CN109815875A - One kind is for versicolor system in internet detection - Google Patents
One kind is for versicolor system in internet detection Download PDFInfo
- Publication number
- CN109815875A CN109815875A CN201910043342.4A CN201910043342A CN109815875A CN 109815875 A CN109815875 A CN 109815875A CN 201910043342 A CN201910043342 A CN 201910043342A CN 109815875 A CN109815875 A CN 109815875A
- Authority
- CN
- China
- Prior art keywords
- data
- model
- picture
- versicolor
- original image
- 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
Landscapes
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The present invention provides one kind for versicolor system in internet detection, including Model Management Service device end and data management system, Model Management Service device end includes reading data and initialization module, for original image data to be read and initialized from data management system, the original image data include: base-line data, conventional application scenarios data or unconventional application scenarios data;It further include model construction module, for constructing colour switching network model, the output layer of the colour switching network model calculates the element in transformation matrix according to the picture feature extracted, to obtain transformation matrix Di‑σ;Utilize transformation matrix Di‑σTo any one original image data that the reading data and initialization module are read, by it from the picture under shooting condition under any illumination condition, the picture being converted under standard illumination condition.The present invention can obtain a large amount of data, without considering illumination condition.
Description
Technical field
The present invention relates to internet detection field, more particularly relates to a kind of versicolor system.
Background technique
With the fast development of artificial intelligence field machine learning techniques, internet detection field has also won fast development,
It is referred in " one kind is for measurement detection system and method neural network based in internet detection " pre- using neural network
Survey the system and method for model.
The weakness of this method is to need largely to cover the training data of all the case where being likely to occur, and the distribution of data is close
Collection degree must be relatively strong, and in reality, since the reaction time of test paper is limited, acquisition data often have higher
Cost.
Summary of the invention
In order to solve the problems in the existing technology, it is generated present invention thus provides a kind of using color transformed network
The form of sample enhances the methods of data.
One kind is for versicolor system in internet detection, including Model Management Service device end and data management system
System, Model Management Service device end includes reading data and initialization module, for reading from data management system and initially
Change original image data, the original image data include: base-line data, conventional application scenarios data or unconventional applied field
Scape data;
It further include model construction module, for constructing colour switching network model, the colour switching network model construction
Being includes initial data input layer → convolutional layer → pond layer → convolutional layer → pond layer → output layer, the initial data input
Layer carries out data prediction to the original image data, and the convolutional layer carries out feature extraction to data treated data,
The pond layer carries out Feature Compression to the data after feature extraction, and the output layer goes out to convert according to the feature calculation extracted
Element in matrix, to obtain the transformation matrix D of 3 parameters, 4 parameters or 9 parametersi-o;
To any one original image data that the reading data and initialization module are read, by it from any illumination item
Picture under part under shooting condition, the picture being converted under standard illumination condition;
Conversion formula is as follows:
foutput=Dl-ofinput
Wherein, foutputFor result images function, finputFor input picture function, and Di-oFor transformation matrix.
Further, the base-line data refers to: in data acquisition, exterior light when fixed acquisition takes, mobile phone picture
Machine exposure compensating, white balance state, characteristic value, the mobile phone posture of test paper glazing correction unit etc. carry out test paper picture shooting, right
Collected data carry out feature extraction to calculate corresponding value as base-line data.
Further, the conventional application scenarios data refer to: same test paper, various without fixation in data acquisition
Control condition factor, it is only necessary to which each factor meets the definition of conventional scenario application, can take pictures, and data obtained are passed through
Final data obtained after collection apparatus algorithm
Further, the unconventional scene application data refer to: fixed various control condition factors when acquisition make it not
Meet conventional scenario application conditions, control condition factor include shooting angle, illumination colour temperature, brightness, camera white balance mode,
Camera exposure compensation model, data obtained pass through data obtained after collection apparatus algorithm.
Further, the transformation matrix D of following 3 parameter is obtained by the colour switching networki-o:
Wherein, θr、θg, θbThe respectively feature vector of rgb value in three directions.
The invention has the benefit that
1. the difficulty and cost of data acquisition can be reduced.
2. a large amount of valid data can be obtained, the picture number under standardization illumination condition is converted by the data of nonstandardized technique
According to.
Detailed description of the invention
When considered in conjunction with the accompanying drawings, can be good at understanding that structure of the invention, principle, work are special with reference to following description
Point and advantage, but attached drawing described herein as is used to that of the invention is explained further, and accompanying schematic figure is intended merely to preferably right
The present invention is illustrated, and does not constitute improper restriction to the present invention, in which:
Fig. 1 illustrates color transformed network and master network, and (i.e. " one kind is for survey neural network based in internet detection
Amount detection systems and method " in the neural network mentioned) workflow that combines.
Specific embodiment
Below with reference to example and attached drawing, the invention will be further described, it is noted that following embodiment is only
Be it is schematical, be not intended to limitation the present invention.
Of the invention includes Model Management Service device end, the model management for versicolor system in internet detection
Server end includes reading data and initialization module, model construction module, model training module, model testing module.
The wherein reading data and initialization module: for being read from data management system and initialization data, simultaneously
Data are pre-processed, are normalized etc. with data processing operations.A large amount of image datas are stored in the data management system.
The image data includes: base-line data, conventional application scenarios data, unconventional application scenarios data.
The base-line data refers to: in data acquisition, fixing exterior light photograph when acquisition, mobile phone camera exposure is mended
Repay, characteristic value, mobile phone posture etc. of white balance state, test paper glazing correction unit carry out test paper picture shooting, every test paper is adopted
Collect such as 5-10 data, acquire multiple test paper, feature extraction is carried out to collected data and is made with calculating corresponding value
For base-line data, the average value of multi collect data can be taken as base-line data at this time.The feature can be picture
RGB, HARR, HOG, SIFT, LBP are equivalent.The method of the feature extraction can use prior art progress.The base-line data
Value be training data sample true value, that is, label value.That is every test paper finally corresponds to a label value.
The routine application scenarios data refer to: each data of middle shooting is obtained according to following methods for same
Test paper, in data acquisition without fixed various control condition factors, it is only necessary to each factor meets the definition of conventional scenario application,
It can take pictures, i.e., in such a case, it is possible to arbitrarily shoot picture, every test paper acquires repeatedly respectively, number obtained
According to passing through collection apparatus algorithm and then being for example averaged algorithm finally data obtained, as conventional applied field
The data shot in scape, final every test paper is corresponding to obtain a conventional application scenarios data.Multiple test paper are acquired, routine is obtained
Application scenarios data set.
The unconventional scene application data refer to: fixed various control condition factors when acquisition make it be unsatisfactory for routine
Scene application conditions, control condition factor include shooting angle, illumination colour temperature, brightness, camera white balance mode, camera exposure
Compensation model etc..Data obtained pass through data obtained after collection apparatus algorithm, and every test paper acquires repeatedly respectively,
Data obtained by collection apparatus algorithm and then be for example averaged algorithm finally data obtained as non-
The data shot in conventional application scenarios.Multiple test paper are acquired, unconventional application scenarios data set is obtained.
At initialization model management server end, after confirming the step of data management server end is opened,
The model construction module carries out initialization model:
To any one image data, picture under shooting condition under any illumination condition is converted into standard illumination item
Picture under part;
Assuming that picture code storage in a manner of RGB, then conversion formula is as follows:
foutput=Dl-ofinput
Wherein, foutputFor result images function, finputFor input picture function, and Di-oFor transformation matrix;
It can also use and be converted based on the VonKries model (also known as diagonal model) in the specific rule of Beer-Lambert.
It itself for internet detection, can be in major network if whole system is constructed using neural network
Present networks are added on network, i.e. addition colour switching network;Problem is changed into the parameter for how estimating transformation matrix.
Colour switching network model construction is as follows:
Initial data input layer → convolutional layer → pond layer → convolutional layer → pond layer → output layer (3 parameters of output, 4 ginsengs
Several or 9 parameter colour switching matrixes).Input layer carries out data prediction, and convolutional layer carries out feature extraction, and pond layer carries out feature
Compression, output layer calculate the element in transformation matrix according to the feature extracted.
By taking 3 parameter transformation matrixes as an example, exported as follows by colour switching network:
Wherein, θr、θg, θbThe respectively feature vector of rgb value in three directions.
In conjunction with the formula in S2, the image data of initial data input layer can be converted to obtain standardization illumination condition
Under image data:
Wherein, c is some Color Channel;θcFor the feature vector on channel c in colour switching matrix;I, j is that image exists
Coordinate in two-dimensional space;For the pixel value of original image preferred coordinates and channel;For respective coordinates and channel after transformation
Pixel value.
According to " a method of for measurement detection system and method neural network based in internet detection " it mentions,
Colour switching network can be trained, and utilizes back-propagation algorithm and gradient descent method Optimal Parameters.
It should be noted that heretofore described model can take the arbitrary number of plies and layer connection structure.Model parameter
Quantity can also be different.Depending on actual conditions.
Basic verifying can also be carried out for being trained to model including model training module.
And model testing module, it tests for the predictive ability to model.
The invention patent the method is bound in any other model and uses.
The picture of generation can be used to train " a kind of for measurement detection system neural network based in internet detection
And method " described in model.
Although having been combined embodiment to be described in detail the present invention, it should be understood by those skilled in the art that
, the present invention is not limited only to specific embodiment, on the contrary, becoming in the various amendments without departing from the application spirit and essence
Shape and replacement are all fallen among the protection scope of the application.
Claims (5)
1. one kind is for versicolor system in internet detection, which is characterized in that including Model Management Service device end sum number
According to management system, Model Management Service device end includes reading data and initialization module, for reading from data management system
Take and initialize original image data, the original image data include: base-line data, conventional application scenarios data or very
Advise application scenarios data;
It further include model construction module, for constructing colour switching network model, the colour switching network model is configured to wrap
Include initial data input layer → convolutional layer → pond layer → convolutional layer → pond layer → output layer, the initial data input layer pair
The original image data carry out data prediction, and treated that data carry out feature extraction to data for the convolutional layer, described
Pond layer carries out Feature Compression to the data after feature extraction, and the output layer goes out transformation matrix according to the feature calculation extracted
In element, thus obtain 3 parameters, 4 parameters or 9 parameters transformation matrix Di-o;
To any one original image data that the reading data and initialization module are read, by it under any illumination condition
Picture under shooting condition, the picture being converted under standard illumination condition;
Conversion formula is as follows:
foutput=Dl-ofinput
Wherein, foutputFor result images function, finputFor input picture function, and Di-oFor transformation matrix.
2. one kind according to claim 1 is used for versicolor system in internet detection,
It is characterized in that, the base-line data refers to: in data acquisition, fixing exterior light photograph, the mobile phone camera when acquisition
Exposure compensating, white balance state, characteristic value, mobile phone posture etc. of test paper glazing correction unit carry out test paper picture shooting, to adopting
The data collected carry out feature extraction to calculate corresponding value as base-line data.
3. according to claim 1 a kind of for versicolor system in internet detection, which is characterized in that described normal
Rule application scenarios data refer to: same test paper, in data acquisition without fixed various control condition factors, it is only necessary to it is each because
Element meets the definition of conventional scenario application, can take pictures, and data obtained pass through final institute after collection apparatus algorithm
The data of acquisition.
4. according to claim 1 a kind of for versicolor system in internet detection, which is characterized in that described non-
Conventional scenario application data refer to: fixed various control condition factors when acquisition make it be unsatisfactory for conventional scenario application conditions, control
Condition element processed includes shooting angle, illumination colour temperature, brightness, camera white balance mode, camera exposure compensation model, is obtained
Data by data obtained after collection apparatus algorithm.
5. according to claim 1 a kind of for versicolor system in internet detection, which is characterized in that pass through institute
It states colour switching network and obtains the transformation matrix D of following 3 parameteri-o:
Wherein, θr、θg, θbThe respectively feature vector of rgb value in three directions.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910043342.4A CN109815875B (en) | 2019-01-17 | 2019-01-17 | System for color transformation in Internet detection |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910043342.4A CN109815875B (en) | 2019-01-17 | 2019-01-17 | System for color transformation in Internet detection |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109815875A true CN109815875A (en) | 2019-05-28 |
CN109815875B CN109815875B (en) | 2023-04-07 |
Family
ID=66603427
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910043342.4A Active CN109815875B (en) | 2019-01-17 | 2019-01-17 | System for color transformation in Internet detection |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109815875B (en) |
Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH0723198A (en) * | 1993-07-01 | 1995-01-24 | Fuji Xerox Co Ltd | Color image forming device |
US5907629A (en) * | 1996-11-15 | 1999-05-25 | Funt; Brian Vicent | Method of estimating chromaticity of illumination using neural networks |
CN1561031A (en) * | 2004-02-20 | 2005-01-05 | ϣ | Interconnected network web information issuing method |
CN101002188A (en) * | 2004-05-05 | 2007-07-18 | 佳能株式会社 | Generation of color measurements from transform-based color profiles and creation of transform-based profiles by a measurement-based color management system |
US20090002545A1 (en) * | 2007-06-29 | 2009-01-01 | Nokia Corporation | Method and system for color management in digital imaging |
CN103065334A (en) * | 2013-01-31 | 2013-04-24 | 金陵科技学院 | Color cast detection and correction method and device based on HSV (Hue, Saturation, Value) color space |
CN103617596A (en) * | 2013-10-12 | 2014-03-05 | 中山大学 | Image color style transformation method based on flow pattern transition |
CN104778466A (en) * | 2015-04-16 | 2015-07-15 | 北京航空航天大学 | Detection method combining various context clues for image focus region |
CN107578453A (en) * | 2017-10-18 | 2018-01-12 | 北京旷视科技有限公司 | Compressed image processing method, apparatus, electronic equipment and computer-readable medium |
CN108985217A (en) * | 2018-07-10 | 2018-12-11 | 常州大学 | A kind of traffic sign recognition method and system based on deep space network |
-
2019
- 2019-01-17 CN CN201910043342.4A patent/CN109815875B/en active Active
Patent Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH0723198A (en) * | 1993-07-01 | 1995-01-24 | Fuji Xerox Co Ltd | Color image forming device |
US5907629A (en) * | 1996-11-15 | 1999-05-25 | Funt; Brian Vicent | Method of estimating chromaticity of illumination using neural networks |
CN1561031A (en) * | 2004-02-20 | 2005-01-05 | ϣ | Interconnected network web information issuing method |
CN101002188A (en) * | 2004-05-05 | 2007-07-18 | 佳能株式会社 | Generation of color measurements from transform-based color profiles and creation of transform-based profiles by a measurement-based color management system |
US20090002545A1 (en) * | 2007-06-29 | 2009-01-01 | Nokia Corporation | Method and system for color management in digital imaging |
CN103065334A (en) * | 2013-01-31 | 2013-04-24 | 金陵科技学院 | Color cast detection and correction method and device based on HSV (Hue, Saturation, Value) color space |
CN103617596A (en) * | 2013-10-12 | 2014-03-05 | 中山大学 | Image color style transformation method based on flow pattern transition |
CN104778466A (en) * | 2015-04-16 | 2015-07-15 | 北京航空航天大学 | Detection method combining various context clues for image focus region |
CN107578453A (en) * | 2017-10-18 | 2018-01-12 | 北京旷视科技有限公司 | Compressed image processing method, apparatus, electronic equipment and computer-readable medium |
CN108985217A (en) * | 2018-07-10 | 2018-12-11 | 常州大学 | A kind of traffic sign recognition method and system based on deep space network |
Non-Patent Citations (4)
Title |
---|
FRANCISCO JAVIER LOPEZ-RUBIO等: "Local color transformation analysis for sudden illumination change detection", 《IMAGE AND VISION COMPUTING》 * |
SATYADHYAN CHICKERUR等: "Color Image Restoration Using Neural Network Model", 《JOURNAL OF UNIVERSAL COMPUTER SCIENCE》 * |
洪亮等: "基于模拟退火算法优化BP神经网络的色彩空间转换", 《包装工程》 * |
袁磊等: "色彩空间转换的BP神经网络模型", 《襄樊学院学报》 * |
Also Published As
Publication number | Publication date |
---|---|
CN109815875B (en) | 2023-04-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110188795B (en) | Image classification method, data processing method and device | |
CN110569695B (en) | Image processing method and device based on loss assessment image judgment model | |
CN108140130B (en) | Image processing method and non-transitory computer-readable medium | |
KR102629380B1 (en) | Method for Distinguishing a Real Three-Dimensional Object from a Two-Dimensional Spoof of the Real Object | |
Tursun et al. | An objective deghosting quality metric for HDR images | |
CN110599395B (en) | Target image generation method, device, server and storage medium | |
JP6489551B2 (en) | Method and system for separating foreground from background in a sequence of images | |
CN112040834A (en) | Eyeball tracking method and system | |
Bellavia et al. | Dissecting and reassembling color correction algorithms for image stitching | |
CN110502986A (en) | Identify character positions method, apparatus, computer equipment and storage medium in image | |
CN102693528B (en) | Noise suppressed in low light images | |
CN108537788B (en) | Camouflage effect evaluation method and device, computer equipment and storage medium | |
JP6840957B2 (en) | Image similarity calculation device, image processing device, image processing method, and recording medium | |
CN110493595A (en) | The detection method and device of camera, storage medium and electronic device | |
Crispell et al. | Pix2face: Direct 3d face model estimation | |
JP7463186B2 (en) | Information processing device, information processing method, and program | |
CN115393404A (en) | Double-light image registration method, device and equipment and storage medium | |
CN113628134B (en) | Image noise reduction method and device, electronic equipment and storage medium | |
Chen et al. | Face super resolution based on parent patch prior for VLQ scenarios | |
CN112330671A (en) | Method and device for analyzing cell distribution state, computer equipment and storage medium | |
CN109815875A (en) | One kind is for versicolor system in internet detection | |
CN109886923A (en) | It is a kind of for internet detection in measurement detection system and method based on machine learning | |
CN109829848A (en) | A kind of system and method for Image space transformation neural network based in internet detection | |
CN113191376A (en) | Image processing method, image processing device, electronic equipment and readable storage medium | |
CN109831352B (en) | Detection sample generation system and method based on countermeasure generation network for Internet detection |
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 |