CN102170516B - Color space transition method based on fuzzy logic and neural network - Google Patents
Color space transition method based on fuzzy logic and neural network Download PDFInfo
- Publication number
- CN102170516B CN102170516B CN 201110086627 CN201110086627A CN102170516B CN 102170516 B CN102170516 B CN 102170516B CN 201110086627 CN201110086627 CN 201110086627 CN 201110086627 A CN201110086627 A CN 201110086627A CN 102170516 B CN102170516 B CN 102170516B
- Authority
- CN
- China
- Prior art keywords
- color space
- fuzzy
- point
- model
- subspace
- 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.)
- Expired - Fee Related
Links
- 238000000034 method Methods 0.000 title claims abstract description 31
- 238000013528 artificial neural network Methods 0.000 title abstract description 9
- 230000007704 transition Effects 0.000 title abstract 6
- 238000006243 chemical reaction Methods 0.000 claims abstract description 40
- 230000001537 neural effect Effects 0.000 claims description 44
- 238000012549 training Methods 0.000 claims description 22
- 230000006870 function Effects 0.000 claims description 18
- 230000009466 transformation Effects 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000011478 gradient descent method Methods 0.000 claims description 4
- 238000012886 linear function Methods 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 239000007787 solid Substances 0.000 claims description 3
- 238000012795 verification Methods 0.000 claims description 3
- 238000005070 sampling Methods 0.000 abstract description 4
- 238000012545 processing Methods 0.000 abstract description 3
- 210000005036 nerve Anatomy 0.000 abstract 2
- 230000000295 complement effect Effects 0.000 abstract 1
- 230000004075 alteration Effects 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000002068 genetic effect Effects 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 238000002620 method output Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000011002 quantification Methods 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Images
Landscapes
- Color Image Communication Systems (AREA)
- Image Processing (AREA)
- Facsimile Image Signal Circuits (AREA)
Abstract
The invention discloses a color space transition method based on the fuzzy logic and the neural network, comprising steps of selecting sampling points of input color space, collecting modeling data of the sampling points, creating a fuzzy nerve color space transition model, and inputting the modeling data of the sampling points into the fuzzy nerve color space transition model to accomplish the transition of the color space. The color space transition method based on the fuzzy logic and the neural network of the invention makes full use of fuzzy identification on advantage of rapidity in WAN firstly, and the input color space is divided into a plurality of subspaces; then the invention also makes full use of neural network on advantages of strong ability and high precision on self-adaption and identification in local subspace and reasonably achieves the complementary advantages between fuzzy and neural network arithmetic, thereby greatly improving the processing speed of model conversion and effectively improving the precision of model conversion.
Description
Technical field
The invention belongs to the printing color management technical field, relate to a kind of method of color space conversion, be specifically related to a kind of color space changover method based on fuzzy theory and neural net.
Background technology
Color Management Technology is in order to solve the correct branch problem of color between different imaging devices, and one of core of Color Management Technology is exactly the mutual conversion of color space model.At present, color space changover method commonly used mainly contains knob Jie fort equation method, three dimensional lookup table (3D LUT), polynomial regression method, and some artificial intelligence approaches, such as artificial neural net, fuzzy logic and genetic algorithm etc.Chroma space belongs to three-dimensional problem of nonlinear mapping, and traditional mathematical model method has certain limitation, is difficult to obtain to use in control is adjusted automatically.Therefore, at present a lot of scholars transfer to research emphasis in the intelligent algorithm research.
For the research of artificial neural network algorithm color space conversion model, often adopt at present the BP neural net.The BP neural network algorithm is essentially gradient descent method, from mathematical angle, the BP algorithm is a kind of optimization method of Local Search, but the problem that will solve when it is when finding the solution the global optimization of complex nonlinear function, algorithm probably is absorbed in local extremum, make failure to train, therefore, its model conversation method is still waiting further raising.Based on the natural language description commonly used of color in the life, have ambiguity, in addition, color notation conversion space has the characteristics of nonlinearity, so people are applied to the color space conversion model with fuzzy control theory.Compare with modern control theory with classical control theory, the main feature of fuzzy control is the Mathematical Modeling that does not need to set up object, and structure is easy, strong robustness, and algorithm is simple, carries out soon, easily realizes etc.Yet in fuzzy model, the selection of fuzzy rule and membership function fully by rule of thumb, in addition, the simple Fuzzy Processing of information will cause the control precision of system to reduce and the dynamic quality variation, improve precision and then must increase quantification progression, thereby cause the rule search expanded range.
Summary of the invention
The purpose of this invention is to provide a kind of color space changover method based on fuzzy theory and neural net, solved the problem that is not suitable for the global optimization of finding the solution the complex nonlinear function that existing color space changover method exists.
The technical solution adopted in the present invention is based on the color space changover method of fuzzy theory and neural net, specifically to implement according to following steps:
Step 1: select the sampled point of input color space, gather the modeling data of sampled point;
Step 2: set up fuzzy neural color space transformation model, the modeling data of the sampled point that step 1 is obtained is input in the fuzzy neural color space transformation model, finishes the conversion of color space.
Characteristics of the present invention also are,
Step 1 is wherein selected the sampled point of input color space, gather the modeling data of sampled point, specifically implement according to following steps: the RGB color space is divided into 27 sub spaces, select cubical central point as the check post of verification model precision, 27 altogether, R, G, B color axis value are respectively 43,129 and 213; Select 729 sample points in the RGB color space, the selection of sampled point is evenly to get a little in the RGB color space, and with R, G, B color axis eight equal parts, the R of training sample, G, B value are taken as respectively 0,32,64,96,128,160,192,224 and 255, have 729 groups of data.
Step 2 is wherein set up fuzzy neural color space transformation model, specifically implements according to following steps:
A. at first, the color value of input RGB color space conversion point is defined as X(r, g, b);
B. set RGB subspace radius, be defined as δ, the initial value of setting the fuzzy subspace radius δ of division is 5;
C. input RGB color space and corresponding CIEL
*a
*b
*Color space sample set matrix;
D. establish centered by the X point, calculate X
Left(r-δ, g-δ, b-δ) and X
RightIf the coordinate of (r+ δ, g+ δ, b+ δ) point is X
LeftAnd X
RightPoint exceeds RGB color space, then X
LeftAnd X
RightSpot projection is to the surface of RGB colour solid;
E. adopt color space conversion fuzzy model output X
LeftAnd X
RightCorresponding point coordinates corresponding to CIEL*a*b* color space is defined as Y
Left(L
1, a
1, b
1) and Y
Right(L
2, a
2, b
2);
F. both ask for take X as the center of circle, δ is the sampled point in the RGB subspace of radius, again with Y
LeftY
RightSample set for the sampled point in the CIEL*ab color sub-spaces of diameter;
If g. the quantity of sample set intermediate samples point is very few, then δ=δ+5, again calculation procedure d~f;
If h. the quantity of sample set intermediate samples point is too much, then δ=δ-1, again calculation procedure d~f;
If the reasonable quantity of sample set intermediate samples point i. is then at RGB color sub-spaces and CIEL
*a
*b
*Adopt neural net in the subspace, utilize subspace sample point training network, the color space conversion model of the built-in BP neural net that is based in the subspace;
J. input the X coordinate figure, utilize the color space conversion model of subspace BP neural net, the CIEL that output X is ordered
*a
*b
*The coordinate figure Y(L in space, a, b), finish color space conversion.
Color space conversion fuzzy model among the step e wherein adopts three inputs, three output fuzzy controllers.
Neural net in the step I wherein, adopt 4 layers of BP network configuration, comprise an output layer and three hidden layers, the nodes of each hidden layer is 20, the neural transferring function of network hidden layer adopts log-sigmoid type function logsig(), the output layer neural transferring function adopts pure linear function purelin(), the training function adopts elasticity gradient descent method trainrp function, set correct training parameter, wherein maximum frequency of training is selected 1000 times, training precision selects 1, and learning rate is set as 0.2.
The quantity of sample set intermediate samples point wherein is between 50 and 100.
The invention has the beneficial effects as follows, at first given full play to the strong advantage of the rapidity of fuzzy recognition in the wide area space, will input color space and dynamically be divided into some subspaces; Then given full play to again the strong and high advantage of precision of neural net Adaptive Identification ability in Local Subspace, rationally realized mutual supplement with each other's advantages fuzzy and neural network algorithm, both greatly improved the processing speed of model conversion, again Effective Raise the precision of model conversion.
Description of drawings
Fig. 1 is with 27 check color lump value input models, the aberration graph of a relation of model output valve and color lump actual measured value in the inventive method;
Fig. 2 be in the inventive method with the RGB color value input model of 729 modeling color lumps, the chromaticity difference diagram of model output valve and color lump actual measured value;
Fig. 3 is the program flow diagram of modeling in the inventive method;
Fig. 4 is the graph of a relation of forgiving modeling point quantity and the average color difference of check color lump in the subspace of model in the inventive method.
Embodiment
The present invention is described in detail below in conjunction with the drawings and specific embodiments.
The present invention is based on the color space changover method of fuzzy theory and neural net, specifically implement according to following steps:
Step 1: select the sampled point of input color space, gather the modeling data of sampled point.
The check color lump: the RGB color space is divided into 27 sub spaces (i.e. 27 sub-cubes), selects cubical central point as the check post of verification model precision, 27 altogether, R, G, B color axis value are respectively 43,129 and 213.
The modeling color lump: select 729 sample points in the RGB color space, the selection of sampled point is evenly to get a little in the RGB color space, and with R, G, B color axis eight equal parts, the R of training sample, G, B value are taken as respectively 0,32,64,96,128,160,192,224 and 255, total 729(is 9
3) the group data.
Step 2: set up fuzzy neural color space transformation model, the modeling data of the sampled point that step 1 is obtained is input in the fuzzy neural color space transformation model, finishes the conversion of color space.
In order to improve the precision of color space conversion, utilize respectively fuzzy model and neural net characteristics separately, this patent proposes at first to adopt the color space conversion fuzzy model that color space is divided into some subspaces, the inner method output sampled point that adopts the BP neural net in the subspace.The program circuit of model is as shown in Figure 3:
(1) at first, the color value of input RGB color space conversion point is defined as X(r, g, b).
(2) set RGB subspace radius, be defined as δ, in model, the initial value of at first setting the fuzzy subspace radius δ of division is 5.
(3) input RGB color space and corresponding CIEL
*a
*b
*Color space sample set matrix.
(4) establish centered by the X point, calculate X
Left(r-δ, g-δ, b-δ) and X
RightThe coordinate of (r+ δ, g+ δ, b+ δ) point.If X
LeftAnd X
RightPoint exceeds RGB color space, then X
LeftAnd X
RightSpot projection is to the surface of RGB colour solid.
(5) adopt color space conversion fuzzy model output X
LeftAnd X
RightCorresponding CIEL
*a
*b
*The point coordinates that color space is corresponding is defined as Y
Left(L
1, a
1, b
1) and Y
Right(L
2, a
2, b
2).The color space conversion fuzzy model adopts three inputs, three output fuzzy controllers.
(6) both ask for take X as the center of circle, δ is the sampled point in the RGB subspace of radius, again with Y
LeftY
RightCIEL for diameter
*a
*b
*The sample set of the sampled point in the color sub-spaces.
(7) if the quantity of sample set intermediate samples point is very few, then δ=δ+5, again calculation procedure (4)~(6).
(8) if the quantity of sample set intermediate samples point is too much, then δ=δ-1, again calculation procedure (4)~(6).
(9) if the reasonable quantity of sample set intermediate samples point, then at RGB color sub-spaces and CIEL
*a
*b
*Adopt neural net in the subspace, utilize subspace sample point training network, the color space conversion model of the built-in BP neural net that is based in the subspace.Neural net adopts 4 layers of BP network configuration (output layer and 3 hidden layers), the nodes of each hidden layer is 20, the neural transferring function of network hidden layer adopts log-sigmoid type function logsig(), the output layer neural transferring function adopts pure linear function purelin(), the training function adopts elasticity gradient descent method trainrp function.Set correct training parameter, wherein maximum frequency of training is selected 1000 times, and training precision selects 1, and learning rate is set as 0.2.
(10) input the X coordinate figure, utilize the color space conversion model of subspace BP neural net, the CIEL that output X is ordered
*a
*b
*The coordinate figure Y(L in space, a, b), finish color space conversion.
For guaranteeing the conversion precision of BP neural net in the subspace, determine sampled point quantity in the model subspace, namely how many sampled point guarantee model conversation precision the subspace has.For this reason, select 729 sample points in the RGB color space, the selection of sampled point is evenly to get a little in the RGB color space, and with R, G, B color axis eight equal parts, the R of training sample, G, B value are taken as respectively 0,32,64,96,128,160,192,224 and 255, total 729(is 9
3) the group data, measure its L
*, a
*, b
*Value.Input point x(r in rgb space, g, b), calculate the distance between x point and the color space sampled point, select respectively the sampled point 15,20 minimum with x point distance, 30,40,50,60,70,80,90,100,110 and 120 points are take the x point as the center of circle, forgive respectively this 15,20,30,40,50,60,70,80,90,100, the space of 110 and 120 modeling points is radius, RGB is inputted color space dynamically be divided into some subspaces according to the coordinate of input point, in the subspace, adopt above-mentioned BP Artificial Neural Network Structures, set up the BP neural net color space conversion model of dividing based on dynamic subspace.Fig. 4 is the graph of a relation of forgiving modeling point quantity and the average color difference of check color lump in the subspace of model in the inventive method, as seen from Figure 4, increase model conversion precision with sampled point in the subspace improves constantly, when bringing up to certain precision, the model conversion precision tends to be steady, when comprise in the subspace sampling number greater than 50 less than 120 the time, the model conversion precision is between 1.6 to 1.8, increasing the subspace sample points can cause BP neural model training difficulty to increase again, training time prolongs, even the network training failure, therefore, can adopt in the subspace sampled point quantity between 50 and 100, as the zone of reasonableness of subspace sampled point quantity.
Model accuracy is analyzed:
Incoming inspection point is after model conversation, and checking the conversion average color difference of color lumps for 27 is 1.92, and maximum aberration is 2.84, and minimum aberration is 0.95, and its aberration distribution map as shown in Figure 1.
With 729 modeling color value input models, the average color difference of model output valve and actual measured value is 1.82, and wherein maximum aberration is 8.78, and minimum aberration is 0.13, its aberration distribution map as shown in Figure 2, its most of color lump value of chromatism is less than 4.0.
By the fuzzy division to the input color sub-spaces, in fuzzy subspace, adopt BP neural network model output color value, its model output accuracy is better than independent BP neural model and fuzzy model, and the increase of sample point quantity can not affect the conversion speed of model, can be higher if continue the conversion precision of its model of increase sample point quantity.
Claims (4)
1. based on the color space changover method of fuzzy theory and neural net, it is characterized in that, specifically implement according to following steps:
Step 1: select the sampled point of input color space, gather the modeling data of sampled point;
Step 2: set up fuzzy neural color space transformation model, the modeling data of the sampled point that step 1 is obtained is input in the fuzzy neural color space transformation model, finishes the conversion of color space; The described fuzzy neural color space transformation model of setting up, specifically implement according to following steps:
A. at first, the color value of input RGB color space conversion point is defined as X(r, g, b);
B. set RGB subspace radius, be defined as δ, the initial value of setting the fuzzy subspace radius δ of division is 5;
C. input RGB color space and corresponding CIEL
*a
*b
*Color space sample set matrix;
D. establish centered by the X point, calculate X
Left(r-δ, g-δ, b-δ) and X
RightIf the coordinate of (r+ δ, g+ δ, b+ δ) point is X
LeftAnd X
RightPoint exceeds RGB color space, then X
LeftAnd X
RightSpot projection is to the surface of RGB colour solid;
E. adopt color space conversion fuzzy model output X
LeftAnd X
RightCorresponding point coordinates corresponding to CIEL*a*b* color space is defined as Y
Left(L
1, a
1, b
1) and Y
Right(L
2, a
2, b
2);
F. both ask for take X as the center of circle, δ is the sampled point in the RGB subspace of radius, again with Y
LeftY
RightSample set for the sampled point in the CIEL*ab color sub-spaces of diameter;
If g. the quantity of sample set intermediate samples point is very few, then δ=δ+5, again calculation procedure d~f;
If h. the quantity of sample set intermediate samples point is too much, then δ=δ-1, again calculation procedure d~f;
If the reasonable quantity of sample set intermediate samples point i. is then at RGB color sub-spaces and CIEL
*a
*b
*Adopt neural net in the subspace, utilize subspace sample point training network, the color space conversion model of the built-in BP neural net that is based in the subspace; Described neural net, adopt 4 layers of BP network configuration, comprise an output layer and three hidden layers, the nodes of each hidden layer is 20, the neural transferring function of network hidden layer adopts log-sigmoid type function logsig(), the output layer neural transferring function adopts pure linear function purelin(), the training function adopts elasticity gradient descent method trainrp function, set correct training parameter, wherein maximum frequency of training is selected 1000 times, training precision selects 1, and learning rate is set as 0.2;
J. input the X coordinate figure, utilize the color space conversion model of subspace BP neural net, the CIEL that output X is ordered
*a
*b
*The coordinate figure Y(L in space, a, b), finish color space conversion.
2. the color space changover method based on fuzzy theory and neural net according to claim 1, it is characterized in that, described step 1 is selected the sampled point of input color space, gather the modeling data of sampled point, specifically implement according to following steps: the RGB color space is divided into 27 sub spaces, select cubical central point as the check post of verification model precision, 27 altogether, R, G, B color axis value are respectively 43,129 and 213; Select 729 sample points in the RGB color space, the selection of sampled point is evenly to get a little in the RGB color space, and with R, G, B color axis eight equal parts, the R of training sample, G, B value are taken as respectively 0,32,64,96,128,160,192,224 and 255, have 729 groups of data.
3. the color space changover method based on fuzzy theory and neural net according to claim 1 is characterized in that, the color space conversion fuzzy model among the described step e adopts three inputs, three output fuzzy controllers.
4. the color space changover method based on fuzzy theory and neural net according to claim 1 is characterized in that, the quantity of described sample set intermediate samples point is between 50 and 100.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN 201110086627 CN102170516B (en) | 2011-04-07 | 2011-04-07 | Color space transition method based on fuzzy logic and neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN 201110086627 CN102170516B (en) | 2011-04-07 | 2011-04-07 | Color space transition method based on fuzzy logic and neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN102170516A CN102170516A (en) | 2011-08-31 |
CN102170516B true CN102170516B (en) | 2013-03-06 |
Family
ID=44491477
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN 201110086627 Expired - Fee Related CN102170516B (en) | 2011-04-07 | 2011-04-07 | Color space transition method based on fuzzy logic and neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN102170516B (en) |
Families Citing this family (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102514963A (en) * | 2011-11-23 | 2012-06-27 | 上海电力学院 | Correcting control method with function of automatically identifying colors |
CN102998976A (en) * | 2012-11-14 | 2013-03-27 | 东南大学 | Online real-time control method for intelligent seismic reduction structure |
CN103354073B (en) * | 2013-06-13 | 2016-01-20 | 南京信息工程大学 | A kind of LCD color deviation correction method |
CN104918030A (en) * | 2015-06-05 | 2015-09-16 | 河海大学 | Color space conversion method based on ELM extreme learning machine |
CN110035268B (en) * | 2019-04-08 | 2021-01-26 | 深圳市帧彩影视科技有限公司 | Color space conversion method and equipment based on fuzzy inference |
CN110310266A (en) * | 2019-06-26 | 2019-10-08 | 江苏理工学院 | A kind of acetes chinensis method based on T-S fuzzy neural network |
CN110378393A (en) * | 2019-06-26 | 2019-10-25 | 江苏理工学院 | A kind of mixing printing product acetes chinensis method based on PSO-GSA-SVM |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1419361A (en) * | 2002-10-31 | 2003-05-21 | 上海交通大学 | Intelligent monitoring administration method for picture on network |
-
2011
- 2011-04-07 CN CN 201110086627 patent/CN102170516B/en not_active Expired - Fee Related
Also Published As
Publication number | Publication date |
---|---|
CN102170516A (en) | 2011-08-31 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN102170516B (en) | Color space transition method based on fuzzy logic and neural network | |
CN106295799B (en) | A kind of implementation method of deep learning multilayer neural network | |
CN107679522B (en) | Multi-stream LSTM-based action identification method | |
CN108081266A (en) | A kind of method of the mechanical arm hand crawl object based on deep learning | |
CN108304826A (en) | Facial expression recognizing method based on convolutional neural networks | |
CN105389843B (en) | Global illumination real-time rendering method based on radial basis function neural network fitting | |
CN104270816B (en) | The adaptive dynamic fingerprint base construction method of LED visible light indoor locating system | |
CN108446794A (en) | One kind being based on multiple convolutional neural networks combination framework deep learning prediction techniques | |
CN106683171A (en) | GPU multi-thread texture mapping SystemC modeling structure | |
CN115019510B (en) | Traffic data restoration method based on dynamic self-adaptive generation countermeasure network | |
CN103530700B (en) | Urban distribution network saturation loading Comprehensive Prediction Method | |
CN108304925A (en) | A kind of pond computing device and method | |
Zhu et al. | Change detection based on the combination of improved SegNet neural network and morphology | |
CN103354073B (en) | A kind of LCD color deviation correction method | |
CN107871034A (en) | Tolerance assignment multi-objective optimization design of power method based on mutative scale learning aid algorithm | |
CN110555189B (en) | Spatial interpolation method based on reverse computing thinking | |
CN109902807A (en) | A kind of hot modeling method of many-core chip distribution formula based on Recognition with Recurrent Neural Network | |
CN106780722B (en) | A kind of differential mode scale Forest Scene construction method of the same race and system | |
Zou et al. | [Retracted] Dynamic Modeling and Simulation Analysis of Virtual Product Design and Creation Based on OSG Artificial Intelligence Platform | |
CN111144290A (en) | Multi-view-based flight situation accurate sensing method | |
CN106021177B (en) | A kind of laser scanning data compressing method based on scan line | |
CN109635848A (en) | A kind of deep learning method based on feature clustering | |
Zhang et al. | Colour correction method of interior decoration engineering based on dense convolution neural network | |
Luo et al. | Study on deep learning political culture communication system in Universities under the perspective of postmodern media | |
CN109922133A (en) | Parameterized treatment method towards electric system load unloading |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20130306 |