CN108492261A - A kind of image enchancing method and computing device - Google Patents
A kind of image enchancing method and computing device Download PDFInfo
- Publication number
- CN108492261A CN108492261A CN201810122975.XA CN201810122975A CN108492261A CN 108492261 A CN108492261 A CN 108492261A CN 201810122975 A CN201810122975 A CN 201810122975A CN 108492261 A CN108492261 A CN 108492261A
- Authority
- CN
- China
- Prior art keywords
- image
- core
- frequency response
- low frequency
- kernel
- 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 103
- 230000004044 response Effects 0.000 claims abstract description 81
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 75
- 230000002708 enhancing effect Effects 0.000 claims abstract description 56
- 238000012795 verification Methods 0.000 claims abstract description 12
- 230000009466 transformation Effects 0.000 claims description 43
- 230000008569 process Effects 0.000 claims description 14
- 238000012545 processing Methods 0.000 description 28
- 238000004891 communication Methods 0.000 description 15
- 238000010586 diagram Methods 0.000 description 13
- 230000033001 locomotion Effects 0.000 description 6
- 230000006870 function Effects 0.000 description 5
- 230000004913 activation Effects 0.000 description 4
- 230000014509 gene expression Effects 0.000 description 4
- 230000008901 benefit Effects 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 235000013399 edible fruits Nutrition 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000005284 excitation Effects 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 230000002093 peripheral effect Effects 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
- 238000004458 analytical method Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000000151 deposition Methods 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001151 other effect Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000003672 processing method Methods 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
- G06T5/77—Retouching; Inpainting; Scratch removal
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20048—Transform domain processing
- G06T2207/20052—Discrete cosine transform [DCT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20172—Image enhancement details
- G06T2207/20192—Edge enhancement; Edge preservation
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (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)
Abstract
The invention discloses a kind of image enchancing method, this method is suitable for executing in computing device, including step:It is decomposed according to the image of the predetermined decomposition verification input of dct transform, obtains the low frequency response and high frequency response of described image;By low frequency response input the first enhancing network of image, enhanced low frequency response is obtained;By high frequency response input the second enhancing network of image, enhanced high frequency response is obtained;And according to enhanced low frequency response and the enhanced high frequency response reconstruction image, obtain enhanced image.The present invention discloses the computing device for executing the above method together.
Description
Technical field
The present invention relates to technical field of image processing, especially a kind of image enchancing method and computing device.
Background technology
Image enhancement technique is a kind of image processing techniques of entirety or local characteristics that purposefully emphasizing image.Pass through
Original unsharp image is apparent from or is emphasized certain interested features, different objects in enlarged image by image enhancement
Difference between feature inhibits uninterested feature, to improve picture quality, abundant information amount, reinforces image interpretation and knowledge
Other effect, meets the needs of certain special analysis.
Image enhancement traditional algorithm can be divided into two major classes, frequency domain Enhancement Method and spatial domain Enhancement Method.The former will
Image regards a kind of 2D signal as, and the signal enhancing based on two-dimensional Fourier transform is carried out to it.The latter is using various
Filter enhances image.At the same time, using the methods of histogram, HDR, PCA, Laplacian, Wavelet to figure
The algorithm that brightness, details, color of picture etc. are enhanced also is widely used.In short, traditional algorithm for image enhancement
It is a kind of relatively general algorithm, advantage is that have stronger versatility, the disadvantage is that sometimes for relevant parameter is manually adjusted, ability
Obtain relatively good enhancing effect.
In consideration of it, needing a kind of effective image enhancement schemes of energy, ideal enhancing effect can be obtained.
Invention content
For this purpose, the present invention provides a kind of image enchancing method and computing device, on trying hard to solve or at least alleviate
At least one problem existing for face.
According to an aspect of the invention, there is provided a kind of image enchancing method, this method is suitable for holding in computing device
Row, including step:It is decomposed according to the image of the predetermined decomposition verification input of dct transform, the low frequency for obtaining described image is rung
Should and high frequency response;By low frequency response input the first enhancing network of image, enhanced low frequency response is obtained;By the height of image
Frequency response should input the second enhancing network, obtain enhanced high frequency response;And according to enhanced low frequency response and the increasing
High frequency response reconstruction image after strong, obtains enhanced image.
Optionally, further include the step for the predetermined decomposition core for generating dct transform in image enchancing method according to the present invention
Suddenly:Calculate the transformation kernel of two-dimensional dct transform;Coded treatment is carried out to transformation kernel, to generate new transformation kernel;And according to new transformation
The predetermined decomposition core of karyogenesis dct transform.
Optionally, in image enchancing method according to the present invention, coded treatment is carried out to generate new transformation to transformation kernel
The step of core includes:The sequence scanned according to zigzag resequences to transformation kernel, obtains new transformation kernel.
Optionally, in image enchancing method according to the present invention, according to the predetermined decomposition of new transformation karyogenesis dct transform
The step of core includes:New transformation kernel is rotated into 180 degree, transformation kernel after being rotated;And to being converted after new transformation kernel and rotation
Core does process of convolution, generates the predetermined decomposition core of dct transform.
Optionally, in image enchancing method according to the present invention, according to the figure of the predetermined decomposition verification input of dct transform
As decomposed, obtained image low frequency response and high frequency response the step of include:The predetermined decomposition core of dct transform is carried out pre-
Processing decomposes core and at least one high-frequency decomposition core to obtain at least one low frequency;The image of verification input is decomposed according to low frequency
It is decomposed, obtains the corresponding at least one low frequency response of image;And divided according to the image of high-frequency decomposition verification input
Solution, obtains the corresponding at least one high frequency response of image.
Optionally, in image enchancing method according to the present invention, the predetermined decomposition core of dct transform is pre-processed,
Include to obtain the step of at least one low frequency decomposes core and at least one high-frequency decomposition core:Multiple predetermined decomposition cores are unequal
Ground is divided into multiple classification successively;Calculate the summation of the predetermined decomposition core under each classification:By predetermined point under first classification
The summation for solving core decomposes core as low frequency;With using the summation of the predetermined decomposition core under other each classification as a high frequency
Decompose core.
Optionally, in image enchancing method according to the present invention, the low frequency response of image is defined as:Wherein, New_DCT_decom_
Kernel (0) indicates that low frequency decomposes core, and I indicates that the image of input, New_Band_Responds (0) indicate low frequency response,Table
Show convolution algorithm.
Optionally, in image enchancing method according to the present invention, the high frequency response of image is defined as:
Wherein, New_
DCT_decom_kernel (w) indicates that w-th of high-frequency decomposition core, I indicate the image of input, New_Band_Responds (w) tables
Show w-th of high frequency response,Indicate convolution algorithm.
Optionally, in image enchancing method according to the present invention, the first enhancing network includes at least:First convolutional layer,
Second convolutional layer, third convolutional layer and the first number residual error module between the second convolutional layer and third convolutional layer.
Optionally, in image enchancing method according to the present invention, include in the first convolutional layer in the first enhancing network
The convolution kernel of 64 5*5, first enhances the convolution kernel for including 64 3*3 in the second convolutional layer in network, the first enhancing network
In third convolutional layer in include the convolution kernel of 3 3*3.
Optionally, in image enchancing method according to the present invention, the second enhancing network includes at least:First convolutional layer,
Second convolutional layer, third convolutional layer and the second number residual error module between the second convolutional layer and third convolutional layer.
Optionally, in image enchancing method according to the present invention, include in the first convolutional layer in the second enhancing network
The convolution kernel of 64 3*3, second enhances the convolution kernel for including 64 3*3 in the second convolutional layer in network, the second enhancing network
In third convolutional layer in include the convolution kernel of 3 3*3.
Optionally, in image enchancing method according to the present invention, residual error module is suitable for that the knot after process of convolution will be passed through
Fruit is added with the input of residual error module, the output as the residual error module.
Optionally, in image enchancing method according to the present invention, residual error module includes:Convolution unit is suitable for input
Characteristic do process of convolution, generate new feature data;And processing unit, suitable for the new spy to being generated by convolution unit
Sign data do normalized.
Optionally, in image enchancing method according to the present invention, according to enhanced low frequency response and enhanced height
The step of frequency responsive reconstruction image includes:The processing of the reconstruction image is carried out as follows:Wherein, Enhanced_Band (0) indicates enhanced low frequency response, Enhanced_
Band (w) (w ∈ [1, M]) indicates that enhanced high frequency response, I' indicate that enhanced image, M indicate low frequency response and high frequency
The total number of response.
Optionally, in image enchancing method according to the present invention, for the dct transform of 8*8 sizes, two-dimensional dct becomes
The transformation kernel kernel (i, j) changed is defined as:
Wherein, i ∈ 0~7, j ∈ 0~7, k ∈ 0~7, l ∈ 0~7, patchsize=8, and
Optionally, in image enchancing method according to the present invention, at least one low frequency is generated as follows and decomposes core
With at least one high-frequency decomposition core:
New_DCT_decom_kernel (0)=DCT_decom_kernel (0)
Wherein, New_DCT_decom_kernel (0) expressions low frequency decomposition core, New_DCT_decom_kernel (1)~
New_DCT_decom_kernel (5) indicates that high-frequency decomposition core, DCT_decom_kernel (u) indicate u-th of predetermined decomposition core.
Optionally, in image enchancing method according to the present invention, the first number is 4.
Optionally, in image enchancing method according to the present invention, the second number is 3.
According to another aspect of the invention, a kind of computing device is provided, including:One or more processors;And storage
Device;One or more programs, wherein one or more of programs are stored in the memory and are configured as by described one
A or multiple processors execute, and one or more of programs include the finger for executing the either method in method as described above
It enables.
In accordance with a further aspect of the present invention, a kind of computer-readable storage medium of the one or more programs of storage is provided
Matter, one or more of programs include instruction, and described instruction is when computing device executes so that the computing device executes such as
Either method in the upper method.
Image enhancement schemes according to the present invention first decompose image according to dct transform, be divided into low frequency part and
Then high frequency section carries out enhancing processing using different enhancing networks to low frequency part and high frequency section, obtains enhanced
Low frequency part lays particular emphasis on the color for enhancing image and brightness, and enhanced high frequency section lays particular emphasis on the details of enhancing image.Cause
This this programme also effectively enhances the detailed information of image while enhancing color of image and luminance information.
Description of the drawings
To the accomplishment of the foregoing and related purposes, certain illustrative sides are described herein in conjunction with following description and drawings
Face, these aspects indicate the various modes that can put into practice principles disclosed herein, and all aspects and its equivalent aspect
It is intended to fall in the range of theme claimed.Read following detailed description in conjunction with the accompanying drawings, the disclosure it is above-mentioned
And other purposes, feature and advantage will be apparent.Throughout the disclosure, identical reference numeral generally refers to identical
Component or element.
Fig. 1 shows the organigram of computing device 100 according to an embodiment of the invention;
Fig. 2 shows the flow charts of image enchancing method 200 according to an embodiment of the invention;
Fig. 3 A and Fig. 3 B respectively illustrate the image schematic diagram of two dct transform core according to the ... of the embodiment of the present invention;
Fig. 4 shows the schematic diagram of lookup_table according to an embodiment of the invention;
Fig. 5 A show the image of input, and Fig. 5 B and Fig. 5 C are respectively illustrated checks Fig. 5 A progress using different predetermined decompositions
Result schematic diagram after decomposition;
Fig. 6 shows the structural schematic diagram of the first enhancing network according to an embodiment of the invention;
Fig. 7 shows the structural schematic diagram of residual error module according to an embodiment of the invention;
Fig. 8 shows the structural schematic diagram of the second enhancing network according to an embodiment of the invention;And
Fig. 9 A show the image of an input, and Fig. 9 B show to execute Fig. 9 A and generate after image enchancing method 200
Enhanced image, Fig. 9 C show the original image of Fig. 9 A.
Specific implementation mode
The exemplary embodiment of the disclosure is more fully described below with reference to accompanying drawings.Although showing the disclosure in attached drawing
Exemplary embodiment, it being understood, however, that may be realized in various forms the disclosure without should be by embodiments set forth here
It is limited.On the contrary, these embodiments are provided to facilitate a more thoroughly understanding of the present invention, and can be by the scope of the present disclosure
Completely it is communicated to those skilled in the art.
Fig. 1 is the block diagram of Example Computing Device 100.In basic configuration 102, computing device 100, which typically comprises, is
System memory 106 and one or more processor 104.Memory bus 108 can be used for storing in processor 104 and system
Communication between device 106.
Depending on desired configuration, processor 104 can be any kind of processor, including but not limited to:Microprocessor
Device (μ P), microcontroller (μ C), digital information processor (DSP) or any combination of them.Processor 104 may include all
Cache, processor core such as one or more rank of on-chip cache 110 and second level cache 112 etc
114 and register 116.Exemplary processor core 114 may include arithmetic and logical unit (ALU), floating-point unit (FPU),
Digital signal processing core (DSP core) or any combination of them.Exemplary Memory Controller 118 can be with processor
104 are used together, or in some implementations, and Memory Controller 118 can be an interior section of processor 104.
Depending on desired configuration, system storage 106 can be any type of memory, including but not limited to:Easily
The property lost memory (RAM), nonvolatile memory (ROM, flash memory etc.) or any combination of them.System stores
Device 106 may include operating system 120, one or more apply 122 and program data 124.In some embodiments,
It may be arranged to be operated using program data 124 on an operating system using 122.In some embodiments, computing device
100 are configured as executing image enchancing method 200, and the finger for executing the method 200 is just contained in program data 124
It enables.
Computing device 100 can also include contributing to from various interface equipments (for example, output equipment 142, Peripheral Interface
144 and communication equipment 146) to basic configuration 102 via the communication of bus/interface controller 130 interface bus 140.Example
Output equipment 142 include graphics processing unit 148 and audio treatment unit 150.They can be configured as contribute to via
One or more port A/V 152 is communicated with the various external equipments of such as display or loud speaker etc.Outside example
If interface 144 may include serial interface controller 154 and parallel interface controller 156, they, which can be configured as, contributes to
Via one or more port I/O 158 and such as input equipment (for example, keyboard, mouse, pen, voice-input device, image
Input equipment) or the external equipment of other peripheral hardwares (such as printer, scanner etc.) etc communicated.Exemplary communication is set
Standby 146 may include network controller 160, can be arranged to convenient for via one or more communication port 164 and one
The communication that other a or multiple computing devices 162 pass through network communication link.
Network communication link can be an example of communication media.Communication media can be usually presented as in such as carrier wave
Or the computer-readable instruction in the modulated data signal of other transmission mechanisms etc, data structure, program module, and can
To include any information delivery media." modulated data signal " can be such signal, one in its data set or
It is multiple or it change can the mode of coding information in the signal carry out.As unrestricted example, communication media can
To include the wire medium of such as cable network or private line network etc, and it is such as sound, radio frequency (RF), microwave, infrared
(IR) the various wireless mediums or including other wireless mediums.Term computer-readable medium used herein may include depositing
Both storage media and communication media.
Computing device 100 can be implemented as including desktop computer and the personal computer of notebook computer configuration.When
So, computing device 100 can be implemented as a part for portable (or mobile) electronic equipment of small size, these electronic equipments can be with
It is that such as cellular phone, digital camera, personal digital assistant (PDA), personal media player device, wireless network browsing are set
Standby, personal helmet, application specific equipment or may include any of the above function mixing apparatus.
Below with reference to Fig. 2, the realization stream of image enchancing method 200 according to an embodiment of the invention is elaborated
Journey.
Method 200 starts from step S210, is decomposed, is obtained according to the image of the predetermined decomposition verification input of dct transform
The low frequency response and high frequency response of the image.
According to a kind of realization method, before executing step S210, method 200 further includes generate dct transform predetermined point
The step of solving core.Including following three step.
The first step calculates the transformation kernel of two-dimensional dct transform.
Dct transform refers to discrete cosine transform, be with a kind of relevant transformation of Fourier transformation, be similar to discrete fourier
Transformation, is mainly used for the compression of image or data, spatial domain signal can be transformed on frequency-region signal, has good decorrelation
The performance of property.For example, still image coding standard JPEG, motion picture encoding standard MJPEG and MPEG each standard in
Two-dimensional dct transform is used.
A kind of embodiment according to the present invention shares 64 transformation kernels, Mei Gebian by taking the dct transform of 8*8 sizes as an example
The size for changing core is 8*8, calculates the transformation kernel kernel (i, j) of its two-dimensional dct transform according to the following formula:
Wherein, i ∈ 0~7, j ∈ 0~7, k ∈ 0~7, l ∈ 0~7, patchsize=8, and
Such as Fig. 3 A and Fig. 3 B, respectively illustrate two dct transform core kernel (Isosorbide-5-Nitrae) according to the ... of the embodiment of the present invention and
Image (after being normalized) schematic diagram of kernel (2,2).
Second step carries out coded treatment, to generate new transformation kernel to transformation kernel.According to the suitable of zigzag (zigzag) scanning
Ordered pair transformation kernel is resequenced, and new transformation kernel is obtained.
According to a kind of realization method, new transformation kernel new_kernel (u) generates as follows:
New_kernel (u)=kernel (i, j)
Wherein, kernel (i, j) is transformation kernel, u=lookup_table (i, j), u ∈ 0~63.
As Fig. 4 shows the schematic diagram of numerical value in lookup_table.The longitudinal axis represents the value of coordinate i, and horizontal axis represents coordinate j
Value, sequentially connect lookup_table in numerical value 0,1,2,3,4 ..., just obtain zigzag scanning scanning sequency, wherein
Lookup_table (0,0)=0, lookup_table (0,1)=1, lookup_table (1,0)=2.
Third walks, according to the predetermined decomposition core of new transformation karyogenesis dct transform.According to one embodiment of present invention, raw
Include following two step at the step of predetermined decomposition core:
1. new transformation kernel new_kernel (u) is rotated 180 degree, new transformation kernel rotate_new_ after being rotated
kernel(u);
2. being done at convolution to transformation kernel rotate_new_kernel (u) after new transformation kernel new_kernel (u) and rotation
Reason, generates the predetermined decomposition core DCT_decom_kernel (u) of dct transform.
Optionally, the calculation formula of predetermined decomposition core DCT_decom_kernel (u) indicates as follows:
Wherein, symbolIndicate that convolution, the convolution algorithm between two matrixes belong to the basic common sense of this field, herein not
Make excessive explain.Ibid, u ∈ 0~63.
After generating the predetermined decomposition core of dct transform by above-mentioned three step, according to a kind of realization method, by the figure of input
As being decomposed according to predetermined decomposition core, the frequency response under different DCT predetermined decompositions cores is just obtained, wherein different DCT are predetermined
The frequency response Band_Responds (u) decomposed under core is obtained by the way that input picture is done convolution algorithm with predetermined decomposition core, such as
Described in following formula:
Wherein, I indicates that the image of input, DCT_decom_kernel (u) indicate u-th of predetermined decomposition core, Band_
Responds (u) indicates the frequency response under u-th of predetermined decomposition core, and the value range of u is [0,64].
By taking Lena figures as an example, if Fig. 5 A show the image of input, Fig. 5 B are shown to be checked using first predetermined decomposition
Fig. 5 A decomposed after result schematic diagram, Fig. 5 C are shown checks the knot after Fig. 5 A are decomposed using second predetermined decomposition
Fruit schematic diagram.U is bigger, and corresponding frequency response is more focused on the details (that is, high frequency section) of image.
So according to the embodiment of the present invention, predetermined decomposition core is reconfigured, utilizes what is reconfigured
The picture breakdown of input is low frequency and high frequency by predetermined decomposition core.Wherein, low frequency part, which is laid particular emphasis on, indicates the color of image and bright
Degree, high frequency section lay particular emphasis on the details for indicating image.Specifically, step S210 is executed as follows.
First, the predetermined decomposition core of dct transform is pre-processed, core and at least one is decomposed to obtain at least one low frequency
A high-frequency decomposition core.According to one embodiment of present invention, by multiple predetermined decomposition cores of generation (as above with 8*8 sizes
For dct transform, predetermined decomposition core is 64) it is unequally divided into multiple classification successively, then calculate pre- under each classification
Surely the summation of core is decomposed, specifically, the summation of the predetermined decomposition core under first classification is decomposed into core as low frequency, it will be other every
The summation of predetermined decomposition core under a classification is respectively as a high-frequency decomposition core.
For example, generating a low frequency as follows decomposes core and five high-frequency decomposition cores:
New_DCT_decom_kernel (0)=DCT_decom_kernel (0)
In above formula, New_DCT_decom_kernel (0) indicates that the low frequency generated decomposes core, New_DCT_decom_
Kernel (1)~New_DCT_decom_kernel (5) indicates the five high-frequency decomposition cores generated, DCT_decom_kernel
(u) u-th of predetermined decomposition core is indicated, similarly, the value of u is u ∈ 0~63.
It is of course also possible to be classified to the predetermined decomposition core of dct transform according to actual conditions, combined, at least one is obtained
A low frequency decomposes core and at least one high-frequency decomposition core.Only provide herein it is according to an embodiment of the invention classification, combination side
Formula, the embodiment of the present invention are without limitation.
Secondly, the image for verification input being decomposed according to the low frequency generated is decomposed, and it is corresponding at least to obtain the image
One low frequency response.Certainly, it when there is multiple low frequencies to decompose core, decomposes core according to each low frequency and the image of input is carried out respectively
Processing obtains each low frequency and decomposes the corresponding low frequency response of core.
According to a kind of realization method, the low frequency response of image is defined as:
Wherein, as it was noted above, New_DCT_decom_kernel (0) indicates that low frequency decomposes core, I indicates the figure of input
Picture, New_Band_Responds (0) indicate the low frequency response of image,Indicate convolution algorithm.
Finally, it is decomposed according to the image of the high-frequency decomposition verification input generated, it is corresponding at least to obtain the image
One high frequency response.When there are multiple high-frequency decomposition cores, the image of input is handled respectively according to each high-frequency decomposition core,
Obtain the corresponding high frequency response of each high-frequency decomposition core.
According to a kind of realization method, the high frequency response of image is defined as:
Wherein, New_DCT_decom_kernel (w) indicates that w-th of high-frequency decomposition core, I indicate the image of input, New_
Band_Responds (w) indicates w-th of high frequency response,Indicate convolution algorithm.As it was noted above, the value range of w be [1,
5]。
Then in step S220, by the low frequency response of image input the first enhancing network, to the low frequency part of image into
Row enhancing, to obtain enhanced low frequency response.That is, enhancing color and the brightness portion of image by the first enhancing network
Point.
Realization method according to the present invention, the first enhancing network include at least:First convolutional layer, the second convolutional layer, third
Convolutional layer and the first number residual error module between the second convolutional layer and third convolutional layer.As Fig. 6 is shown according to this
Invention one embodiment first enhancing network 600 structural schematic diagram, first enhancing network 600 include the first convolutional layer 610,
Second convolutional layer 620, third convolutional layer 630 and four residual error moulds between the second convolutional layer 620 and third convolutional layer 630
Block 640.
According to one embodiment of present invention, the convolution kernel of 64 5*5, convolution kernel movement are included in the first convolutional layer 610
Step-length (Stride) be set as 1, meanwhile, processing is filled to the boundary of image, wherein the parameter for filling image boundary
Padding is set as 2, and (expression will all expand Padding=2 2 pixels at 4 edges up and down, i.e. width and height all expands
4 pixels).Similarly, the convolution kernel of 64 3*3, the step-length (Stride) of convolution kernel movement are included in the second convolutional layer 620
It is set as 1, the parameter Padding for filling image boundary is set as 1, and (Padding=1 expressions will all expand at 4 edges up and down
1 pixel is filled, i.e. width and height has all expanded 2 pixels).Include the convolution kernel of 3 3*3, convolution in third convolutional layer 630
The step-length (Stride) of core movement is set as 1, and the parameter Padding for filling image boundary is set as 1.Certainly, in each convolutional layer
In addition to process of convolution, it also may include pondization processing (such as maxpooling), activation processing (such as ReLu), but not limited to this.
Residual error module 640 will be added by the result after process of convolution with the input of the residual error module, as the residual error mould
The output of block.Residual error module 640 includes multiple convolution units 642 and multiple processing units 644.Fig. 7 shows residual error module 640
Structural schematic diagram.Such as Fig. 7, residual error module 640 includes two convolution units 642 and two processing units 644, but is not limited to
This.Wherein, the convolution kernel of 64 3*3 sizes is included in convolution unit 642, the step-length of convolution kernel movement is 1, fills image boundary
Parameter Padding be 1, convolution unit 642 does process of convolution to the characteristic of input, generates new feature data;Processing is single
Member 644 is connected with convolution unit 642, does normalized to the new feature data generated by convolution unit 642, then will processing
Characteristic afterwards is input in next convolution unit 642, and so on.Optionally, processing unit 644 can also be to normalizing
Changing treated, characteristic do activation processing (Nonlinear Mapping is done to characteristic), common excitation function has:
Sigmoid, Tanh (tanh), ReLU, Leaky ReLU, ELU, Maxout etc..According to one embodiment of present invention, it selects
Activation processing is done to the characteristic after normalized with Leaky ReLU excitation functions.First enhancing net according to the present invention
Network 600 simplifies the depth of convolutional network structure, and then reduce and calculate by the way that residual error module is arranged among multiple convolutional layers
Complexity.
According to a kind of realization method, by low frequency response input the first enhancing network 600 of image, through the first convolutional layer 610
The characteristic of this layer is obtained after processing, then the characteristic of this layer is input to the second convolutional layer 620 and is handled, and is passed through successively
After crossing the second convolutional layer 620, multiple residual error modules 640 and the processing of third convolutional layer 630, the characteristic of output indicates enhancing
Low frequency response afterwards.
Then in step S230, by the high frequency response of image input the second enhancing network, to the high frequency section of image into
Row enhancing, to obtain enhanced high frequency response.That is, enhancing the detail section of image by the second enhancing network.Such as
Described previously, each high frequency response is input to the second enhancing network, obtains corresponding increasing by the high frequency response of image totally 5
High frequency response after strong.
Realization method according to the present invention, the second enhancing network include at least:First convolutional layer, the second convolutional layer, third
Convolutional layer and the second number residual error module between the second convolutional layer and third convolutional layer.As Fig. 8 is shown according to this
The schematic network structure of the second enhancing network 800 of invention one embodiment, the second enhancing network 800 include:First convolution
The 810, second convolutional layer 820 of layer, third convolutional layer 830 and three between the second convolutional layer 820 and third convolutional layer 830
Residual error module 840.
According to one embodiment of present invention, the convolution kernel of 64 3*3 is included in the first convolutional layer 810.Second convolutional layer
Include the convolution kernel of 64 3*3 in 820.Include the convolution kernel of 3 3*3 in third convolutional layer 830.Also, enhance net second
In network 800, the step-length (Stride) of the convolution kernel movement in each convolutional layer is 1, meanwhile, the ginseng for filling image boundary
Number Padding be 1 (Padding=1 indicate will all expands 1 pixel at 4 edges up and down, i.e., width and highly all expand
2 pixels are filled).Certainly, in each convolutional layer in addition to process of convolution, it also may include pondization processing (such as maxpooling), activation
It handles (such as ReLu), but not limited to this.Residual error module 840 will pass through the input of result and the residual error module after process of convolution
It is added, the output as the residual error module.Residual error module 840 in second enhancing network 800 enhances residual in network with first
The network structure of difference module 640 can be consistent, including multiple convolution units and multiple processing units, no longer superfluous herein referring to Fig. 7
It states.Second enhancing network 800 according to the present invention simplifies convolution net by the way that residual error module is arranged among multiple convolutional layers
The depth of network structure, and then reduce the complexity calculated.
According to a kind of realization method, by high frequency response input the second enhancing network 800 of image, through the first convolutional layer 810
The characteristic of this layer is obtained after processing, then the characteristic of this layer is input to the second convolutional layer 820 and is handled, and is passed through successively
After crossing the second convolutional layer 820, multiple residual error modules 840 and the processing of third convolutional layer 830, the characteristic of output indicates enhancing
High frequency response afterwards.
Then in step S240, according to enhanced low frequency response (i.e. through obtained by step S220) and enhanced high frequency
The image that response rebuilds input (i.e. through obtained by step S230), obtains enhanced image.
For example, carrying out the processing of reconstruction image as follows, enhanced image I' is generated:
Wherein, Enhanced_Band (0) indicates enhanced low frequency response, Enhanced_Band (w) (w ∈ [1, M])
Indicate that enhanced high frequency response, M indicate the total number of low frequency response and high frequency response, in the present embodiment, M=5.
If Fig. 9 A show that the image of an input, Fig. 9 B are shown to being generated after Fig. 9 A execution image enchancing methods 200
Enhanced image, Fig. 9 C show the original image (that is, truthful data) of Fig. 9 A, and the comparison from above-mentioned three figure can
It arrives, through treated the image of image enchancing method 200 while highlighting with color, effectively enhances image detail
(as shown in house in figure, branch).
Image enhancement schemes according to the present invention first decompose image according to dct transform, be divided into low frequency part and
Then high frequency section carries out enhancing processing using different enhancing networks to low frequency part and high frequency section, obtains enhanced
Low frequency part lays particular emphasis on the color for enhancing image and brightness, and enhanced high frequency section lays particular emphasis on the details of enhancing image.Cause
This this programme effectively enhances the detailed information of image while enhancing color of image and luminance information.
In addition, multiple low frequency response to being obtained through DCT predetermined decomposition nuclear decomposition and high frequency response, are respectively adopted individually
Enhancing network is handled, and the difficulty of image enhancement is reduced.Each enhancing network only needs to learn the image to some frequency range
Enhanced.
Further, the image enhancement network structure based on dct transform used by this programme can also be extended to all
In such as image processing method of image denoising, image super-resolution pixel-by-pixel, there is very strong application.
It should be appreciated that in order to simplify the disclosure and help to understand one or more of each inventive aspect, it is right above
In the description of exemplary embodiment of the present invention, each feature of the invention be grouped together into sometimes single embodiment, figure or
In person's descriptions thereof.However, the method for the disclosure should be construed to reflect following intention:I.e. claimed hair
The bright feature more features required than being expressly recited in each claim.More precisely, as the following claims
As book reflects, inventive aspect is all features less than single embodiment disclosed above.Therefore, it then follows specific real
Thus the claims for applying mode are expressly incorporated in the specific implementation mode, wherein each claim itself is used as this hair
Bright separate embodiments.
Those skilled in the art should understand that the module of the equipment in example disclosed herein or unit or groups
Part can be arranged in equipment as depicted in this embodiment, or alternatively can be positioned at and the equipment in the example
In different one or more equipment.Module in aforementioned exemplary can be combined into a module or be segmented into addition multiple
Submodule.
Those skilled in the art, which are appreciated that, to carry out adaptively the module in the equipment in embodiment
Change and they are arranged in the one or more equipment different from the embodiment.It can be the module or list in embodiment
Member or component be combined into a module or unit or component, and can be divided into addition multiple submodule or subelement or
Sub-component.Other than such feature and/or at least some of process or unit exclude each other, it may be used any
Combination is disclosed to all features disclosed in this specification (including adjoint claim, abstract and attached drawing) and so to appoint
Where all processes or unit of method or equipment are combined.Unless expressly stated otherwise, this specification (including adjoint power
Profit requires, abstract and attached drawing) disclosed in each feature can be by providing the alternative features of identical, equivalent or similar purpose come generation
It replaces.
The present invention discloses together:
A9, the method as described in any one of A1-8, wherein the first enhancing network includes at least:First convolutional layer,
Second convolutional layer, third convolutional layer and the first number residual error module between the second convolutional layer and third convolutional layer.
A10, the method as described in A9, wherein include 64 5*5 in the first convolutional layer in the first enhancing network
Convolution kernel, described first enhances the convolution kernel for including 64 3*3 in the second convolutional layer in network, and described first enhances in network
Third convolutional layer in include the convolution kernel of 3 3*3.
A11, the method as described in any one of A1-10, wherein the second enhancing network includes at least:First convolution
Layer, the second convolutional layer, third convolutional layer and the second number residual error module between the second convolutional layer and third convolutional layer.
A12, the method as described in A11, wherein include 64 3*3 in the first convolutional layer in the second enhancing network
Convolution kernel, include the convolution kernel of 64 3*3 in the second convolutional layer in the second enhancing network, described second enhances network
In third convolutional layer in include the convolution kernel of 3 3*3.
A13, the method as described in A9 or 11, wherein the residual error module be suitable for will by result after process of convolution with
The input of the residual error module is added, the output as the residual error module.
A14, the method as described in A11 or 13, wherein the residual error module includes:Convolution unit is suitable for the spy to input
Sign data do process of convolution, generate new feature data;And processing unit, suitable for the new feature number to being generated by convolution unit
According to doing normalized.
A15, the method as described in any one of A1-14, wherein described according to enhanced low frequency response and enhanced
High frequency response rebuilds the step of image and includes:The processing of the reconstruction image is carried out as follows:
Wherein, Enhanced_Band (0) indicates enhanced low frequency response, Enhanced_Band (w) (w ∈ [1, M])
Indicate that enhanced high frequency response, I' indicate that enhanced image, M indicate the total number of low frequency response and high frequency response.
A16, the method as described in A2, wherein be directed to the dct transform of 8*8 sizes, the transformation kernel of two-dimensional dct transform
Kernel (i, j) is defined as:
Wherein, i ∈ 0~7, j ∈ 0~7, k ∈ 0~7, l ∈ 0~7, patchsize=8, and
A17, the method as described in A16, wherein generate at least one low frequency as follows and decompose core and at least one high
Frequency division solution core:
New_DCT_decom_kernel (0)=DCT_decom_kernel (0)
Wherein, New_DCT_decom_kernel (0) expressions low frequency decomposition core, New_DCT_decom_kernel (1)~
New_DCT_decom_kernel (5) indicates that high-frequency decomposition core, DCT_decom_kernel (u) indicate u-th of predetermined decomposition core.
A18, the method as described in A9, wherein first number is 4.
A19, the method as described in A11, wherein second number is 3.
In addition, it will be appreciated by those of skill in the art that although some embodiments described herein include other embodiments
In included certain features rather than other feature, but the combination of the feature of different embodiments means in of the invention
Within the scope of and form different embodiments.For example, in the following claims, embodiment claimed is appointed
One of meaning mode can use in any combination.
Various technologies described herein are realized together in combination with hardware or software or combination thereof.To the present invention
Method and apparatus or the process and apparatus of the present invention some aspects or part can take embedded tangible media, such as it is soft
The form of program code (instructing) in disk, CD-ROM, hard disk drive or other arbitrary machine readable storage mediums,
Wherein when program is loaded into the machine of such as computer etc, and is executed by the machine, the machine becomes to put into practice this hair
Bright equipment.
In the case where program code executes on programmable computers, computing device generally comprises processor, processor
Readable storage medium (including volatile and non-volatile memory and or memory element), at least one input unit, and extremely
A few output device.Wherein, memory is configured for storage program code;Processor is configured for according to the memory
Instruction in the said program code of middle storage executes method of the present invention.
By way of example and not limitation, computer-readable medium includes computer storage media and communication media.It calculates
Machine readable medium includes computer storage media and communication media.Computer storage media storage such as computer-readable instruction,
The information such as data structure, program module or other data.Communication media is generally modulated with carrier wave or other transmission mechanisms etc.
Data-signal processed embodies computer-readable instruction, data structure, program module or other data, and includes that any information passes
Pass medium.Above any combination is also included within the scope of computer-readable medium.
In addition, be described as herein can be by the processor of computer system or by executing for some in the embodiment
The combination of method or method element that other devices of the function are implemented.Therefore, have for implementing the method or method
The processor of the necessary instruction of element forms the device for implementing this method or method element.In addition, device embodiment
Element described in this is the example of following device:The device is used to implement performed by the element by the purpose in order to implement the invention
Function.
As used in this, unless specifically stated, come using ordinal number " first ", " second ", " third " etc.
Description plain objects are merely representative of the different instances for being related to similar object, and are not intended to imply that the object being described in this way must
Must have the time it is upper, spatially, in terms of sequence or given sequence in any other manner.
Although the embodiment according to limited quantity describes the present invention, above description, the art are benefited from
It is interior it is clear for the skilled person that in the scope of the present invention thus described, it can be envisaged that other embodiments.Additionally, it should be noted that
The language that is used in this specification primarily to readable and introduction purpose and select, rather than in order to explain or limit
Determine subject of the present invention and selects.Therefore, without departing from the scope and spirit of the appended claims, for this
Many modifications and changes will be apparent from for the those of ordinary skill of technical field.For the scope of the present invention, to this
The done disclosure of invention is illustrative and not restrictive, and it is intended that the scope of the present invention be defined by the claims appended hereto.
Claims (10)
1. a kind of image enchancing method, the method is suitable for executing in computing device, the method includes the steps:
It is decomposed according to the image of the predetermined decomposition verification input of dct transform, obtains the low frequency response and high frequency of described image
Response;
By low frequency response input the first enhancing network of described image, enhanced low frequency response is obtained;
By high frequency response input the second enhancing network of described image, enhanced high frequency response is obtained;And
Described image is rebuild according to the enhanced low frequency response and the enhanced high frequency response, obtains enhanced figure
Picture.
2. the method as described in claim 1 further includes the steps that the predetermined decomposition core for generating dct transform:
Calculate the transformation kernel of two-dimensional dct transform;
Coded treatment is carried out to the transformation kernel, to generate new transformation kernel;And
According to the predetermined decomposition core of the new transformation karyogenesis dct transform.
3. method as claimed in claim 2, wherein described to carry out the step of coded treatment is to generate new transformation kernel to transformation kernel
Including:
The sequence scanned according to zigzag resequences to the transformation kernel, obtains new transformation kernel.
4. method as claimed in claim 2, wherein the basis newly converts the step of the predetermined decomposition core of karyogenesis dct transform
Suddenly include:
The new transformation kernel is rotated into 180 degree, transformation kernel after being rotated;And
Process of convolution is done to transformation kernel after the new transformation kernel and the rotation, generates the predetermined decomposition core of dct transform.
5. the method as described in any one of claim 1-4, wherein according to the figure of the predetermined decomposition verification input of dct transform
As decomposed, obtained image low frequency response and high frequency response the step of include:
The predetermined decomposition core of dct transform is pre-processed, core and at least one high frequency division are decomposed to obtain at least one low frequency
Solve core;
The image that verification input is decomposed according to the low frequency is decomposed, and is obtained the corresponding at least one low frequency of described image and is rung
It answers;And
It is decomposed according to the image of high-frequency decomposition verification input, obtains the corresponding at least one high frequency sound of described image
It answers.
6. method as claimed in claim 5, wherein the predetermined decomposition core to dct transform pre-processed, with obtain to
Lacking a step of low frequency decomposes core and at least one high-frequency decomposition core includes:
Multiple predetermined decomposition cores are unequally divided into multiple classification successively;
Calculate the summation of the predetermined decomposition core under each classification:
The summation of predetermined decomposition core under first classification is decomposed into core as low frequency;With
Using the summation of the predetermined decomposition core under other each classification as a high-frequency decomposition core.
7. method as claimed in claim 6, wherein the low frequency response of image is defined as:
Wherein, New_DCT_decom_kernel (0) indicates that low frequency decomposes core, and I indicates the image of input, New_Band_
Responds (0) indicates low frequency response,Indicate convolution algorithm.
8. method as claimed in claim 5, wherein the high frequency response of image is defined as:
Wherein, New_DCT_decom_kernel (w) indicates that w-th of high-frequency decomposition core, I indicate the image of input, New_Band_
Responds (w) indicates w-th of high frequency response,Indicate convolution algorithm.
9. a kind of computing device, including:
One or more processors;With
Memory;
One or more programs, wherein one or more of programs are stored in the memory and are configured as by described one
A or multiple processors execute, and one or more of programs include for executing according in claim 1-8 the methods
The instruction of either method.
10. a kind of computer readable storage medium of the one or more programs of storage, one or more of programs include instruction,
Described instruction is when computing device executes so that the computing device executes appointing in the method according to claim 1-8
One method.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810122975.XA CN108492261A (en) | 2018-02-07 | 2018-02-07 | A kind of image enchancing method and computing device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810122975.XA CN108492261A (en) | 2018-02-07 | 2018-02-07 | A kind of image enchancing method and computing device |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108492261A true CN108492261A (en) | 2018-09-04 |
Family
ID=63344705
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810122975.XA Pending CN108492261A (en) | 2018-02-07 | 2018-02-07 | A kind of image enchancing method and computing device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108492261A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111161189A (en) * | 2019-12-31 | 2020-05-15 | 杭州电子科技大学 | Single image re-enhancement method based on detail compensation network |
CN112561843A (en) * | 2020-12-11 | 2021-03-26 | 北京百度网讯科技有限公司 | Method, apparatus, device and storage medium for processing image |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160364843A1 (en) * | 2012-09-21 | 2016-12-15 | Htc Corporation | Image enhancement methods and systems using the same |
CN106296600A (en) * | 2016-08-01 | 2017-01-04 | 大连理工大学 | A kind of contrast enhancement process decomposed based on wavelet image |
CN107451981A (en) * | 2017-08-15 | 2017-12-08 | 四川长虹电器股份有限公司 | Picture noise level estimation method based on DCT and gradient covariance matrix |
CN107492070A (en) * | 2017-07-10 | 2017-12-19 | 华北电力大学 | A kind of single image super-resolution computational methods of binary channels convolutional neural networks |
-
2018
- 2018-02-07 CN CN201810122975.XA patent/CN108492261A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160364843A1 (en) * | 2012-09-21 | 2016-12-15 | Htc Corporation | Image enhancement methods and systems using the same |
CN106296600A (en) * | 2016-08-01 | 2017-01-04 | 大连理工大学 | A kind of contrast enhancement process decomposed based on wavelet image |
CN107492070A (en) * | 2017-07-10 | 2017-12-19 | 华北电力大学 | A kind of single image super-resolution computational methods of binary channels convolutional neural networks |
CN107451981A (en) * | 2017-08-15 | 2017-12-08 | 四川长虹电器股份有限公司 | Picture noise level estimation method based on DCT and gradient covariance matrix |
Non-Patent Citations (5)
Title |
---|
JIANRUI CAI等: "Learning a Deep Single Image Contrast Enhancer from Multi-Exposure Images", 《IEEE TRANSACTIONS ON IMAGE PROCESSING》 * |
KAI ZHANG等: "Beyond a Gaussian Denoiser Residual Learning of Deep CNN for Image Denoising", 《IEEE TRANSACTIONS ON IMAGE PROCESSING》 * |
KAIMING HE等: "Deep Residual Learning for Image Recognition", 《2016 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 * |
张亚飞等: "基于分块DCT同态滤波的彩色图像增强算法", 《计算机工程与设计》 * |
王俊杰著: "《数字水印与信息安全技术研究》", 31 August 2014, 知识产权出版社 * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111161189A (en) * | 2019-12-31 | 2020-05-15 | 杭州电子科技大学 | Single image re-enhancement method based on detail compensation network |
CN112561843A (en) * | 2020-12-11 | 2021-03-26 | 北京百度网讯科技有限公司 | Method, apparatus, device and storage medium for processing image |
CN112561843B (en) * | 2020-12-11 | 2024-07-02 | 上饶市纯白数字科技有限公司 | Method, apparatus, device and storage medium for processing image |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Zha et al. | From rank estimation to rank approximation: Rank residual constraint for image restoration | |
JP6874168B2 (en) | Image style conversion methods and devices, equipment, and storage media | |
Wang et al. | Multi-scale dilated convolution of convolutional neural network for image denoising | |
Mairal et al. | Learning multiscale sparse representations for image and video restoration | |
CN110610237A (en) | Quantitative training method and device of model and storage medium | |
CN110310229A (en) | Image processing method, image processing apparatus, terminal device and readable storage medium storing program for executing | |
CN108648163A (en) | A kind of Enhancement Method and computing device of facial image | |
Liu et al. | Graph-based joint dequantization and contrast enhancement of poorly lit JPEG images | |
CN108038823A (en) | Image-type becomes the training method of network model, image-type becomes method and computing device | |
CN109325928A (en) | A kind of image rebuilding method, device and equipment | |
CN109978063A (en) | A method of generating the alignment model of target object | |
CN107454284A (en) | A kind of video denoising method and computing device | |
Yang et al. | Analyzing the effect of JPEG compression on local variance of image intensity | |
CN110569961A (en) | neural network training method and device and terminal equipment | |
CN107886516A (en) | The method and computing device that hair moves towards in a kind of calculating portrait | |
CN110148100A (en) | A kind of image conversion method, device, storage medium and computer equipment | |
CN109671061A (en) | A kind of image analysis method, calculates equipment and storage medium at device | |
CN109712082A (en) | The method and device of figure is repaired in cooperation | |
CN109978137A (en) | A kind of processing method of convolutional neural networks | |
CN108492261A (en) | A kind of image enchancing method and computing device | |
CN114978189A (en) | Data coding method and related equipment | |
Shiau et al. | A low-cost hardware architecture for illumination adjustment in real-time applications | |
Yang et al. | IBM3D: integer BM3D for efficient image denoising | |
CN107481203A (en) | A kind of image orientation filtering method and computing device | |
CN103745443B (en) | The method and apparatus for improving picture quality |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180904 |
|
RJ01 | Rejection of invention patent application after publication |