CN109615589A - Remove the method, apparatus and terminal device of picture noise - Google Patents
Remove the method, apparatus and terminal device of picture noise Download PDFInfo
- Publication number
- CN109615589A CN109615589A CN201811290243.8A CN201811290243A CN109615589A CN 109615589 A CN109615589 A CN 109615589A CN 201811290243 A CN201811290243 A CN 201811290243A CN 109615589 A CN109615589 A CN 109615589A
- Authority
- CN
- China
- Prior art keywords
- image
- image processing
- processing block
- network
- noise reduction
- 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 88
- 238000012545 processing Methods 0.000 claims abstract description 273
- 239000010410 layer Substances 0.000 claims abstract description 177
- 230000009467 reduction Effects 0.000 claims abstract description 73
- 230000008569 process Effects 0.000 claims abstract description 42
- 238000011946 reduction process Methods 0.000 claims abstract description 22
- 230000005540 biological transmission Effects 0.000 claims abstract description 21
- 239000011229 interlayer Substances 0.000 claims abstract description 14
- 238000005070 sampling Methods 0.000 claims abstract description 10
- 238000003860 storage Methods 0.000 claims abstract description 9
- 238000012549 training Methods 0.000 claims description 25
- 238000004891 communication Methods 0.000 description 7
- 238000010586 diagram Methods 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 6
- 230000003287 optical effect Effects 0.000 description 4
- 230000005236 sound signal Effects 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 238000004590 computer program Methods 0.000 description 3
- 230000001133 acceleration Effects 0.000 description 2
- 230000000712 assembly Effects 0.000 description 2
- 238000000429 assembly Methods 0.000 description 2
- 239000002131 composite material Substances 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000002093 peripheral effect Effects 0.000 description 2
- 238000012163 sequencing technique Methods 0.000 description 2
- KLDZYURQCUYZBL-UHFFFAOYSA-N 2-[3-[(2-hydroxyphenyl)methylideneamino]propyliminomethyl]phenol Chemical compound OC1=CC=CC=C1C=NCCCN=CC1=CC=CC=C1O KLDZYURQCUYZBL-UHFFFAOYSA-N 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 201000001098 delayed sleep phase syndrome Diseases 0.000 description 1
- 208000033921 delayed sleep phase type circadian rhythm sleep disease Diseases 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000003825 pressing Methods 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- 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
The disclosure is directed to a kind of method and apparatus, mobile terminal and storage mediums for removing picture noise, the method comprise the steps that image to be processed is inputted in trained image noise reduction model in advance, wherein, described image noise reduction model includes multiple images process block, include multiple network layers in other image processing blocks in addition to first image processing block, maximum pond layer or up-sampling layer are provided between adjacent two image processing block;From the first image processing block, noise reduction process successively is carried out to the image to be processed according to network inter-layer data transmission rule in each described image process block, the target image after obtaining noise reduction.Noise reduction process is carried out to image by the method for the removal noise image of the disclosure, picture noise can either be removed and be able to maintain image detail again.
Description
Technical field
This disclosure relates to technical field of image processing more particularly to a kind of method, apparatus and terminal for removing picture noise
Equipment.
Background technique
Noise is the major reason of image interference, piece image in practical applications there may be various noises,
These noises may generate in the transmission, it is also possible to generate in the processing such as quantization.Image noise reduction is one in image preprocessing
Widely used technology, effect are the signal-to-noise ratio in order to improve image, the desired character of prominent image.
Image noise reduction can be carried out in time domain, can also be carried out in frequency, but be all based on noise after all and signal exists
Different distributions rule on frequency domain, i.e., signal is mainly distributed on low frequency region and noise is mainly distributed on high-frequency region.However by
It is also distributed about high-frequency region in the details of image, therefore key when image noise reduction processing is to protect while reducing picture noise
Stay image detail.Current more common image denoising method is low-pass filtering method, and this method filters the radio-frequency component of image
It removes, although can achieve the purpose that reduce picture noise, this kind of mode can destroy image detail.As it can be seen that compeling to be essential at present
Those skilled in the art are wanted to provide a kind of picture noise that removes that can either be reduced picture noise and be able to maintain image detail again
Method.
Summary of the invention
To overcome the problems in correlation technique, present disclose provides it is a kind of remove picture noise method, apparatus and
Terminal device.
According to the first aspect of the embodiments of the present disclosure, a kind of method for removing picture noise is provided, wherein the method
It include: to input image to be processed in trained image noise reduction model in advance, wherein described image noise reduction model includes multiple
Image processing block, includes multiple network layers in other image processing blocks in addition to first image processing block, at adjacent two image
Maximum pond layer or up-sampling layer are provided between reason block;From the first image processing block, successively according to each described image at
It manages network inter-layer data transmission rule in block and noise reduction process is carried out to the image to be processed, the target image after obtaining noise reduction.
Optionally, described image noise reduction model includes the first image processing block being sequentially arranged, the second image processing block, the
Three image processing blocks, the 4th image processing block, the 5th image processing block, the 6th image processing block, the 7th image processing block, the 8th
Image processing block, the 9th image processing block and the tenth image processing block, wherein the first image process block is described first
Image processing block;Last layer network in second image processing block respectively with each network layer in the tenth image processing block
Foundation has data connection;Last layer network in the third image processing block respectively with each net in the 9th image processing block
Network layers foundation has data connection;Last layer network in 4th image processing block is respectively and in the 8th image processing block
Each network layer foundation has data connection;Last layer network in 5th image processing block respectively with the 7th image processing block
In each network layer foundation have data connection.
It optionally, include six network layers, each figure in addition to first image processing block in each described image process block
As the convolution kernel size of first three network layer of process block is 5 × 5, the convolution kernel sizes of rear three network layers is 1 × 1.
Optionally, each network layer is formed by Conv layers, Relu layers and Hardtanh layers respectively.
Optionally, before described by the step in image to be processed input in advance trained image noise reduction model, institute
State method further include: acquire the image pattern of preset quantity;Carry out each described image sample plus make an uproar respectively processing, obtains noise
Image pattern;Each described image sample is formed into training image pair with corresponding noise image sample;According to each training figure
As being trained to image noise reduction model.
According to the second aspect of an embodiment of the present disclosure, a kind of device for removing picture noise is provided, wherein described device packet
Include: input module is configured as inputting image to be processed in trained image noise reduction model in advance, wherein described image
Noise reduction model includes multiple images process block, includes multiple networks in other image processing blocks in addition to first image processing block
Layer, maximum pond layer or up-sampling layer are provided between adjacent two image processing block;Noise reduction process module is configured as from the head
A image processing block rises, successively according to network inter-layer data transmission rule in each described image process block to the image to be processed
Carry out noise reduction process, the target image after obtaining noise reduction.
Optionally, described image noise reduction model includes the first image processing block being sequentially arranged, the second image processing block, the
Three image processing blocks, the 4th image processing block, the 5th image processing block, the 6th image processing block, the 7th image processing block, the 8th
Image processing block, the 9th image processing block and the tenth image processing block, wherein the first image process block is described first
Image processing block;Last layer network in second image processing block respectively with each network layer in the tenth image processing block
Foundation has data connection;Last layer network in the third image processing block respectively with each net in the 9th image processing block
Network layers foundation has data connection;Last layer network in 4th image processing block is respectively and in the 8th image processing block
Each network layer foundation has data connection;Last layer network in 5th image processing block respectively with the 7th image processing block
In each network layer foundation have data connection.
It optionally, include six network layers, each figure in addition to first image processing block in each described image process block
As the convolution kernel size of first three network layer of process block is 5 × 5, the convolution kernel sizes of rear three network layers is 1 × 1.
Optionally, each network layer is formed by Conv layers, Relu layers and Hardtanh layers respectively.
Optionally, described device further include: acquisition module is configured as inputting image to be processed in the input module
Before in preparatory trained image noise reduction model, the image pattern of preset quantity is acquired;Add module of making an uproar, be configured to by
Each described image sample carries out adding processing of making an uproar, and obtains noise image sample;Composite module is configured as each described image sample
Training image pair is formed with corresponding noise image sample;Training module is configured as according to each training image to image
Noise reduction model is trained.
According to the third aspect of an embodiment of the present disclosure, a kind of terminal device is provided, comprising: processor;It is handled for storage
The memory of device executable instruction;Wherein, the processor is configured to the method for executing any of the above-described kind of removal picture noise.
According to a fourth aspect of embodiments of the present disclosure, a kind of non-transitorycomputer readable storage medium is provided, when described
When instruction in storage medium is executed by the processor of terminal device, so that terminal device executes any of the above-described kind of removal image and makes an uproar
The method of sound.
According to a fifth aspect of the embodiments of the present disclosure, it provides according to a kind of computer program product, when the computer journey
When instruction in sequence product is executed by the processor of mobile terminal, so that terminal device executes any of the above-described kind of removal picture noise
Method.
The technical scheme provided by this disclosed embodiment can include the following benefits:
Image to be processed is inputted trained packet in advance by the scheme for the removal picture noise that embodiment of the disclosure provides
In the image noise reduction model of the process block containing multiple images, from first image processing block, successively according to net in each image processing block
Network inter-layer data transmission rule carries out noise reduction process to image to be processed, and the target image after obtaining noise reduction is dropped by the image
Although the process that the image processing block of the preceding preset quantity in model of making an uproar carries out noise reduction process removal noise to image can have figure
As loss in detail, but the image detail data of loss can be transmitted to below by the image processing block of preceding preset quantity by cross-layer
Image processing block in, image detail is made up, image noise reduction and image detail can be combined.
It should be understood that above general description and following detailed description be only it is exemplary and explanatory, not
The disclosure can be limited.
Detailed description of the invention
The drawings herein are incorporated into the specification and forms part of this specification, and shows the implementation for meeting the disclosure
Example, and together with specification for explaining the principles of this disclosure.
Fig. 1 is a kind of step flow chart of method for removing picture noise shown according to an exemplary embodiment;
Fig. 2 is a kind of step flow chart of method for removing picture noise shown according to an exemplary embodiment;
Fig. 3 is image noise reduction model schematic;
Fig. 4 is a kind of block diagram of device for removing picture noise shown according to an exemplary embodiment;
Fig. 5 is a kind of structural block diagram of terminal device shown according to an exemplary embodiment;
Fig. 6 is a kind of structural block diagram of terminal device shown according to an exemplary embodiment.
Specific embodiment
Example embodiments are described in detail here, and the example is illustrated in the accompanying drawings.Following description is related to
When attached drawing, unless otherwise indicated, the same numbers in different drawings indicate the same or similar elements.Following exemplary embodiment
Described in embodiment do not represent all implementations consistent with this disclosure.On the contrary, they be only with it is such as appended
The example of the consistent device and method of some aspects be described in detail in claims, the disclosure.
Fig. 1 is a kind of flow chart of method for removing picture noise shown according to an exemplary embodiment, as shown in Figure 1
Removal picture noise method in terminal device, comprising the following steps:
Step 101: image to be processed is inputted in trained image noise reduction model in advance.
Wherein, image noise reduction model includes multiple images process block, other image procossings in addition to first image processing block
Include multiple network layers in block, maximum pond layer or up-sampling layer are provided between adjacent two image processing block.Image noise reduction mould
3 channel images of input can be mapped as 16 channel images by the first image processing block in type, therefore first image processing block is only
Comprising a network layer, which can be used 1 × 1 convolution kernel, this kind setting, which can be avoided, to be destroyed in image to be processed
Noise profile.
It should be noted that in the specific number for the image processing block for including in image noise reduction model, each image processing block
The design parameters such as the convolution nucleus number of the network layer number, each network layer that include, can be by those skilled in the art according to practical need
It asks and is configured, this is not particularly limited in the embodiment of the present disclosure.
Step 102: from first image processing block, successively according to network inter-layer data transmission rule in each image processing block
Noise reduction process is carried out to image to be processed, the target image after obtaining noise reduction.
When in image input picture noise reduction model to be processed, inputted in first image processing block first by first image procossing
Block carries out that treated to it and outputs data to the first network layer of second image processing block, presses in the second image processing block
After the processing of each network inter-layer data transmission rule, by the output of the last one network layer of the second image processing block through second
The maximum pond layer or up-sampling layer being arranged between image processing block and third image procossing are transmitted to third image procossing
The first network layer of block, and so on carry out data transmission in each image processing block in image noise reduction model.Except first figure
As following following rule when carrying out data transmission inside each network layer in other image processing blocks outside process block: sequence exists
When preceding network layer after the data processing of input to exporting, each network layer conduct to sequence after the network layer is exported respectively
Input.
Each image processing block is corresponding with the size and number of channels of characteristic pattern, corresponding characteristic pattern size and number of channels
Parallel link between image processing block all the same.When carrying out noise reduction process to image, the image processing block of two parallel links
Between have data transmission, the preceding image processing block that specifically sorts in the image processing block of two parallel links the last one
The output of network layer, each network layer of the posterior image processing block of input sequencing is as a part of input data respectively.
The method of the removal picture noise provided in the embodiment of the present disclosure is not only suitable for the removal to single image noise, again
The removal of noise suitable for video.When noise is removed in video, each video frame images are sequentially input to image
The processing of above-mentioned removal noise is executed in noise reduction model.
Image to be processed is inputted trained packet in advance by the method for the removal picture noise shown in the present exemplary embodiment
In the image noise reduction model of the process block containing multiple images, from first image processing block, successively according to net in each image processing block
Network inter-layer data transmission rule carries out noise reduction process to image to be processed, and the target image after obtaining noise reduction is dropped by the image
Although the process that the image processing block of the preceding preset quantity in model of making an uproar carries out noise reduction process removal noise to image can have figure
As loss in detail, but the image detail data of loss can be transmitted to below by the image processing block of preceding preset quantity by cross-layer
Image processing block in, image detail is made up, image noise reduction and image detail can be combined.
Fig. 2 is a kind of flow chart of method for removing picture noise shown according to an exemplary embodiment, as shown in Figure 2
Removal picture noise method for including the following steps in terminal device.
Step 201: acquiring the image pattern of preset quantity.
Preset quantity and the size of image pattern can be configured according to actual needs by those skilled in the art, this
This is not particularly limited in open embodiment.Such as: 891,900 or 1000 etc. are set by preset quantity.Figure
Decent is no noise added image.Image pattern may be sized to 512 × 512 pixels.The size of image pattern
The extraction of characteristics of image will be influenced, therefore needs the size according to actual demand flexible setting image pattern.
Step 202: carry out each image pattern plus make an uproar respectively processing, obtains noise image sample.
To single image sample carry out plus make an uproar handle when, to image pattern be added mean value be 0, standard deviation is 0.043 to 0.2
Between Gaussian noise.
Step 203: each image pattern is formed into training image pair with corresponding noise image sample.
The number of training image pair is identical as the picture number after addition noise.
Step 204: being trained according to each training image to image noise reduction model.
When according to image to being trained to image noise reduction model, the noise image sample of training image centering is inputted into figure
It is compared as noise reduction model must export the no noise added image pattern with training image centering, then adjusts image drop
The parameter of model of making an uproar restrains image noise reduction model constantly.Repeat to instruct to image noise reduction model by each training image
Practice, until image noise reduction model receives default degree of convergence or frequency of training reaches preset times.
Can be L1+10 × L2 of output and no noise added image pattern difference by Loss constraint when training, wherein L1 is
One norm, L2 are two norms;Batchsize is criticized and is sized to 2, is criticized defeated into image noise reduction model having a size of each training
The training image entered is to number;Training parameter lr, that is, learning rate is set as 0.01, carries out 2000 training altogether.When loss is less than
It can be determined that training process has restrained when 0.015 or so.Different batches of image patterns can have any different in the training process, once in a while
There is biggish loss, shows some noises of physical presence in no noise added image pattern.It can also for the setting of loss
To increase the constraint of TV norm, the clarity of picture after output denoising can also be increased only with L1 norm.
About the selection of batchsize, if can not only aggravate the calculating space of GPU using excessive batchsize such as 32
Demand, and train the result of the obtained smaller batchsize of result poor.This is because can be added using lesser batchsize
The randomness in direction, can prevent local convergence to a certain extent when big gradient decline, therefore will in the embodiment of the present disclosure
Batchsize is criticized and is sized to 2.Above-mentioned cited 0.01 is not limited to for lr parameter, those skilled in the art
It can the convergence effect of image noise reduction model be adaptively adjusted according to actual needs and in training process.
Step 205: image to be processed is inputted in trained image noise reduction model in advance.
Wherein, image noise reduction model includes multiple images process block, other image procossings in addition to first image processing block
Include multiple network layers in block, maximum pond layer or up-sampling layer are provided between adjacent two image processing block.At each image
Reason block is corresponding with the size and number of channels of characteristic pattern, can be by those skilled in the art according to practical need during specific implementation
It asks and is configured.
For image noise reduction model in the embodiment of the present disclosure referring to fig. 3, to the removal picture noise of the disclosure
Method be illustrated.It should be noted that at the specific number for the image processing block for including in image noise reduction model, each image
The design parameters such as the convolution nucleus number of the network layer number, each network layer that include in reason block, it is not limited to one cited by Fig. 3
Kind set-up mode, can be configured according to actual needs by those skilled in the art, not do and have to this in the embodiment of the present disclosure
Body limitation.
As shown in figure 3, image noise reduction model in the embodiment of the present disclosure includes the first image processing block being sequentially arranged, the
Two image processing blocks, third image processing block, the 4th image processing block, the 5th image processing block, the 6th image processing block, the 7th
Image processing block, the 8th image processing block, the 9th image processing block and the tenth image processing block, wherein at this ten images
Reason block arrays from left to right in Fig. 3.First image processing block is first image processing block.Except head in each image processing block
It include six network layers outside a image processing block.
Last layer network in second image processing block has data with each network layer foundation in the tenth image processing block respectively
Connection;Last layer network in third image processing block has data company with each network layer foundation in the 9th image processing block respectively
It connects;Last layer network in 4th image processing block has data connection with each network layer foundation in the 8th image processing block respectively;
Last layer network in 5th image processing block has data connection with each network layer foundation in the 7th image processing block respectively.
Wherein, each image processing block is corresponding with the size and number of channels of characteristic pattern, corresponding characteristic pattern size and logical
Parallel link between quantity image processing block all the same is crossed, the size of characteristic pattern can be indicated with W × H, channel channel table
Show.Referring to image noise reduction model shown in Fig. 3, the parallel link between image processing block is mainly reflected under each image processing block
Side's connection with the arrow.By taking the parallel link between the second image processing block and the tenth image processing block as an example, from second in Fig. 3
Only extend one article of arrow curve in 6th network layer of image processing block to the tenth image processing block, the second image processing block
It needs to extend six articles of arrows in 6th network layer, this six articles of arrow curves are connected respectively to six networks of the tenth image processing block
Layer is inputted as data.Other several network layers across image processing block connect also similarly.The image that the embodiment of the present disclosure provides
Noise reduction model can be described as Dense-DenseNet i.e. DDN model.
In image noise reduction model shown in Fig. 3, the second image processing block, third image processing block, the 4th image procossing
Block, the 5th image processing block are respectively arranged with maximum pond layer between adjacent two process block of the 6th image processing block;6th image
Processing module, the 7th image processing block, the 8th image processing block, the 9th image processing block and the tenth image processing block adjacent two
Up-sampling layer is respectively arranged between process block, this kind of set-up mode can realize different resolution between different images process block
Characteristic pattern.
Ledgement connection in Fig. 3 in each image processing block between vertical line indicates front net in the same image processing block
Input of the output of network layers as network layer below.First image processing block in image noise reduction model can be by 3 channels of input
Image is mapped as 16 channel images, therefore first image processing block only includes a network layer, which can be used 1 × 1
Convolution kernel, this kind setting can be avoided the noise profile destroyed in image to be processed, remove first image in the image noise reduction model
Other image processing modules outside process block include six network layers, and first three of the image processing block comprising six network layers is a
The convolution kernel size of network layer is 5 × 5, and the convolution kernel size of rear three network layers is 1 × 1, this is because if network layer is all provided with
It is set to 5 × 5 convolution kernel, the coloured silk that obtained denoising result has " pressing " is made an uproar residual.
Network layer in each image processing block is formed by Conv layers, Relu layers and Hardtanh layers respectively.Wherein,
Hardtanh layers of threshold value is selected as (0,4), and not set BN (i.e. BatchNorm layers) is although layer is due to BN layers of energy in network layer
Enough accelerate the convergence rate of network, but original image color space structure can be destroyed after being added BN layers, leads to the figure of output
As will appear serious colour cast in different Epoch.
Step 206: from first image processing block, successively according to network inter-layer data transmission rule in each image processing block
Noise reduction process is carried out to image to be processed, the target image after obtaining noise reduction.
When in image input picture noise reduction model to be processed, inputted in first image processing block first by first image procossing
Block carries out that treated to it and outputs data to the first network layer of second image processing block, presses in the second image processing block
After the processing of each network inter-layer data transmission rule, by the output of the last one network layer of the second image processing block through second
The maximum pond layer being arranged between image processing block and third image procossing is transmitted to the first net of third image processing block
Network layers, and so on carry out data transmission in each image processing block in image noise reduction model.In addition to first image processing block
Other image processing blocks in each network layer inside follow following rule when carrying out data transmission: sort preceding network layer
When to exporting after the data processing of input, each network layer to sequence after the network layer is exported respectively as input.
Each image processing block is corresponding with the size and number of channels of characteristic pattern, corresponding characteristic pattern size and passes through quantity
Parallel link between image processing block all the same.When carrying out noise reduction process to image, the image processing block of two parallel links
Between have data transmission, the preceding image processing block that specifically sorts in the image processing block of two parallel links the last one
The output of network layer, each network layer of the posterior image processing block of input sequencing is as a part of input data respectively.
Referring to shown in Fig. 3, image to be processed inputs the first image processing block, through treated the output of the first image processing block
Input of the data as first network layer in the second image processing block, the first network layer of the second image processing block are handled,
The process flow of six layer networks in the second image processing block are as follows: first network layer will be defeated respectively after the data processing received
Enter to second, third, fourth, fifth and the 6th network layer, the second network layer will input respectively after the data processing received
The network layer of third, the four, the 5th and the 6th, third network layer will input the four, the 5th after the data processing received respectively
And the 6th network layer, the 4th network layer will input the 5th and the 6th network layer respectively after the data processing received, the 5th
Network layer will input the 6th network layer after the data processing received, the 6th network layer will be defeated respectively after the data processing received
Enter in six network layers of the tenth image processing block and inputted as a part of data, while the 6th network of the second image processing block
Layer will also be input to the maximum pond layer between third image processing block after the data processing received, pass through the maximum pond layer
The first network layer of third image processing block is sent data to after processing.Follow above-mentioned similar network inter-layer data transmission rule
Noise reduction process is successively then carried out to image to be processed using each image processing block and carries out cross-layer non-adjacent image processing block
Processing, is finally completed the removal of picture noise, obtains target image.
Image to be processed is inputted trained packet in advance by the method for the removal picture noise shown in the present exemplary embodiment
In the image noise reduction model of the process block containing multiple images, from first image processing block, successively according to net in each image processing block
Network inter-layer data transmission rule carries out noise reduction process to image to be processed, and the target image after obtaining noise reduction is dropped by the image
Although the process that the image processing block of the preceding preset quantity in model of making an uproar carries out noise reduction process removal noise to image can have figure
As loss in detail, but the image detail data of loss can be transmitted to below by the image processing block of preceding preset quantity by cross-layer
Image processing block in, image detail is made up, image noise reduction and image detail can be combined.
Fig. 4 is a kind of block diagram of device for removing picture noise shown according to an exemplary embodiment, referring to Fig. 4 dress
Set includes: input module 401 and noise reduction process module 402.
Input module 401 is configured as inputting image to be processed in trained image noise reduction model in advance, wherein
Described image noise reduction model includes multiple images process block, includes in other image processing blocks in addition to first image processing block
Multiple network layers are provided with maximum pond layer or up-sampling layer between adjacent two image processing block;
Noise reduction process module 402 is configured as from the first image processing block, is successively handled according to each described image
Network inter-layer data transmission rule carries out noise reduction process to the image to be processed in block, the target image after obtaining noise reduction.
Optionally, described image noise reduction model includes the first image processing block being sequentially arranged, the second image processing block, the
Three image processing blocks, the 4th image processing block, the 5th image processing block, the 6th image processing block, the 7th image processing block, the 8th
Image processing block, the 9th image processing block and the tenth image processing block, wherein the first image process block is described first
Image processing block;Last layer network in second image processing block respectively with each network layer in the tenth image processing block
Foundation has data connection;Last layer network in the third image processing block respectively with each net in the 9th image processing block
Network layers foundation has data connection;Last layer network in 4th image processing block is respectively and in the 8th image processing block
Each network layer foundation has data connection;Last layer network in 5th image processing block respectively with the 7th image processing block
In each network layer foundation have data connection.
It optionally, include six network layers, each figure in addition to first image processing block in each described image process block
As the convolution kernel size of first three network layer of process block is 5 × 5, the convolution kernel sizes of rear three network layers is 1 × 1.
Optionally, each network layer is formed by Conv layers, Relu layers and Hardtanh layers respectively.
Optionally, described device further include: acquisition module 403 is configured as figure to be processed in the input module 401
As acquiring the image pattern of preset quantity before inputting in trained image noise reduction model in advance;Add module 404 of making an uproar, is matched
It is set to processing that each described image sample is carried out plus made an uproar respectively, obtains noise image sample;Composite module 405, be configured as by
Each described image sample forms training image pair with corresponding noise image sample;Training module 406 is configured as according to each institute
Training image is stated to be trained image noise reduction model.
About the device in above-described embodiment, wherein modules execute the concrete mode of operation in related this method
Embodiment in be described in detail, no detailed explanation will be given here.
Fig. 5 is a kind of block diagram of terminal device shown according to an exemplary embodiment.With terminal in the embodiment of the present disclosure
Equipment be server for be illustrated.Referring to Fig. 5, it further comprises one that terminal device 500, which includes processing component 501,
Or multiple processors, and the memory resource as representated by memory 502, it can be by the execution of processing component 501 for storing
Instruction, such as application program.The application program stored in memory 502 may include that one or more each is right
The module of Ying Yuyi group instruction.In addition, processing component 501 is configured as executing instruction, to execute above-mentioned Fig. 1, shown in Fig. 2
The method for removing picture noise.
Terminal device 500 can also include that a power supply module 503 is configured as executing the power supply pipe of terminal device 500
Reason, a wired or wireless network interface 504 are configured as terminal device 500 being connected to network and an input and output (I/
O) interface 505.Terminal device 500 can be operated based on the operating system for being stored in memory 502, such as Windows
ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM or similar.
Fig. 6 is a kind of block diagram of terminal device 600 shown according to an exemplary embodiment.For example, terminal device 600 can
To be mobile phone, computer, digital broadcasting terminal, messaging device, game console, tablet device, Medical Devices are good for
Body equipment, personal digital assistant etc..
Referring to Fig. 6, terminal device 600 may include following one or more components: processing component 602, memory 604,
Power supply module 606, multimedia component 608, audio component 610, the interface 612 of input/output (I/O), sensor module 614,
And communication component 616.
The integrated operation of the usual controlling terminal equipment 600 of processing component 602, such as with display, call, data are logical
Letter, camera operation and record operate associated operation.Processing component 602 may include one or more processors 620 to hold
Row instruction, to perform all or part of the steps of the methods described above.In addition, processing component 602 may include one or more moulds
Block, convenient for the interaction between processing component 602 and other assemblies.For example, processing component 602 may include multi-media module, with
Facilitate the interaction between multimedia component 608 and processing component 602.
Memory 604 is configured as storing various types of data to support the operation in terminal device 600.These data
Example include any application or method for being operated on terminal device 600 instruction, contact data, telephone directory
Data, message, picture, video etc..Memory 604 can by any kind of volatibility or non-volatile memory device or it
Combination realize, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM) is erasable
Except programmable read only memory (EPROM), programmable read only memory (PROM), read-only memory (ROM), magnetic memory, fastly
Flash memory, disk or CD.
Power supply module 606 provides electric power for the various assemblies of terminal device 600.Power supply module 606 may include power supply pipe
Reason system, one or more power supplys and other with for terminal device 600 generate, manage, and distribute the associated component of electric power.
Multimedia component 608 includes the screen of one output interface of offer between terminal device 600 and user.One
In a little embodiments, screen may include liquid crystal display (LCD) and touch panel (TP).If screen includes touch panel, screen
Curtain may be implemented as touch screen, to receive input signal from the user.Touch panel includes one or more touch sensings
Device is to sense the gesture on touch, slide, and touch panel.The touch sensor can not only sense touch or sliding action
Boundary, but also detect duration and pressure associated with the touch or slide operation.In some embodiments, more matchmakers
Body component 608 includes a front camera and/or rear camera.When terminal device 600 is in operation mode, as shot mould
When formula or video mode, front camera and/or rear camera can receive external multi-medium data.Each preposition camera shooting
Head and rear camera can be a fixed optical lens system or have focusing and optical zoom capabilities.
Audio component 610 is configured as output and/or input audio signal.For example, audio component 610 includes a Mike
Wind (MIC), when terminal device 600 is in operation mode, when such as call mode, recording mode, and voice recognition mode, microphone
It is configured as receiving external audio signal.The received audio signal can be further stored in memory 604 or via logical
Believe that component 616 is sent.In some embodiments, audio component 610 further includes a loudspeaker, is used for output audio signal.
I/O interface 612 provides interface between processing component 602 and peripheral interface module, and above-mentioned peripheral interface module can
To be keyboard, click wheel, button etc..These buttons may include, but are not limited to: home button, volume button, start button and lock
Determine button.
Sensor module 614 includes one or more sensors, for providing the state of various aspects for terminal device 600
Assessment.For example, sensor module 614 can detecte the state that opens/closes of terminal device 600, the relative positioning of component, example
As the component be terminal device 600 display and keypad, sensor module 614 can also detect terminal device 600 or
The position change of 600 1 components of terminal device, the existence or non-existence that user contacts with terminal device 600, terminal device 600
The temperature change of orientation or acceleration/deceleration and terminal device 600.Sensor module 614 may include proximity sensor, be configured
For detecting the presence of nearby objects without any physical contact.Sensor module 614 can also include optical sensor,
Such as CMOS or ccd image sensor, for being used in imaging applications.In some embodiments, which may be used also
To include acceleration transducer, gyro sensor, Magnetic Sensor, pressure sensor or temperature sensor.
Communication component 616 is configured to facilitate the communication of wired or wireless way between terminal device 600 and other equipment.
Terminal device 600 can access the wireless network based on communication standard, such as WiFi, 2G or 3G or their combination.Show at one
In example property embodiment, communication component 616 receives broadcast singal or broadcast from external broadcasting management system via broadcast channel
Relevant information.In one exemplary embodiment, the communication component 616 further includes near-field communication (NFC) module, short to promote
Cheng Tongxin.For example, radio frequency identification (RFID) technology, Infrared Data Association (IrDA) technology, ultra wide band can be based in NFC module
(UWB) technology, bluetooth (BT) technology and other technologies are realized.
In the exemplary embodiment, terminal device 600 can be by one or more application specific integrated circuit (ASIC), number
Word signal processor (DSP), digital signal processing appts (DSPD), programmable logic device (PLD), field programmable gate array
(FPGA), controller, microcontroller, microprocessor or other electronic components are realized, for executing shown in above-mentioned Fig. 1 to Fig. 2
Removal picture noise method.
In the exemplary embodiment, a kind of non-transitorycomputer readable storage medium including instruction, example are additionally provided
It such as include the memory 604 of instruction, above-metioned instruction can be executed by the processor 620 of terminal device 600 to complete above-mentioned Fig. 1 to figure
The method of picture noise is removed shown in 2.For example, the non-transitorycomputer readable storage medium can be ROM, random
Access memory (RAM), CD-ROM, tape, floppy disk and optical data storage devices etc..
In the exemplary embodiment, a kind of computer program product is additionally provided, when in the computer program product
When instruction is executed by the processor 620 of terminal device 600, so that terminal device 600 is completed above-mentioned Fig. 1 and is gone to shown in Fig. 2
Except the method for picture noise.
Those skilled in the art after considering the specification and implementing the invention disclosed here, will readily occur to its of the disclosure
Its embodiment.This application is intended to cover any variations, uses, or adaptations of the disclosure, these modifications, purposes or
Person's adaptive change follows the general principles of this disclosure and including the undocumented common knowledge in the art of the disclosure
Or conventional techniques.The description and examples are only to be considered as illustrative, and the true scope and spirit of the disclosure are by following
Claim is pointed out.
It should be understood that the present invention is not limited to the precise structure already described above and shown in the accompanying drawings, and
And various modifications and changes may be made without departing from the scope thereof.The scope of the present invention is limited only by the attached claims.
Claims (10)
1. a kind of method for removing picture noise, which is characterized in that the described method includes:
Image to be processed is inputted in trained image noise reduction model in advance, wherein described image noise reduction model includes multiple
Image processing block, includes multiple network layers in other image processing blocks in addition to first image processing block, at adjacent two image
Maximum pond layer or up-sampling layer are provided between reason block;
From the first image processing block, successively according to network inter-layer data transmission rule in each described image process block to institute
It states image to be processed and carries out noise reduction process, the target image after obtaining noise reduction.
2. the method according to claim 1, wherein described image noise reduction model includes the first figure being sequentially arranged
As process block, the second image processing block, third image processing block, the 4th image processing block, the 5th image processing block, the 6th image
Process block, the 7th image processing block, the 8th image processing block, the 9th image processing block and the tenth image processing block, wherein institute
Stating the first image processing block is the first image processing block;
Last layer network in second image processing block has with each network layer foundation in the tenth image processing block respectively
Data connection;
Last layer network in the third image processing block has with each network layer foundation in the 9th image processing block respectively
Data connection;
Last layer network in 4th image processing block has with each network layer foundation in the 8th image processing block respectively
Data connection;
Last layer network in 5th image processing block has with each network layer foundation in the 7th image processing block respectively
Data connection.
3. the method according to claim 1, wherein in each described image process block in addition to first image processing block
It include six network layers, the convolution kernel size of first three network layer of each described image process block is 5 × 5, rear three networks
The convolution kernel size of layer is 1 × 1.
4. the method according to claim 1, wherein each network layer respectively by Conv layers, Relu layers and
Hardtanh layers of composition.
5. the method according to claim 1, wherein image to be processed is inputted trained figure in advance described
Before the step in noise reduction model, the method also includes:
Acquire the image pattern of preset quantity;
Carry out each described image sample plus make an uproar respectively processing, obtains noise image sample;
Each described image sample is formed into training image pair with corresponding noise image sample;
Image noise reduction model is trained according to each training image.
6. a kind of device for removing picture noise, which is characterized in that described device includes:
Input module is configured as inputting image to be processed in trained image noise reduction model in advance, wherein described image
Noise reduction model includes multiple images process block, includes multiple networks in other image processing blocks in addition to first image processing block
Layer, maximum pond layer or up-sampling layer are provided between adjacent two image processing block;
Noise reduction process module is configured as from the first image processing block, successively according to net in each described image process block
Network inter-layer data transmission rule carries out noise reduction process to the image to be processed, the target image after obtaining noise reduction.
7. device according to claim 6, which is characterized in that described image noise reduction model includes the first figure being sequentially arranged
As process block, the second image processing block, third image processing block, the 4th image processing block, the 5th image processing block, the 6th image
Process block, the 7th image processing block, the 8th image processing block, the 9th image processing block and the tenth image processing block, wherein institute
Stating the first image processing block is the first image processing block;
Last layer network in second image processing block has with each network layer foundation in the tenth image processing block respectively
Data connection;
Last layer network in the third image processing block has with each network layer foundation in the 9th image processing block respectively
Data connection;
Last layer network in 4th image processing block has with each network layer foundation in the 8th image processing block respectively
Data connection;
Last layer network in 5th image processing block has with each network layer foundation in the 7th image processing block respectively
Data connection.
8. device according to claim 6, which is characterized in that in each described image process block in addition to first image processing block
It include six network layers, the convolution kernel size of first three network layer of each described image process block is 5 × 5, rear three networks
The convolution kernel size of layer is 1 × 1.
9. a kind of terminal device characterized by comprising
Processor;
Memory for storage processor executable instruction;
Wherein, the processor is configured to the method for executing removal picture noise as described in any one in claim 1-5.
10. a kind of non-transitorycomputer readable storage medium, when the instruction in the storage medium is by the processing of terminal device
When device executes, so that the method that terminal device is able to carry out removal picture noise as described in any one in claim 1-5.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811290243.8A CN109615589A (en) | 2018-10-31 | 2018-10-31 | Remove the method, apparatus and terminal device of picture noise |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811290243.8A CN109615589A (en) | 2018-10-31 | 2018-10-31 | Remove the method, apparatus and terminal device of picture noise |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109615589A true CN109615589A (en) | 2019-04-12 |
Family
ID=66003033
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811290243.8A Pending CN109615589A (en) | 2018-10-31 | 2018-10-31 | Remove the method, apparatus and terminal device of picture noise |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109615589A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111598807A (en) * | 2020-05-16 | 2020-08-28 | 武汉青铜科技信息咨询有限公司 | Automobile part detection data sharing system and method based on block chain |
CN112446369A (en) * | 2020-11-05 | 2021-03-05 | 深圳中兴网信科技有限公司 | Bill processing method, system and storage medium |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20180060649A1 (en) * | 2016-08-30 | 2018-03-01 | Irida Labs S.A. | Fast, embedded, hybrid video face recognition system |
CN108717680A (en) * | 2018-03-22 | 2018-10-30 | 北京交通大学 | Spatial domain picture steganalysis method based on complete dense connection network |
-
2018
- 2018-10-31 CN CN201811290243.8A patent/CN109615589A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20180060649A1 (en) * | 2016-08-30 | 2018-03-01 | Irida Labs S.A. | Fast, embedded, hybrid video face recognition system |
CN108717680A (en) * | 2018-03-22 | 2018-10-30 | 北京交通大学 | Spatial domain picture steganalysis method based on complete dense connection network |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111598807A (en) * | 2020-05-16 | 2020-08-28 | 武汉青铜科技信息咨询有限公司 | Automobile part detection data sharing system and method based on block chain |
CN112446369A (en) * | 2020-11-05 | 2021-03-05 | 深圳中兴网信科技有限公司 | Bill processing method, system and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106339680B (en) | Face key independent positioning method and device | |
CN106558025B (en) | Picture processing method and device | |
CN107590461B (en) | Face recognition method and related product | |
CN110706179B (en) | Image processing method and electronic equipment | |
CN106548468B (en) | The method of discrimination and device of image definition | |
CN108510987A (en) | Method of speech processing and device | |
CN106937039A (en) | A kind of imaging method based on dual camera, mobile terminal and storage medium | |
CN105809704A (en) | Method and device for identifying image definition | |
CN104243814B (en) | Analysis method, image taking reminding method and the device of objects in images layout | |
CN106778531A (en) | Face detection method and device | |
CN106682736A (en) | Image identification method and apparatus | |
WO2019011098A1 (en) | Unlocking control method and relevant product | |
CN105957037B (en) | Image enchancing method and device | |
CN105959554A (en) | Video shooting apparatus and method | |
CN105354793A (en) | Facial image processing method and device | |
CN109934275A (en) | Image processing method and device, electronic equipment and storage medium | |
CN106101307A (en) | Mobile terminal | |
CN109672830A (en) | Image processing method, device, electronic equipment and storage medium | |
CN105654092B (en) | Feature extracting method and device | |
CN107463903A (en) | Face key independent positioning method and device | |
CN105513021A (en) | Image noise reduction device and method | |
CN104867112B (en) | Photo processing method and device | |
CN108108671A (en) | Description of product information acquisition method and device | |
CN107341777A (en) | image processing method and device | |
CN110931028B (en) | Voice processing method and device and electronic equipment |
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: 20190412 |