end carry out wavelet decomposition to multiple sample images, obtain corresponding multiple low resolution sample images；Hair
Sending end or receiving end are according to multiple sample images and corresponding multiple low resolution sample image training neural networks；After training
Neural network be stored in transmitting terminal and receiving end as initial neural network.
Optionally, transmitting terminal or receiving end are according to multiple sample images and corresponding multiple low resolution sample image training
Neural network may include: transmitting terminal or receiving end according to the content of low resolution sample image to low resolution sample image into
Row classification；The low resolution sample image of each classification is trained using corresponding neural network, obtains multiple classifications
Neural network.
Optionally, transmitting terminal to receiving end send wavelet coefficient may include: transmitting terminal to wavelet coefficient according to default pressure
Compression algorithm is encoded, and code stream is obtained；Obtained code stream is sent to receiving end by transmitting terminal, wherein receiving end is receiving code
It is decoded after stream according to default compression algorithm, obtains wavelet coefficient.
Present invention also provides a kind of storage medium, which includes the program of storage, wherein in program operation
Equipment where control storage medium executes the image processing method of the above embodiment of the present invention offer.
Present invention also provides a kind of processor, the processor is for running program, wherein program executes this hair when running
Bright image processing method provided by the above embodiment.
As a kind of alternative embodiment of above-described embodiment, the present invention is by the image processing method under a kind of concrete application scene
The step of method, is described as follows:
As shown in figure 3, the image processing method in the embodiment carries out the process of wavelet decomposition to original image are as follows: first
By whole original image x (n1, n2) decomposed using 5/3 small echo of promotion, it is low to row data acquisition by 5/3 boosting algorithm
Frequency signal simultaneously passes through 5/3 boosting algorithm to row data acquisition high-frequency signal, uses to row data low frequency signal and adopts under 2 times of column data
Sample obtains x1(n1, n2/ 2) x, is obtained using 2 times of down-samplings of column data to row data high-frequency signal2(n1, n2/ 2), to x1(n1, n2/
2) low frequency signal is obtained to column data using 5/3 boosting algorithm, obtains LL subband wavelet coefficient xLL{n1/ 2, n2/ 2 }, to x1(n1,
n2/ 2) high-frequency signal is obtained to column data using 5/3 boosting algorithm, obtains LH subband wavelet coefficient xLH{n1/ 2, n2/ 2 }, to x2
(n1, n2/ 2) low frequency signal is obtained to column data using 5/3 boosting algorithm, obtains HL subband wavelet coefficient xHL{n1/ 2, n2/ 2 },
To x2(n1, n2/ 2) high-frequency signal is obtained to column data using 5/3 boosting algorithm, obtains HH subband wavelet coefficient xHH{n1/ 2, n2/
2 }, that is, after wavelet decomposition tetra- sub-band informations of available LL, LH, HL and HH, wherein L indicate low-pass filtering after obtain
Low frequency signal, H indicate high-pass filtering after high-frequency signal.By this isolation a sub-picture divide in order to it is multiple compared with
Small image information.Then compressed encoding and transmission are carried out to the image after wavelet transformation.Utilize this method for compressing image one
The lossless compression of picture may be implemented in aspect, is on the one hand adapted to different network environments and realizes Delamination Transmission.
As shown in figure 4, the image processing method in the embodiment realizes the super-resolution of image by artificial neural network
The process of reconstruction is as follows: (1) choosing the biggish clearly high-resolution sample image of quantity as training sample, to each sample
Image obtains low-resolution image, obtains sample pair, each sample using 5/3 wavelet decomposition, progress down-sampling and low-pass filtering
To including an original image and corresponding low-resolution image, sample is to for training neural network parameter；(2) sample is utilized
To training neural network under the conditions of unsupervised, neural network is depth convolutional neural networks, wherein when training neural network
Constraint condition is: guarantee available corresponding high-definition picture after low-resolution image input depth convolutional neural networks,
And the high-definition picture and original sample image similarity degree meet error constraints condition, specifically, low-resolution image
Image is exported by deconvolution after into depth convolutional neural networks, the image of output is high-definition picture, by output
High-definition picture is compared with original high-resolution sample image, calculates error, corrective networks, then by next sample
This carries out training next time to input depth convolutional neural networks；It optionally, can be according to the content difference of sample image point
Multiple neural networks, the picture material of the corresponding classification of each neural network, for example, dividing sample image are not trained
Class, judgement are character image or picture, and different classes of image is trained using different neural networks；(3) pass through
After a large amount of sample training, available trained text neural network and picture neural network can after training is completed
To choose the low resolution picture with sample same type, whether test neural network can use the picture and generates corresponding high score
Resolution picture.By above step, that is, the training of the neural network for image super-resolution rebuilding is completed, to network parameter
It saves, saves two trained neural networks in the form of data in coding side (transmitting terminal) and decoding end (receiving end).
Subsequent in use, the trained neural network can be used as " flight data recorder ", Lai Shixian low-resolution image surpasses
Resolution reconstruction.
As shown in figure 5, the process of the image processing method of coding side (transmitting terminal) is as follows:
Original image to be sent does level of decomposition using 5/3 small echo, low-resolution image is obtained, by low resolution figure
As carrying out compressed encoding, and transmission code stream to LL low frequency sub-band using existing image compression algorithm.
Level of decomposition is being done using 5/3 small echo, after obtaining low-resolution image, also according to original image and low resolution
Image is updated the neural network of coding side, specifically, low-resolution image input classifier is identified, judges class
Type, determination is character image or picture, if it is character image, then calls the neural network of character image to character image
Carry out image super-resolution rebuilding and optimize text neural network according to reconstructed results then to call picture if it is picture
The neural network of image carries out image super-resolution rebuilding to picture and optimizes picture neural network, according to reconstructed results
After optimization reaches preset times, the parameter of the neural network after optimization is sent to decoding end, decoding end is receiving update
The neural network of decoding end storage can be updated after parameter.
It is transmitted to as shown in fig. 6, transmitting terminal after determining 5/3 wavelet decomposition low-resolution image, carries out compressed encoding
The process of receiving end, the image processing method of decoding end (receiving end) is as follows:
Receiving end is decoded data, obtains low-resolution image, and input classifier is identified, which can be with
It is classifier identical with the classifier of transmitting terminal, judges to be the corresponding nerve of calling after character image or picture
Network carries out image super-resolution rebuilding, generates high-definition picture, exports image.Wherein, the neural network of decoding end can be with
It updates, specifically, transmitting terminal is after sending undated parameter, current neural network is updated by decoding end.
Optionally, in the image processing method that the embodiment provides, using 5/3 small echo as wavelet basis tectonic network,
In practical applications, other wavelet basis can also be used and carry out tectonic network, for example, Haar small echo or 9/7 small echo etc..
Image processing method provided in an embodiment of the present invention can write program using C language or C Plus Plus, wherein C language
Say the better performances of program, the code organization of C++ is relatively good but operational efficiency is lower than C programmer, preferably using C language into
Row programming.
The image processing method that the embodiment provides will be the characteristics of wavelet decomposition and deep neural network image super-resolution
It combines, proposes a kind of technical solution of new image compressing transmission, generate sample database using the method for 5/3 wavelet decomposition,
Sample database preliminary classification is that character image and picture two major classes are other, using sample database training deep neural network, is answered
Two neural networks for character image and picture；Then, 5/3 wavelet decomposition is utilized to image in transmitting terminal, it will be former
The LL subband wavelet coefficient of figure carries out coding transmission；Finally, in receiving end according to image category, it is utilized respectively text or picture
Neural network carries out super-resolution rebuilding to picture, obtains high-definition picture and exports.
Traditional picture compression method has been difficult to have greatly improved on the basis of existing algorithm, which mentions
This method for compressing image based on neural network image super-resolution that the image processing method of confession proposes, inherently provides
New compression of images thinking, can significantly promote compression efficiency.It is passed using the low-resolution image of level-one wavelet decomposition
It, can be by original picture compression improved efficiency 2 times or more when defeated.The program can use the spy of the unsupervised self study of neural network
Point, the optimization of invisible completion neural network in image encoding process, constantly promotion network performance, makes to generate picture quality not
It is disconnected to be promoted.
Verifying can obtain, and when only compressing to the LL subband wavelet coefficient that level-one wavelet decomposition obtains, code stream size is only
Be the 40%~50% of original bit stream, compression efficiency compared with the prior art promoted 2 times or more, and decode generate high-definition picture with
Original image difference is little, can satisfy vision needs, in addition, the unsupervised learning feature of neural network is utilized in the program, makes
Picture quality available continuous promotion during use, so that the Neural Network Self-learning of decoding end optimizes, system
Performance constantly improve with using the time elongated.
The image processing method that the embodiment provides can be applied in computer screen picture transmission, and screen-picture transmission can
Applied to scenes such as video conference, remote desktops.These application scenarios are higher to network bandwidth requirement, and the size of bandwidth will be direct
Influence the quality of image and the smooth degree of picture.Therefore, the image processing method of the program utilizes wavelet transformation and depth mind
The characteristics of network, is proposed a kind of new compression of images encoding and decoding transmission method, can be by original compression using this method
Efficiency improves 2 times or more.
It should be noted that attached drawing flow chart though it is shown that logical order, but in some cases, can be with
Shown or described step is executed different from sequence herein.
Present invention also provides a kind of embodiments of image processing apparatus.The device can be to be set to and go back to image
The device of former one end, can be used for executing the image processing method provided in an embodiment of the present invention for being restored to image
Method.
Fig. 7 is a kind of schematic diagram of optional image processing apparatus according to an embodiment of the present invention, as shown in fig. 7, the dress
It sets including receiving unit 10, generation unit 20 and reconstruction unit 30, wherein receiving unit is used to receive the number of transmitting terminal transmission
According to, and data determine wavelet coefficient based on the received, wherein it include to original image in the data that transmitting terminal is sent by small echo
The wavelet coefficient obtained after decomposition；Generation unit is used to generate low-resolution image according to wavelet coefficient；Reconstruction unit is used for root
Super-resolution rebuilding is carried out to low-resolution image according to neural network, obtains high-definition picture.
The embodiment is by receiving unit, for receiving the data of transmitting terminal transmission, and based on the received data determine it is small
Wave system number, wherein include the wavelet coefficient obtained after wavelet decomposition to original image in the data that transmitting terminal is sent；It generates
Unit, for generating low-resolution image according to wavelet coefficient；Reconstruction unit is used for according to neural network to low-resolution image
Super-resolution rebuilding is carried out, obtains high-definition picture, the method for compressing image solved in the related technology can not meet simultaneously
The technical issues of higher compression efficiency and higher image restoring quality, and then realize and can combine higher compression
The technical effect of efficiency and higher image restoring quality.
Further, reconstruction unit comprises determining that module, for determining current neural network；First receiving module is used
In reception undated parameter, wherein undated parameter is the excellent to current neural network progress according to original image of transmitting terminal transmission
Change the parameter of obtained neural network；Update module is obtained for being updated according to undated parameter to current neural network
The neural network of update；First rebuilds module, for carrying out super-resolution to low-resolution image according to the neural network of update
It rebuilds.
Further, device further include: decomposition unit obtains pair for carrying out wavelet decomposition to multiple sample images
The multiple low resolution sample images answered；Training unit, for according to multiple sample images and corresponding multiple low resolution samples
This image trains neural network；Storage unit, for being stored in hair for the neural network after training as initial neural network
Sending end and receiving end, wherein determine current mind when the data sent for the first time according to transmitting terminal carry out super-resolution rebuilding
It is initial neural network through network.
Further, reconstruction unit includes: selecting module, for the content selection neural network according to low-resolution image
Classification；Second rebuilds module, for carrying out super-resolution rebuilding to low-resolution image according to the neural network of selection.
Further, receiving unit includes: the second receiving module, for receiving transmitting terminal transmitted stream；Decoder module,
For being decoded according to default compression algorithm to code stream, wavelet coefficient is obtained, wherein default compression algorithm is transmitting terminal to warp
Cross the compression algorithm used when the wavelet coefficient obtained after wavelet decomposition is compressed.
Fig. 8 is the schematic diagram of another optional image processing apparatus according to an embodiment of the present invention, which can be
It is set to the device of the one end compressed to image, can be used for executing provided in an embodiment of the present invention for carrying out image
The image processing method of compression, as shown in figure 8, the device includes decomposition unit 11, generation unit 21 optimizes unit 31 and sends
Unit 41, wherein decomposition unit is used to carry out wavelet decomposition to original image, obtains wavelet coefficient；Generation unit is used for basis
Wavelet coefficient generates low-resolution image；Optimize unit to be used for according to original image and low-resolution image to current nerve net
Network optimizes to obtain the undated parameter of neural network；Transmission unit is used to wavelet coefficient and undated parameter being sent to reception
End.
Further, decomposition unit is also used to carry out wavelet decomposition to multiple sample images, obtains corresponding multiple low points
Resolution sample image, the device further include: training unit, for according to multiple sample images and corresponding multiple low resolution samples
This image trains neural network；Storage unit, for being stored in hair for the neural network after training as initial neural network
Sending end and receiving end, wherein determine that current neural network is first when optimizing to first original image to be sent
The neural network of beginning.
Further, training unit includes: categorization module, for the content according to low resolution sample image to low resolution
Rate sample image is classified；Training module uses different nerve nets for the low resolution sample image to each classification
Network is trained, and obtains the neural network of multiple classifications.
Further, transmission unit includes: decoder module, for being calculated according to default compression wavelet coefficient and undated parameter
Method is encoded, and code stream is obtained；Sending module, for obtained code stream to be sent to receiving end.
Fig. 9 is the schematic diagram of another optional image processing apparatus according to an embodiment of the present invention, which can use
In the image processing method for executing the corresponding embodiment offer of Fig. 2 of the present invention, as shown in figure 9, the device includes decomposition unit 12,
Transmission unit 22, generation unit 32 and reconstruction unit 42, wherein decomposition unit is used to carry out original image by transmitting terminal small
Wave Decomposition obtains wavelet coefficient；Transmission unit is used to send wavelet coefficient to receiving end by transmitting terminal；Generation unit is for leading to
It crosses receiving end and low-resolution image is generated according to wavelet coefficient；Reconstruction unit is for receiving end according to neural network to low resolution
Image carries out super-resolution rebuilding, obtains high-definition picture.
Above-mentioned device may include processor and memory, and said units can be used as program unit and be stored in storage
In device, above procedure unit stored in memory is executed by processor to realize corresponding function.
Memory may include the non-volatile memory in computer-readable medium, random access memory (RAM) and/
Or the forms such as Nonvolatile memory, if read-only memory (ROM) or flash memory (flash RAM), memory include that at least one is deposited
Store up chip.
The sequence of above-mentioned the embodiment of the present application does not represent the advantages or disadvantages of the embodiments.
In above-described embodiment of the application, all emphasizes particularly on different fields to the description of each embodiment, do not have in some embodiment
The part of detailed description, reference can be made to the related descriptions of other embodiments.In several embodiments provided herein, it should be appreciated that
It arrives, disclosed technology contents can be realized in other ways.
Wherein, the apparatus embodiments described above are merely exemplary, such as the division of the unit, can be one
Kind of logical function partition, there may be another division manner in actual implementation, for example, multiple units or components can combine or
It is desirably integrated into another system, or some features can be ignored or not executed.Another point, it is shown or discussed it is mutual it
Between coupling, direct-coupling or communication connection can be through some interfaces, the INDIRECT COUPLING or communication link of unit or module
It connects, can be electrical or other forms.
It, can also be in addition, each functional unit in each embodiment of the application can integrate in one processing unit
It is that each unit physically exists alone, can also be integrated in one unit with two or more units.Above-mentioned integrated list
Member both can take the form of hardware realization, can also realize in the form of software functional units.
If the integrated unit is realized in the form of SFU software functional unit and sells or use as independent product
When, it can store in a computer readable storage 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 for personal computer, server or network equipment etc.) execute each embodiment the method for the application whole or
Part steps.And storage medium above-mentioned includes: that USB flash disk, read-only memory (ROM, Read-Only Memory), arbitrary access are deposited
Reservoir (RAM, Random Access Memory), mobile hard disk, magnetic or disk etc. be various to can store program code
Medium.
The above is only the preferred embodiment of the application, it is noted that for the ordinary skill people of the art
For member, under the premise of not departing from the application principle, several improvements and modifications can also be made, these improvements and modifications are also answered
It is considered as the protection scope of the application.
Claims (11)
1. a kind of image processing method characterized by comprising
The data that transmitting terminal is sent are received, and data determine wavelet coefficient based on the received, wherein the number that the transmitting terminal is sent
It include the wavelet coefficient obtained after wavelet decomposition to original image in；
Low-resolution image is generated according to the wavelet coefficient；
Super-resolution rebuilding is carried out to the low-resolution image according to neural network, obtains high-definition picture；
It wherein, include the wavelet coefficient packet obtained after wavelet decomposition to original image in the data that the transmitting terminal is sent
It includes: using 5/3 lossless wavelet-decomposing method of level-one to the original image, the original image passes through 5/3 lossless small echo
After wavelet-decomposing method as wavelet basis is decomposed, obtain four subband wavelet coefficients: LL subband wavelet coefficient, LH subband are small
Wave system number, HL subband wavelet coefficient, HH subband wavelet coefficient, it is small in four subband wavelet coefficients to choose the LL subband
Wave system number is transmitted；
Wherein, the method also includes: to multiple sample images carry out wavelet decomposition, obtain corresponding multiple low resolution samples
Image；According to the multiple sample image and corresponding multiple low resolution sample image training neural networks；It will train
Neural network afterwards is stored in the transmitting terminal and receiving end as initial neural network, wherein according to the transmitting terminal
The data sent for the first time carry out determining that current neural network is the initial neural network when super-resolution rebuilding；
Wherein, the multiple sample image and corresponding multiple low resolution sample image composing training sample databases, training sample
It include multiple samples pair in library, each sample is to including a sample image and corresponding low resolution sample image, to mind
When being trained through network, by each sample to gradually inputting neural network.
2. the method according to claim 1, wherein being surpassed according to neural network to the low-resolution image
Resolution reconstruction includes:
Determine current neural network；
Receive undated parameter, wherein the undated parameter is working as according to the original image to described for transmitting terminal transmission
The parameter for the neural network that preceding neural network optimizes；
The current neural network is updated according to the undated parameter, the neural network updated；
Super-resolution rebuilding is carried out to the low-resolution image according to the neural network of the update.
3. the method according to claim 1, wherein being surpassed according to neural network to the low-resolution image
Resolution reconstruction includes:
According to the classification of the content selection neural network of the low-resolution image；
Super-resolution rebuilding is carried out to the low-resolution image according to the neural network of selection.
4. a kind of image processing method characterized by comprising
Transmitting terminal carries out wavelet decomposition to original image, obtains wavelet coefficient；
The transmitting terminal sends the wavelet coefficient to receiving end；
The receiving end generates low-resolution image according to the wavelet coefficient；
The receiving end carries out super-resolution rebuilding to the low-resolution image according to neural network, obtains high resolution graphics
Picture；
It wherein, include the wavelet coefficient packet obtained after wavelet decomposition to original image in the data that the transmitting terminal is sent
It includes: using 5/3 lossless wavelet-decomposing method of level-one to the original image, the original image passes through 5/3 lossless small echo
After wavelet-decomposing method as wavelet basis is decomposed, obtain four subband wavelet coefficients: LL subband wavelet coefficient, LH subband are small
Wave system number, HL subband wavelet coefficient, HH subband wavelet coefficient, it is small in four subband wavelet coefficients to choose the LL subband
Wave system number is transmitted；
Wavelet decomposition is carried out to original image in transmitting terminal, before obtaining wavelet coefficient, the method also includes: the transmitting terminal
Or the receiving end carries out wavelet decomposition to multiple sample images, obtains corresponding multiple low resolution sample images；The hair
Sending end or the receiving end are according to the multiple sample image and corresponding multiple low resolution sample image training nerves
Network；The transmitting terminal and the receiving end are stored in using the neural network after training as initial neural network.
5. according to the method described in claim 4, it is characterized in that, being obtained in transmitting terminal to original image progress wavelet decomposition
After wavelet coefficient, the method also includes:
The transmitting terminal generates low-resolution image according to the wavelet coefficient；
The transmitting terminal according to the original image and the low-resolution image neural network current to the transmitting terminal into
Row optimization；
After the transmitting terminal is to optimization of the neural network Jing Guo preset times, the transmitting terminal is sent more to the receiving end
New parameter, wherein the undated parameter is the parameter of the neural network in the transmitting terminal after preset times optimization；
The receiving end is after receiving the undated parameter, according to the undated parameter to the neural network of the receiving end
It is updated the neural network to be updated, and the neural network based on the update is to subsequently received wavelet coefficient pair
The low-resolution image answered carries out super-resolution rebuilding.
6. according to the method described in claim 4, it is characterized in that, the transmitting terminal or the receiving end are according to the multiple sample
This image and corresponding multiple low resolution sample image training neural networks include:
The transmitting terminal or the receiving end are according to the content of the low resolution sample image to the low resolution sample graph
As classifying；
The low resolution sample image of each classification is trained using corresponding neural network, obtains the nerve of multiple classifications
Network.
7. according to the method described in claim 4, it is characterized in that, the transmitting terminal sends the wavelet coefficient packet to receiving end
It includes:
The transmitting terminal encodes the wavelet coefficient according to default compression algorithm, obtains code stream；
The obtained code stream is sent to the receiving end by the transmitting terminal, wherein the receiving end is receiving the code
It is decoded after stream according to the default compression algorithm, obtains the wavelet coefficient.
8. a kind of image processing apparatus characterized by comprising
Receiving unit, for receiving the data of transmitting terminal transmission, and data determine wavelet coefficient based on the received, wherein described
It include the wavelet coefficient obtained after wavelet decomposition to original image in the data that transmitting terminal is sent；
Generation unit, for generating low-resolution image according to the wavelet coefficient；
Reconstruction unit obtains high-resolution for carrying out super-resolution rebuilding to the low-resolution image according to neural network
Image；
It wherein, include the wavelet coefficient packet obtained after wavelet decomposition to original image in the data that the transmitting terminal is sent
It includes: using 5/3 lossless wavelet-decomposing method of level-one to the original image, the original image passes through 5/3 lossless small echo
After wavelet-decomposing method as wavelet basis is decomposed, obtain four subband wavelet coefficients: LL subband wavelet coefficient, LH subband are small
Wave system number, HL subband wavelet coefficient, HH subband wavelet coefficient, it is small in four subband wavelet coefficients to choose the LL subband
Wave system number is transmitted；
Wherein, wavelet decomposition is carried out to multiple sample images, obtains corresponding multiple low resolution sample images；According to described more
A sample image and corresponding multiple low resolution sample image training neural networks；Using the neural network after training as
Initial neural network is stored in the transmitting terminal and receiving end, wherein in the data sent for the first time according to the transmitting terminal
It carries out determining that current neural network is the initial neural network when super-resolution rebuilding；
Wherein, the multiple sample image and corresponding multiple low resolution sample image composing training sample databases, training sample
It include multiple samples pair in library, each sample is to including a sample image and corresponding low resolution sample image, to mind
When being trained through network, by each sample to gradually inputting neural network.
9. device according to claim 8, which is characterized in that the reconstruction unit includes:
Determining module, for determining current neural network；
First receiving module, for receiving undated parameter, wherein the undated parameter is transmitting terminal transmission according to
The parameter for the neural network that original image optimizes the current neural network；
Update module, for being updated according to the undated parameter to the current neural network, the nerve updated
Network；
First rebuilds module, for carrying out Super-resolution reconstruction to the low-resolution image according to the neural network of the update
It builds.
10. device according to claim 8, which is characterized in that the reconstruction unit includes:
Selecting module, for the classification according to the content selection neural network of the low-resolution image；
Second rebuilds module, for carrying out super-resolution rebuilding to the low-resolution image according to the neural network of selection.
11. a kind of image processing apparatus characterized by comprising
Decomposition unit obtains wavelet coefficient for carrying out wavelet decomposition to original image by transmitting terminal；
Transmission unit, for sending the wavelet coefficient to receiving end by the transmitting terminal；
Generation unit, for generating low-resolution image according to the wavelet coefficient by the receiving end；
Reconstruction unit carries out super-resolution rebuilding to the low-resolution image according to neural network for the receiving end, obtains
To high-definition picture；
It wherein, include the wavelet coefficient packet obtained after wavelet decomposition to original image in the data that the transmitting terminal is sent
It includes: using 5/3 lossless wavelet-decomposing method of level-one to the original image, the original image passes through 5/3 lossless small echo
After wavelet-decomposing method as wavelet basis is decomposed, obtain four subband wavelet coefficients: LL subband wavelet coefficient, LH subband are small
Wave system number, HL subband wavelet coefficient, HH subband wavelet coefficient, it is small in four subband wavelet coefficients to choose the LL subband
Wave system number is transmitted；
Wavelet decomposition, before obtaining wavelet coefficient, the transmitting terminal or the receiving end pair are carried out to original image in transmitting terminal
Multiple sample images carry out wavelet decomposition, obtain corresponding multiple low resolution sample images；The transmitting terminal or the reception
End is according to the multiple sample image and corresponding multiple low resolution sample image training neural networks；After training
Neural network is stored in the transmitting terminal and the receiving end as initial neural network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710326762.4A CN107087201B (en) | 2017-05-10 | 2017-05-10 | Image processing method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710326762.4A CN107087201B (en) | 2017-05-10 | 2017-05-10 | Image processing method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107087201A CN107087201A (en) | 2017-08-22 |
| CN107087201B true CN107087201B (en) | 2019-11-08 |
Family
ID=59611522
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710326762.4A Active CN107087201B (en) | 2017-05-10 | 2017-05-10 | Image processing method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107087201B (en) |
Families Citing this family (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108876864B (en) * | 2017-11-03 | 2022-03-08 | 北京旷视科技有限公司 | Image encoding and decoding method, apparatus, electronic device, and computer-readable medium |
| CN108197488B (en) * | 2017-12-25 | 2020-04-14 | 大国创新智能科技(东莞)有限公司 | Information hiding and extraction method and system based on big data and neural network |
| KR102749961B1 (en) | 2018-01-04 | 2025-01-09 | 삼성전자주식회사 | Video playback device and controlling method thereof |
| CN108564546B (en) * | 2018-04-18 | 2020-08-04 | 厦门美图之家科技有限公司 | Model training method and device and photographing terminal |
| CN108960425B (en) * | 2018-07-05 | 2022-04-19 | 广东工业大学 | Rendering model training method, system, equipment, medium and rendering method |
| KR102865289B1 (en) * | 2019-08-14 | 2025-09-29 | 삼성전자주식회사 | Electronic apparatus and control method thereof |
| CN112801872A (en) * | 2021-02-02 | 2021-05-14 | 上海眼控科技股份有限公司 | Image processing method, device, equipment and storage medium |
| CN114841308A (en) * | 2022-03-17 | 2022-08-02 | 阿里巴巴（中国）有限公司 | Super-resolution reconstruction method, device and equipment for cloud desktop image and storage medium |
| CN116051369A (en) * | 2022-11-01 | 2023-05-02 | 深信服科技股份有限公司 | Virtual desktop image processing method, device and storage medium |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10302727B2 (en) * | 2015-07-02 | 2019-05-28 | The General Hospital Corporation | System and method for high resolution diffusion imaging |
| CN106067161A (en) * | 2016-05-24 | 2016-11-02 | 深圳市未来媒体技术研究院 | A kind of method that image is carried out super-resolution |
| CN106204447A (en) * | 2016-06-30 | 2016-12-07 | 北京大学 | The super resolution ratio reconstruction method with convolutional neural networks is divided based on total variance |
| CN106228512A (en) * | 2016-07-19 | 2016-12-14 | 北京工业大学 | Based on learning rate adaptive convolutional neural networks image super-resolution rebuilding method |
-
2017
- 2017-05-10 CN CN201710326762.4A patent/CN107087201B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN107087201A (en) | 2017-08-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107087201B (en) | Image processing method and device | |
| Li et al. | Learning convolutional networks for content-weighted image compression | |
| US8223837B2 (en) | Learning-based image compression | |
| EP3583777A1 (en) | A method and technical equipment for video processing | |
| DE102023133577A1 (en) | SPEECH AUDIO COMPRESSION USING NEURAL NETWORKS | |
| KR20210018668A (en) | Downsampling image processing system and mehod using deep learning neural network and image streaming server system | |
| CN113409216A (en) | Image restoration method based on frequency band self-adaptive restoration model | |
| WO2020238439A1 (en) | Video quality-of-service enhancement method under restricted bandwidth of wireless ad hoc network | |
| KR20240025629A (en) | Video compression using optical flow | |
| CN115409716B (en) | Video processing method, device, storage medium and equipment | |
| CN117061763A (en) | Vincent video generation method, device, electronic equipment and readable storage medium | |
| JP2004505520A (en) | Video coding method using wavelet decomposition | |
| CN117857795B (en) | Method and system for compressing and decompressing limit image by combining generation and entropy estimation | |
| CN120455700A (en) | Method and system for restoring compressed images with raindrops | |
| CN113225554A (en) | Image coding and decoding method and device based on neural network, storage medium and terminal | |
| Jeny et al. | Improving image compression with adjacent attention and refinement block | |
| CN120281925A (en) | Intelligent image compression encoder based on conditional reversible neural network | |
| CN112637609B (en) | Image real-time transmission method, sending end and receiving end | |
| CN115063494B (en) | Mars image compression method, device, computer equipment and storage medium | |
| CN113497938B (en) | Method and device for image compression and decompression based on variational autoencoder | |
| CN116228895A (en) | Video generation method, deep learning model training method, device and equipment | |
| Mahajan et al. | Pixelpyramids: Exact inference models from lossless image pyramids | |
| JP7752353B2 (en) | Terminal, program, and information processing method for displaying low-resolution images as high-resolution images | |
| CN116132737B (en) | Data processing method, live broadcast method and device, electronic device, and storage medium | |
| Xiong et al. | Deep feature compression with collaborative coding of image texture |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |