CN110428473A - A kind of confrontation based on auxiliary variable generates the color image gray processing method of network - Google Patents
A kind of confrontation based on auxiliary variable generates the color image gray processing method of network Download PDFInfo
- Publication number
- CN110428473A CN110428473A CN201910529133.0A CN201910529133A CN110428473A CN 110428473 A CN110428473 A CN 110428473A CN 201910529133 A CN201910529133 A CN 201910529133A CN 110428473 A CN110428473 A CN 110428473A
- Authority
- CN
- China
- Prior art keywords
- network
- image
- color image
- gray
- confrontation
- 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.)
- Granted
Links
Classifications
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- 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
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/90—Determination of colour characteristics
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of, and the confrontation based on auxiliary variable generates the color image gray processing method of network, the following steps are included: step A: examining whether input picture is color image, gradient correlation similarity gray processing (GcsDecolor) algorithm is then used to carry out gray processing processing to it if color image, and replicate image after gray processing, obtain the contrast images that three parts of gray level images generate network as confrontation；Step B: confrontation of the design based on auxiliary variable generates network (AV-GAN), training AV-GAN network；Step C: by color image by having trained the AV-GAN network completed to test, final gray level image is obtained.The present invention keeps color image gray processing computational efficiency higher, and can be reserved for the notable feature of color image, make gray level image can retaining color sequence, preferably reflect the structural similarity between colored and gray level image.
Description
Technical field
The invention belongs to technical field of computer vision more particularly to color image gray processing technical application, specially one
Confrontation of the kind based on auxiliary variable generates the color image gray processing method of network.
Background technique
In today that Digital Media and science and technology rapidly develop, color image technology is widely used already, but
Gray level image is still few with its data volume and the characteristics such as operation facilitates are active in various directions.Firstly, gray processing processing has warp
Advantage in Ji, in order to save printing cost, many textbooks, the paper delivered, that most of newpapers and periodicals tend to export price is cheap
Gray level image with distinct contrast；Gray processing is for helping colour blindness crowd to be also of great significance, and color barrier patient is not by sympathizing with
Condition demand chooses different platforms, and it is negative can to solve color barrier patient bring due to color cannot be distinguished to a certain extent
It influences.Secondly, some image processing techniques, for gray level image can more easily operation, for example, currently being ground extensively by everybody
The popular domain studied carefully such as pattern-recognition and machine vision all select the gray level image table for rapidly using data volume small when pretreatment
Show color image, the processing speed of subsequent algorithm can not only be improved in this way, additionally it is possible to greatly improve algorithm synthesis application actual effect.
Finally, gray processing processing also has application in terms of Graphic Arts, the black-and-white photography for such as obtaining gray level image also continues to obtain one
A little shutterbugs' pursues.Therefore, the research of color image gray processing is of great significance and value.
Color is that 3D vector is mapped to 1D scalar to gradation conversion target, it is substantially a reduction process, unfortunate
Be that the end, it is inevitably by information loss.Therefore, many discoloration methods, which have, is proposed to come from human perception
Solve the viewpoint of this problem.Conventional method gray scale for conveying color can be roughly divided into two classes: local directed complete set method and
Global adaptation method.In the first kind, the mapping value of color to the gray scale of pixel is usually spatially varying, and depends on local face
Color distribution.For example, Bala and Eschbach propose a kind of radio-frequency component that the method for retaining color boundary passes through addition coloration
To realize luminance channel.Neumann et al. rebuilds gray level image measurement color and brightness contrast from color image gradual change
Spend the comparison as gradient color space.Smith et al., which decomposes, to be divided the image into several frequency contents and is adjusted using color
The combination weight of chrominance channel.Although they, which have, retains local feature, if mapped the advantages of color constancy, region can be uneven
The variation in area is converted evenly.
Global Algorithm is broadly divided into the dimensionality reduction class based on transformation and is based on the optimization of color difference (pixel color contrast)
Two kinds of algorithm.For the method for the dimensionality reduction class of transformation, using PCA transformation dimensionality reduction as main representative；For pixel color contrast
Method, the thought of this kind of algorithms are to comprehensively utilize cromogram when constructing the mapping function from color image to gray level image
As the brightness value information and color contrast information of pixel, and more as far as possible original color image adjacent area is mapped to gray level image
Different colours comparative information, to increase the contrast of gray level image.After having constructed mapping objects function, one is reconstructed
Then optimization equation corresponding with mapping objects function is obtained by solving optimization equation closest to target brightness value
Gray level image.Due to mapping function be it is flexible and changeable, different mapping functions can be constructed according to different purposes, thus will
After color image information is mapped to gray level image, it is possible that the identical colouring information mapping of different zones in original color image
The case where for identical gray value, there may be the different colouring information of different zones is mapped as same ash in color image
The case where angle value.Its purpose is mainly to discriminate between the feature between the neighbor pixel in color image with different colours.Finally
Result it is related with the color of color image pixel point and its neighborhood information.
There are two disadvantages for these existing methods: robustness and high calculating cost.In order to solve these difficulties, Yi Xieyan
Study carefully personnel to rethink using the simple RGB2GRAY model of tradition.In particular, it assumes that gray scale output is color image
The linear combination of middle RGB channel, i.e.,Wherein Ir, Ig, IbIt respectively represents
RGB color channel components.In the rgb2gray function of classical Matlab, the weight of all images is fixed.Recently, have
Researcher attempts that channel weight is adaptive selected under certain measures.The solution space of the discretizations linear parametric model such as Lu with
66 candidate values, then determine one reach can magnitude candidate as best solution, this is current most fast
Algorithm.Liu et al. proposes new gradient correlation similarity (Gcs) model, and input color figure is carried out between each channel
The gray level image of picture and generation, the color sequence for preferably reflecting the differentiability and color to grey of keeping characteristics are converted.He
With minimum function Gcs value determined solution linear parametric model induce discrete search candidate image.
Summary of the invention
The purpose of the present invention is to provide a kind of, and the confrontation based on auxiliary variable generates the color image gray processing side of network
Method, to solve the problems mentioned in the above background technology.
To achieve the above object, the invention provides the following technical scheme: a kind of confrontation based on auxiliary variable generates network
Color image gray processing method, comprising the following steps:
Step A: examining whether input picture is color image, and gradient correlation similarity gray scale is then used if color image
Change (GcsDecolor) algorithm and carry out gray processing processing to it, and image after gray processing is replicated, obtains three parts of gray processings
Image generates the contrast images of network as confrontation.
Step B: confrontation of the design based on auxiliary variable generates network (AV-GAN), training AV-GAN network.
Step C: by color image by having trained the AV-GAN network completed to test, final gray processing figure is obtained
Picture.
Further, the step A are as follows:
Assuming that input color image is rgb format, wherein R, G, B represent RGB channel, use GcsDecolor algorithm pair
The color image of input carries out the processing of gray processing, obtains gray level image.
It is constrained to 1 using single order multinomial function c={ r, g, b } and by the sum of weight, calculates original color images
Each channel in gradient amplitude and obtained gray level image between whole pixel similitude, it may be assumed that
Next, using gradient correlation come the holding of description scheme, and obtained in each channel of rgb space
Similarity is calculated between gray level image and original image, obtains the channel of three gray processings, then three gray level images are carried out
Adduction, obtains final gray level image, finally replicates the image after obtained gray processing, obtain three parts of identical ashes
Degreeization image generates the reference of network discriminator as confrontation.
Further, the step B are as follows:
It constructs the confrontation based on auxiliary variable and generates network (AV-GAN), and it is trained, it will be in color image
Input of R, G, the B triple channel as network, AV-GAN network include a generator and a discriminator, and generator includes 14
Convolutional layer and several active coatings；Wherein, convolutional layer minimizes the distance between two images as unit of pixel, enables F (xi；
It is θ) i of ConvNet modelithTrained output, by trained loss is defined as:
Wherein, p indicates each pixel, and n indicates the total pixel number in image；Then its overall goal can indicate are as follows:
Wherein, N indicates the sum of training example；The behavior of this loss function is, by average value as a result, to minimum
Change loss.
The input of discriminator is the gray scale picture that color image generates after generator, and wherein the generation of gray scale picture is such as
Under: use a kind of dimensionality reduction technology involved, by addition auxiliary variable and by sample training constrain they realization from colour to
The transformation of grey is related to three features:
(1) it outputs and inputs as gradient field；
(2) by auxiliary variable technology, so that it is identical to output and input variable channel number；
(3) using L1 norm for overcoming the shortcomings of gradient amplitude；
Used network losses function are as follows:Wherein, I={ IR, IG, IB, H={ g, g, g },
On the basis of this loss function, auxiliary variable is now added, then carry out topography's processing, obtain gray level image.
Discriminator is made of 11 coding layers, similar with generator coding, and convolution of each coding layer by stride greater than 1 is transported
Relu activation composition is calculated, standardizes in batches and reveals, the last layer is activated by sigmoid, number of the return one from 0 to 1,
To explain that input as very as false probability, using three width gray level images in step A as benchmark is judged, is judged as true, then
Return to 1, it is different then return to 0.
For AV-GAN network, generator and discriminator are all adjusted on input x, pass through qgTo generator into
Row parametrization, uses qdDiscriminator is parameterized, minimax objective function are as follows:
In the state of guaranteeing its steady operation, it is contemplated that input x in generator and export the L1 difference between y, every
In secondary iteration, discriminator maximizes q according to above formulad, and generator will minimize in the following manner:
Network is trained in this manner, obtains AV-GAN network.
Compared with prior art, the beneficial effects of the present invention are:
The present invention is based on the confrontation of auxiliary variable to generate in the color image gray processing method of network (AV-GAN) to colour
Significance level, the similarity of three sub-spaces of image are furtherd investigate, and fully consider the reference information of three sub-spaces, and
By three sub-spaces by the AV-GAN network of training, more accurate gray level image is obtained.The present invention is related with normalization
Property saves the notable feature of color image, distinguish color image still can in gray level image, and can protect
Color is stayed to sort；And it can preferably reflect the structural similarity between colored and gray level image, obtain higher cromogram
As gray processing ability.By the color image gray processing method, the speed of image gray processing processing can be improved, guarantee image ash
The precision of degreeization processing simultaneously, can be adapted for different scenes, can prevent processing caused by changing due to extraneous factor
Problem of Failure.
Detailed description of the invention
Fig. 1 is flow chart of the invention；
Fig. 2 is AV-GAN network frame figure of the invention；
Fig. 3 is the generator frame diagram in AV-GAN network of the present invention；
Fig. 4 is the discriminator frame diagram in AV-GAN network of the present invention；
Fig. 5 is that Set5 data set is tested by AV-GAN network, obtains final gray level image；
Fig. 6 is the comparison diagram after the color image difference gray processing in Cadik data set.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is described in further detail.The specific embodiments are only for explaining the present invention technical solution described herein, and
It is not limited to the present invention.
Description 1 describes the color image gray processing algorithm of AV-GAN network according to the present invention.
In step: specific implementation is as follows:
Firstly, generating training data: choosing BSDS300 database color image as input；
Secondly, carrying out gray processing to it using gradient correlation similarity gray processing (GcsDecolor) algorithm, threeway is obtained
The gray level image in road.
Assuming that input color image be rgb format, wherein R, G, B represent RGB channel, in order to reduce data processing when
Between picture pixels point is averagely reduced to original half, and the value by reading its each central point obtains the RGB of the pixel
Value and be stored in array, in order to keep feature differentiability into gradation conversion in color, minimize input color and gained gray scale
The distance of pixel difference between image.I.e., it is assumed that input color image is rgb format, wherein indexing R, G, B represents RGB channel,
It enablesIt is the color contrast for having value of symbol with the difference for indicating color pair, and
gx-gyIndicate the gray scale difference value between pixel, the energy function based on classical L2 norm are as follows:P
Pixel is represented to pond, it includes local and non local candidates；To have rgb value and obtain array random alignment, and with former array value
It is poor to make, and obtains referential data, chooses minimum difference, R is chosen according to minimum value, the corresponding place value of G, B determines R, G, B gray scale
Coefficient before change, obtains gray level image.By the way that the difference of remote pixel to be integrated into energy function, enable model
The pixel and long scale contrasted zones of nearest-neighbors are utilized well.
Using single order multinomial function c={ r, g, b } and the sum of weight is constrained to 1.Calculate original color images
Each channel in gradient amplitude and obtained gray level image between whole pixel similitude, it may be assumed that
Next, using gradient correlation come the holding of description scheme, rather than common gradient error and in rgb space
Each channel in calculate similarity between obtained gray level image and original image, obtain the channel of three gray processings, then
Three gray level images are summed up, final gray level image is obtained.Finally the image after obtained gray processing is carried out
Duplication, obtains three parts of identical gray level images, and the reference of network discriminator is generated as confrontation.
In stepb:
It constructs the confrontation based on auxiliary variable and generates network (AV-GAN), and it is trained, it will be in color image
Input of R, G, the B triple channel as network, AV-GAN network include a generator and a discriminator, and generator includes 14
Convolutional layer and several active coatings；Wherein, convolutional layer minimizes the distance between two images as unit of pixel, enables F (xi；
It is θ) i of ConvNet modelithTrained output, by trained loss is defined as:
Wherein, p indicates each pixel, and n indicates the total pixel number in image.Then its overall goal can indicate are as follows:
Wherein, N indicates the sum of training example.The behavior of this loss function is, by average value as a result, to minimum
Change loss.
The input of discriminator is the gray scale picture that color image generates after generator, and wherein the generation of gray scale picture is such as
Under: use a kind of dimensionality reduction technology involved, by addition auxiliary variable and by sample training constrain they realization from colour to
The transformation of grey is related to three features:
(1) it outputs and inputs as gradient field；
(2) by auxiliary variable technology, so that it is identical to output and input variable channel number；
(3) using L1 norm for overcoming the shortcomings of gradient amplitude；
Used network losses function are as follows:Wherein, I={ IR, IG, IB, H={ g, g, g },
On the basis of this loss function, auxiliary variable is now added, then carry out topography's processing, obtain gray level image.
Discriminator is made of 11 coding layers, similar with generator coding, and convolution of each coding layer by stride greater than 1 is transported
It calculates, batch standardizes and leakage relu activation composition.The last layer is activated by sigmoid, returns to a number from 0 to 1,
To explain input as very as false probability, in the present invention, using three width gray level images in step A as judging benchmark,
It is judged as true, then returns to 1, it is different then return to 0.
For AV-GAN network, generator and discriminator are all adjusted on input x, pass through qgTo generator into
Row parametrization, uses qdDiscriminator is parameterized, minimax objective function are as follows:
In the state of guaranteeing its steady operation, we consider input x in generator and the L1 difference between y is exported,
In each iteration, discriminator maximizes q according to above formulad, and generator will minimize in the following manner:
Network is trained in this manner, obtains AV-GAN network.
In step C: by color image by having trained the AV-GAN network completed to test, obtaining final gray scale
Change image.
As shown in fig. 6, the method for the present invention (d) Cadik data set carry out qualitative analysis, and with Gcs2 algorithm (a),
Gooch algorithm (b), Smith algorithm (c) comparison.Gooch algorithm and Smith algorithm do not fully consider significant stimulation, and for
Some images can generate flat result.Gcs2 method is related using normalization and saves the notable feature of color image.This hair
Bright algorithm can not only make color image that can distinguish in gray level image, and can sequentially save required color sequence.
The present invention in the color image gray processing algorithm based on AV-GAN the experimental results showed that make to output and input port number
It is identical, more detail gray level image is obtained, while confrontation having been used to generate network, enhances image gray processing effect.
The above only expresses the preferred embodiment of the present invention, and the description thereof is more specific and detailed, but can not be because
This and be interpreted as limitations on the scope of the patent of the present invention.It should be pointed out that for those of ordinary skill in the art, In
Under the premise of not departing from present inventive concept, several deformations can also be made, improves and substitutes, these belong to protection of the invention
Range.Therefore, the scope of protection of the patent of the invention shall be subject to the appended claims.
Claims (3)
1. a kind of color image gray processing method that confrontation based on auxiliary variable generates network, it is characterised in that: including following
Step:
Step A: examining whether input picture is color image, and gradient correlation similarity gray processing is then used if color image
(GcsDecolor) algorithm carries out gray processing processing to it, and image after gray processing is replicated, and obtains three parts of gray processing figures
Contrast images as generating network as confrontation；
Step B: confrontation of the design based on auxiliary variable generates network (AV-GAN), training AV-GAN network；
Step C: by color image by having trained the AV-GAN network completed to test, final gray level image is obtained.
2. a kind of confrontation based on auxiliary variable according to claim 1 generates the color image gray processing method of network,
It is characterized by: the step A are as follows:
Assuming that input color image is rgb format, wherein R, G, B represent RGB channel, using GcsDecolor algorithm to input
Color image carry out gray processing processing, obtain gray level image；
Using single order multinomial function c={ r, g, b } and the sum of weight is constrained to 1, calculates the every of original color images
Whole pixel similitude between gradient amplitude in a channel and obtained gray level image, it may be assumed that
Next, using gradient correlation come the holding of description scheme, and in obtained gray scale in each channel of rgb space
Similarity is calculated between image and original image, obtains the channel of three gray processings, then three gray level images are summed up,
Final gray level image is obtained, finally replicates the image after obtained gray processing, obtains three parts of identical gray processings
Image generates the reference of network discriminator as confrontation.
3. a kind of confrontation based on auxiliary variable according to claim 1 generates the color image gray processing method of network,
It is characterized by: the step B are as follows:
It constructs the confrontation based on auxiliary variable and generates network (AV-GAN), and it is trained, by the R in color image, G, B
Input of the triple channel as network, AV-GAN network include a generator and a discriminator, and generator includes 14 convolution
Layer and several active coatings；Wherein, convolutional layer minimizes the distance between two images as unit of pixel, enables F (xi；θ) it is
The output of the iith training of ConvNet model, by trained loss is defined as:
Wherein, p indicates each pixel, and n indicates the total pixel number in image；Then its overall goal can indicate are as follows:
Wherein, N indicates the sum of training example；The behavior of this loss function is, by average value as a result, to minimize damage
It loses；
The input of discriminator is the gray scale picture that color image generates after generator, and discriminator is made of 11 coding layers,
Similar with generator coding, convolution algorithm, batch of each coding layer by stride greater than 1 standardize and leakage relu activation group
At the last layer is activated by sigmoid, returns to a number from 0 to 1, as false probability, will very be walked to explain input
Three width gray level images in rapid A are judged as very as benchmark judge, then return to 1, different then return to 0；
For AV-GAN network, generator and discriminator are all adjusted on input x, pass through qgGenerator is joined
Numberization uses qdDiscriminator is parameterized, minimax objective function are as follows:
In the state of guaranteeing its steady operation, it is contemplated that input x in generator and export the L1 difference between y, changing every time
Dai Zhong, discriminator maximize q according to above formulad, and generator will minimize in the following manner:
Network is trained in this manner, obtains AV-GAN network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910529133.0A CN110428473B (en) | 2019-06-18 | 2019-06-18 | Color image graying method of confrontation generation network based on auxiliary variable |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910529133.0A CN110428473B (en) | 2019-06-18 | 2019-06-18 | Color image graying method of confrontation generation network based on auxiliary variable |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110428473A true CN110428473A (en) | 2019-11-08 |
| CN110428473B CN110428473B (en) | 2022-06-14 |
Family
ID=68408659
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910529133.0A Active CN110428473B (en) | 2019-06-18 | 2019-06-18 | Color image graying method of confrontation generation network based on auxiliary variable |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110428473B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111696026A (en) * | 2020-05-06 | 2020-09-22 | 华南理工大学 | Reversible gray scale map algorithm and computing device based on L0 regular term |
| CN113033561A (en) * | 2019-12-09 | 2021-06-25 | 财团法人资讯工业策进会 | Image analysis device and image analysis method |
| CN113450272A (en) * | 2021-06-11 | 2021-09-28 | 广州方图科技有限公司 | Image enhancement method based on sinusoidal curve change and application thereof |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106023268A (en) * | 2016-05-30 | 2016-10-12 | 南昌大学 | Color image graying method based on two-step parameter subspace optimization |
| CN107862293A (en) * | 2017-09-14 | 2018-03-30 | 北京航空航天大学 | Radar based on confrontation generation network generates colored semantic image system and method |
| US20180293712A1 (en) * | 2017-04-06 | 2018-10-11 | Pixar | Denoising monte carlo renderings using generative adversarial neural networks |
| CN109635774A (en) * | 2018-12-21 | 2019-04-16 | 中山大学 | A kind of human face synthesizing method based on generation confrontation network |
| CN109635511A (en) * | 2019-01-16 | 2019-04-16 | 哈尔滨工业大学 | A kind of high-rise residential areas forced-ventilated schemes generation design method generating confrontation network based on condition |
| US20190171908A1 (en) * | 2017-12-01 | 2019-06-06 | The University Of Chicago | Image Transformation with a Hybrid Autoencoder and Generative Adversarial Network Machine Learning Architecture |
-
2019
- 2019-06-18 CN CN201910529133.0A patent/CN110428473B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106023268A (en) * | 2016-05-30 | 2016-10-12 | 南昌大学 | Color image graying method based on two-step parameter subspace optimization |
| US20180293712A1 (en) * | 2017-04-06 | 2018-10-11 | Pixar | Denoising monte carlo renderings using generative adversarial neural networks |
| CN107862293A (en) * | 2017-09-14 | 2018-03-30 | 北京航空航天大学 | Radar based on confrontation generation network generates colored semantic image system and method |
| US20190171908A1 (en) * | 2017-12-01 | 2019-06-06 | The University Of Chicago | Image Transformation with a Hybrid Autoencoder and Generative Adversarial Network Machine Learning Architecture |
| CN109635774A (en) * | 2018-12-21 | 2019-04-16 | 中山大学 | A kind of human face synthesizing method based on generation confrontation network |
| CN109635511A (en) * | 2019-01-16 | 2019-04-16 | 哈尔滨工业大学 | A kind of high-rise residential areas forced-ventilated schemes generation design method generating confrontation network based on condition |
Non-Patent Citations (3)
| Title |
|---|
| CHENGCHENG LI 等: ""Fast-Converging Conditional Generative Adversarial Networks for Image Synthesis"", 《2018 25TH IEEE INTERNATIONAL CONFERENCE ON IMAGE PROCESSING (ICIP)》 * |
| QIEGEN LIU 等: ""GcsDecolor: Gradient Correlation Similarity for Efficient Contrast Preserving Decolorization"", 《IEEE TRANSACTIONS ON IMAGE PROCESSING》 * |
| 万里鹏: ""基于生成对抗网络的多属性人脸生成及辅助识别研究"", 《中国优秀博硕士学位论文全文数据库(硕士)信息科技辑》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113033561A (en) * | 2019-12-09 | 2021-06-25 | 财团法人资讯工业策进会 | Image analysis device and image analysis method |
| CN113033561B (en) * | 2019-12-09 | 2023-07-07 | 财团法人资讯工业策进会 | Image analysis device and image analysis method |
| CN111696026A (en) * | 2020-05-06 | 2020-09-22 | 华南理工大学 | Reversible gray scale map algorithm and computing device based on L0 regular term |
| CN111696026B (en) * | 2020-05-06 | 2023-06-23 | 华南理工大学 | Reversible gray scale graph algorithm and computing equipment based on L0 regular term |
| CN113450272A (en) * | 2021-06-11 | 2021-09-28 | 广州方图科技有限公司 | Image enhancement method based on sinusoidal curve change and application thereof |
| CN113450272B (en) * | 2021-06-11 | 2024-04-16 | 广州方图科技有限公司 | Image enhancement method based on sinusoidal variation and application thereof |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110428473B (en) | 2022-06-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108537742B (en) | Remote sensing image panchromatic sharpening method based on generation countermeasure network | |
| WO2019178702A1 (en) | Systems and methods for polygon object annotation and a method of training an object annotation system | |
| CN108038445B (en) | SAR automatic target identification method based on multi-view deep learning framework | |
| CN105046276B (en) | Hyperspectral image band selection method based on low-rank representation | |
| CN110428473A (en) | A kind of confrontation based on auxiliary variable generates the color image gray processing method of network | |
| CN106022363B (en) | A kind of Chinese text recognition methods suitable under natural scene | |
| CN108021930B (en) | Self-adaptive multi-view image classification method and system | |
| CN109033107A (en) | Image search method and device, computer equipment and storage medium | |
| CN106295694A (en) | A kind of face identification method of iteration weight set of constraints rarefaction representation classification | |
| CN110598564B (en) | OpenStreetMap-based high-spatial-resolution remote sensing image transfer learning classification method | |
| CN114821198B (en) | Cross-domain hyperspectral image classification method based on self-supervision and small sample learning | |
| CN110175248B (en) | Face image retrieval method and device based on deep learning and Hash coding | |
| CN108428220A (en) | Satellite sequence remote sensing image sea island reef region automatic geometric correction method | |
| Liu et al. | Spectral shift mitigation for cross-scene hyperspectral imagery classification | |
| CN112002009A (en) | Unsupervised three-dimensional face reconstruction method based on generation of confrontation network | |
| Ding et al. | Robust spatial–spectral block-diagonal structure representation with fuzzy class probability for hyperspectral image classification | |
| CN106023268B (en) | A kind of color image gray processing method based on the optimization of two step Parameter Subspaces | |
| CN113837191B (en) | Cross-star remote sensing image semantic segmentation method based on bidirectional unsupervised domain adaptive fusion | |
| CN111680579A (en) | Remote sensing image classification method for adaptive weight multi-view metric learning | |
| CN114663880A (en) | Three-dimensional target detection method based on multi-level cross-modal self-attention mechanism | |
| CN104778683B (en) | A kind of multi-modality images dividing method based on Functional Mapping | |
| Ma et al. | Light self-Gaussian-attention vision transformer for hyperspectral image classification | |
| CN107184224B (en) | Lung nodule diagnosis method based on bimodal extreme learning machine | |
| CN117079098A (en) | Space small target detection method based on position coding | |
| CN116758340A (en) | Small target detection method based on super-resolution feature pyramid and attention mechanism |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |