CN103354073B - A kind of LCD color deviation correction method - Google Patents

Abstract

The present invention proposes a kind of LCD color deviation correction method.Described method introduces the BP neural network based on genetic algorithm, genetic algorithm carries out global search fast to the initial weight of neural network and threshold value, solve the shortcoming that traditional BP algorithm is easily absorbed in local minimum, accelerate speed of convergence, improve the generalization ability of network; Adopt Lab color space value as the input and output value of neural network, than directly adopting RGB color space vector to have better quantitative classification accuracy, automatically can correct LCD color error ratio, making LCD have better color re-production capabilities.

Description

A kind of LCD color deviation correction method

Technical field

The invention belongs to display technique field, relate to a kind of LCD color deviation correction method.

Background technology

Along with the development of display technique, LCDs is widely used in various electronic equipment.The color requirements of people to LCDs is more and more higher.Chromatic rendition is an important indicator of evaluating liquid crystal display screen, but LCD display is under different photoenvironments, and the color of the image gathered and object itself will be caused to there is certain deviation; Under different hardware device environment, input and output image also can be caused to occur color error ratio.How color error ratio is corrected, become a focus of current research.

LCD display color adopts RGB color space to define usually.But RGB color space and equipment closely related, the color of different equipment display has difference, directly using RGB color vector as input, desirable treatment effect cannot be obtained.Lab color space has device independence, this color space reflect preferably colour vision impression, its aberration calculated and human eye visual perception basically identical.

BP neural network is widely applied in color correction field in recent years.BP neural network is a kind of Multi-layered Feedforward Networks by Back Propagation Algorithm training, its learning rules use gradient descent method, constantly adjusted the weights and threshold of network by backpropagation, make the error sum of squares of network minimum, thus arrive the object of anticipation learning.But in the process of study, there will be crenellated phenomena, be even absorbed in Local Minimum.

Summary of the invention

The present invention is directed to above-mentioned technical matters, propose a kind of LCD color deviation correction method.Described method adopts the GA-BP neural network based on genetic algorithm and BP algorithm, automatically can correct, make LCD have better color re-production capabilities to LCD color error ratio.

The present invention, for solving the problems of the technologies described above, adopts following technical scheme:

A kind of LCD color deviation correction method, comprises the steps:

Step one, with redness, green and blueness for establishment of coordinate system RGB three-dimensional color space, and chooses training sample in this space;

Step 2, choosing the output calibration value of one piece of standard screen for obtaining BP neural network, selecting one piece of screen to be calibrated else for obtaining the input value of BP neural network;

Step 3, exports training sample through screen to be calibrated, uses digital camera to obtain its RGB three-dimensional color space color value V _rGB, and this value is transformed into Lab color space value V _lab, as the input value of BP neural network;

Step 4, exports training sample through standard screen, uses digital camera to obtain its RGB three-dimensional color space RGB color space value V ' _rGB, and this value is transformed into Lab color space value V ' _lab, as the output calibration value of BP neural network;

Step 5, creates BP neural network, adopts the BP neural network with three input neurons, three output neurons and five intrerneurons;

Step 6, is optimized the initial weight of BP neural network and threshold value by genetic algorithm; Obtain the BP neural network with best initial weights and threshold value;

Step 7, training BP neural network: by V _labas the input of BP neural network, V ' _labas the output of BP neural network, according to predetermined excitation function, training function, maximum training iterations and neural metwork training permissible error, training BP neural network;

Step 8, is cured in the hardware configuration of LCD display by the BP neural network trained, and realizes LCD display color and corrects function.

Beneficial effect of the present invention: the present invention proposes a kind of LCD color deviation correction method.Described method introduces the BP neural network based on genetic algorithm, genetic algorithm carries out global search fast to the initial weight of neural network and threshold value, solve the shortcoming that traditional BP algorithm is easily absorbed in local minimum, accelerate speed of convergence, improve the generalization ability of network; Adopt Lab color space value as the input and output value of neural network, than directly adopting RGB color space vector to have better quantitative classification accuracy, automatically can correct LCD color error ratio, making LCD have better color re-production capabilities.

Accompanying drawing explanation

Fig. 1 is a kind of LCD color deviation correction method workflow diagram of the present invention.

Fig. 2 is the GA-BP neural network structure figure that the present invention adopts.

Fig. 3 is the BP neural network workflow diagram that genetic algorithm is improved.

Fig. 4 is aberration correlation curve before and after correcting.

Embodiment

Below in conjunction with the drawings and specific embodiments, illustrate a kind of LCD color deviation correction method that the present invention proposes further.

As shown in Figure 1, a kind of LCD color deviation correction method workflow diagram of the present invention, its detailed process is as follows:

Step one: RGB color space is uniformly distributed the little cubic space of 10*10*10 by the present invention, total obtains 1000 samples.And pick out from 1000 samples representative 400 as train samples.

Step 2: select one piece of screen to be calibrated and one piece of standard screen, use high-definition digital camera to show image to LCD and take, obtain its color value respectively, be designated as RGB.Operation platform is WindowsXP, and development language is VC++6.0, based on OpenGL shape library.

Step 3: screen color character vector to be calibrated extracts and standardization, by CIE1976LAB colour difference formula, RGB color space value is converted into Lab color space value, is designated as Lab.

First by RGB(0-255) color space value, be converted into XYZ (CIE1931 standard colorimetric system) color space value:

$\left\{\begin{array}{c}X=0.412453*R+0.357580*G+0.180423*B\\ Y=0.212671*R+0.715160*G+0.072169*B\\ Z=0.019334*R+0.119193*G+0.950227*B\end{array}\right.---\left(1\right)$

Then be normalized:

$\left\{\begin{array}{c}x=X/(255*0.950456)\\ y=Y/255\\ z=Z/(255*1.088754)\end{array}\right.---\left(2\right)$

Secondly the xyz value obtained is converted to Lab color space value:

$\left\{\begin{array}{c}L=115*y^{1/3},y>0.00856\\ L=903.3*y,y<=0.00856\end{array}\right.---\left(3\right)$

$\left\{\begin{array}{c}a=500*(f\left(x\right)-f\left(y\right))\\ b=200*(\left(f\left(y\right)\right)-f\left(z\right))\end{array}\right.---\left(4\right)$

Wherein f (t)=t ^1/3, work as t>0.008856; F (t)=7.787t+16/116, works as t<=0.008856

Partial data is as shown in table 1, RGB color space value of comparing, and Lab has obvious numerical values recited and positive and negative differentiation.

Table 1 Color Characteristic

Color	R	G	B	L	a	b
							Red	255	0	0	69	80	67
Yellow	255	255	0	112	-22	94
							Blue	0	0	255	48	79	-109
Green	0	255	0	103	-86	83
							In vain	245	245	245	113	-2	0
Ash	124	125	128	91	0.1	-1

Orange	255	128	64	96	18	35
							Purple	128	0	255	65	89	-79
Palm fibre	128	64	0	76	10	66
							Black	8	7	3	35	-2	12

Step 4: treat that standard screen color character vector extracts and standardization, method, as described in step 3, obtains Lab value, is designated as L'a'b'.

Step 5: create initial BP neural network: the BP neural network of the present invention's design adopts 3-5-3 structure, and its topological structure as shown in Figure 2.This structure has 3 input neurons and three output neurons.Lab color space value is as 3 inputs of BP neural network, and L'a'b' color space value is as 3 outputs of neural network, and hidden layer neuron number there is no theoretic guidance, and according to many experiments, the present invention chooses 5 nodes.Excitation function adopts tansigmoid; Training function adopts trainbpx, and utilize rapid bp algorithm Training Multilayer Neural Network, namely have employed momentum or adaptive learning, can reduce the training time, maximum training iterations is within 300, and network training permissible error is set to 0.03.

Step 6: initial weight and threshold value with genetic algorithm optimization BP neural network: GA-BP workflow as shown in Figure 3.By genetic algorithm, global search is fast carried out to initial weight and threshold value, the chromosome of genetic algorithm is exactly connection weights or the threshold value of BP network, the specific mode of each chromosome is encoded, the satisfactory solution of genetic algorithm is exactly the approximate value of the weights and threshold of BP neural network, the best initial weights threshold value of acquisition is assigned to the BP network not yet starting to train.

(1) coding and initialization colony: coded system common in genetic algorithm is 2 scale codings and real coding.Compare 2 scale codings, it is high that real coding has precision, the advantage that speed of searching optimization is fast, and the present invention adopts real coding mode.Input Lab value is converted into real number string.

(2) determination of fitness function, this function is the unique information of genetic algorithm guidance search, and choosing of it determines algorithm quality.This function should be easy to convergence and calculates and can not cause locally optimal solution.

(3) selection, intersection, mutation genetic operation

1, select: the individuality of selecting vitality strong from old population produces the process of new population, the present invention adopts that algorithm of tournament selection method is namely at every turn random selects some individualities from population, then preferably wins as father is individual, constantly repeats until produce N number of individuality.

2, intersect: two chromosomes matched by the mutual switching part gene of certain mode, thus form new individuality.The introducing of crossover operator, determines the ability of searching optimum of genetic algorithm.The present invention adopts middle restructuring:

T ₂=S ₂+λ ₂(S ₁-S ₂)(5)

T ₁=S ₁+λ ₁(S ₂-S ₁)(6)

T in formula ₁for son individuality 1, T ₂for son individuality 2, S ₁for father's individuality 1, S ₂for father's individuality 2, λ ₁, λ ₂for scale factor.

3, make a variation: in biological heredity process, can gene mutation be caused in accidentalia situation, thus occur new species.Introducing mutation operator produces new individual, determines the local search ability of genetic algorithm.The present invention adopts with definitive variation probability random selecting several body from population, and for the individuality chosen, the number on the overall position that need make a variation at random between generation [0-9] is to replace legacy data.

Step 7: training GA-BP neural network: train the BP neural network of weights and threshold after genetic algorithm optimization with training sample (training sample that step one mode obtains).After meeting network training error, training process completes automatically.

Step 8: by the GA-BP neural network that trains by dsp chip, be cured in the hardware configuration of LCD display, realizes LCD display color and corrects function.

Fig. 4 is the comparison diagram of value of chromatism before and after correcting, and horizontal ordinate is the part sample chosen, and ordinate is value of chromatism.Prove that algorithm of the present invention has good correcting feature to LCD image color by experiment.

Simulation result shows, this algorithm ability of searching optimum is strong, and fast convergence rate, overcomes the defect that BP neural network is easily absorbed in Local Minimum, indicates feasibility and the validity of the method.

Claims (1)

1. a LCD color deviation correction method, is characterized in that, comprises the steps:

Step one, with redness, green and blueness for establishment of coordinate system RGB three-dimensional color space, and chooses training sample in this space;

Step 2, choosing the output calibration value of one piece of standard screen for obtaining BP neural network, selecting one piece of screen to be calibrated else for obtaining the input value of BP neural network;

Step 3, exports training sample through screen to be calibrated, uses digital camera to obtain its RGB three-dimensional color space color value V _rGB, and use CIE1976LAB colour difference formula, this value is converted to Lab color space value V _lab, as the input value of BP neural network;

Step 4, exports training sample through standard screen, uses digital camera to obtain its RGB three-dimensional color space color value V ' _rGB, and use CIE1976LAB colour difference formula, this value is converted to Lab color space value V ' _lab, as the output calibration value of BP neural network;

Step 5, creates BP neural network, adopts the BP neural network with three input neurons, three output neurons and five intrerneurons;

Step 6, is optimized the initial weight of BP neural network and threshold value by genetic algorithm; Obtain the BP neural network with best initial weights and threshold value;

Step 7, training BP neural network: by V _labas the input of BP neural network, V ' _labas the output of BP neural network, according to predetermined excitation function, training function, maximum training iterations and neural metwork training permissible error, training BP neural network;

Step 8, is cured in the hardware configuration of LCD display by the BP neural network trained, and realizes LCD display color and corrects function.