A display spectral characterization method based on support vector machines

Abstract

This invention discloses a display spectral characterization method based on support vector machines, belonging to the field of color science and technology. The method includes analyzing the principal components of the display's emission spectrum, further determining the number of principal components, scaling the principal component transformation matrix to obtain the principal component decomposition matrix and reconstruction matrix; using the principal component decomposition matrix and driving value spectra to obtain spectral principal component representations, establishing a regression theoretical model of color driving values ​​and spectral principal components; applying the driving values ​​and spectral principal component training set to design a parameter optimization algorithm to obtain a prediction model; based on the model's prediction of the corresponding emission spectrum principal components, applying the spectral principal component reconstruction matrix to perform inverse principal component transformation for spectral reconstruction; and finally, evaluating the spectral reconstruction accuracy. This invention utilizes support vector machine regression and principal component dimensionality reduction methods, exhibiting good prediction accuracy, robustness, and low redundancy, providing an effective method for display spectral characterization.

Description

Technical Field

[0001] This invention relates to the field of color science and color image display technology, specifically to a display spectral characterization method based on support vector machines. Background Technology

[0002] Display spectral characterization is a key issue in color information management. Its purpose is to establish the conversion relationship between the display's RGB driving signals and the display's output spectral radiance. Modeling the RGB driving signals and spectral radiance involves mapping the device-dependent color space (RGB space) to a device-independent spectral space using a spectral characterization model. Currently, there are several main types of display spectral characterization:

[0003] In 2011, Liu Haoxue et al. proposed a spectral radiance partitioning model (SPRM) based on spectral superposition. This model offers a method for calculating the spectral radiance function of the three primary colors using wavelength partitioning. It fits the measured spectra of the RGB primary colors to a cubic polynomial relationship between the spectral radiance values ​​and the digital drive values. Using this relationship, the spectral radiance curves of the three primary colors corresponding to any digital drive value can be calculated. Then, the display color under any digital drive value can be calculated from the spectral radiance curves based on the color additive relationship. However, this model cannot predict backwards, and the prediction error is large when the RGB values ​​are small, limiting its application.

[0004] In 2020, Zhang Xiaohui et al. used grayscale data to train a PLCC model (Piecewise Linear Interpolation assuming Constant Chromaticity) to perform spectral characterization of liquid crystal displays. Whether considered from a forward or inverse perspective, the accuracy of the grayscale-trained PLCC model is higher than that of the SRPPM (Segmented Spectral Radiance Partition Model) and the GOG (Gain-Bias-Gamma) model, and the inverse operation of the PLCC model is much simpler than that of the SRPPM. However, it requires establishing a spectral characterization model based on the assumptions of channel independence and constant chromaticity. The gain and gamma of the three channels of the display in the PLCC model need to be obtained by establishing a luminance lookup table and piecewise linear interpolation, making model establishment relatively complex.

[0005] In 2020, Ma Xiangcai et al. proposed a method for spectral characterization of LCD displays based on visual characteristics. They utilized principal component analysis (PCA) to reduce the dimensionality of spectral data and employed an RBF neural network to investigate the impact of the input variable range, three different visual weighting functions, and the number of colors on the accuracy of the characterization model. The results showed that PCA dimensionality reduction without affecting accuracy can improve running speed; using the CIE standard chromaticity observer function as the spectral weighting can improve model accuracy. However, because the display emission spectrum and its driving values ​​are used simultaneously for PCA and RBF neural network modeling, data reuse occurs. Therefore, this method lacks analysis of the accuracy of principal component reconstruction of the display emission spectrum and lacks research on the calibration of the display principal component matrix, leading to coupling between the RBF neural network model and the display's emission spectral characteristics.

[0006] In 2021, Yao Jianwei et al. proposed a spectral characterization method for liquid crystal displays based on singular value decomposition (SVD). This method obtains the k most important basis vectors for a single channel through SVD, so the spectral radiance of any single channel is a linear combination of these basis vectors. The corresponding sub-basal coordinates can also be obtained through orthogonality, thus establishing a lookup table. Therefore, for any driving value of a single channel, the corresponding spectral radiance can be linearly represented by the basis vectors, and the combination coefficients are obtained through lookup tables and interpolation techniques. Finally, for any set of driving values ​​R, G, and B, the overall spectral radiance can be predicted by summing the spectra of the single channels. However, this method essentially performs principal component analysis on a single channel. According to the channel independence mechanism, the eigenvectors of the principal components of a single channel are basically in the same direction, so this method cannot perform reverse characterization, resulting in a large prediction error when the RGB values ​​are small.

[0007] As can be seen from the above methods for display spectral characterization, display spectral characterization processes the display driving values ​​to find a suitable conversion model from device-related RGB values ​​to the spectral space. Current generation methods have the following problems:

[0008] 1. Directly modeling the spectral characteristics of RGB driving values ​​results in low accuracy due to data redundancy and severe nonlinearity;

[0009] 2. The spectral superposition method requires the establishment of multiple regression equations, resulting in high model complexity;

[0010] 3. Neural network models are prone to getting stuck in local minima and are constrained by the initial weights and thresholds, which makes model training difficult and results in low accuracy;

[0011] 4. Weighted least squares method has high requirements for data, requires processing of each data item, and has a long calculation time.

[0012] Since display spectral datasets are often represented in the form of complex nonlinear structures, how to accurately and effectively characterize the display spectrum, and provide a fast and efficient method for characterizing the display spectrum, is a problem that urgently needs to be overcome in the field of high-precision and high-fidelity color display of displays.

[0013] Therefore, proposing a display spectral characterization method based on support vector machines to address the difficulties in existing technologies is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0014] In view of this, in order to solve these difficulties, the present invention provides a display spectral characterization method based on support vector machine regression. The purpose of the present invention is to predict the high-dimensional spectral data corresponding to the three driving values ​​of the display. The method uses principal component analysis to reduce the dimensionality of the calibrated high-dimensional spatial spectral data, describes the display spectrum through the principal components of the calibrated data, and combines support vector machine regression to achieve spectral principal component prediction, thus providing an effective method for display spectral characterization.

[0015] To achieve the above objectives, the present invention adopts the following technical solution:

[0016] A display spectral characterization method based on support vector machines includes the following steps:

[0017] S1. Perform principal component analysis on the radiance of the display's emission spectrum to obtain the orthogonal composition of the display's emission spectrum;

[0018] S2. Based on the RGB three-primary-color mixing mechanism of the display color information, determine the number of principal components, conduct principal component transformation matrix calibration experiments, and obtain the decomposition matrix and reconstruction matrix.

[0019] S3. Based on the experiment, construct data pairs of display color driving values ​​and principal components of emission spectrum, and establish a theoretical model for the regression between support vector machine color driving values ​​and principal components of display emission spectrum.

[0020] S4. Design parameter optimization algorithm, determine the theoretical model parameters of S3, and obtain the prediction model;

[0021] S5. Input the color driving value into the spectral principal component prediction model to obtain the spectral emission principal component prediction value corresponding to the driving value;

[0022] S6. Further combine the principal component reconstruction matrix to perform principal component inverse transformation, and perform principal component-based display emission spectrum reconstruction.

[0023] S7. Evaluate the accuracy of the reconstructed emission spectrum.

[0024] The above method, optionally, includes S1 specifically including:

[0025] S11. Define the characteristic color patches for principal component analysis, including primary color patches and neutral color patches. The driving value for the primary color patches is... The driving value of neutral color blocks ,in They are The values ​​of each element in the sample color block are N, where N is a positive integer.

[0026] S12. The display shows a sample color patch, and a spectrophotometer is used in a dark room to measure the emission spectrum of the sample color patch on the display.

[0027] S13. Perform principal component analysis on the measured N sets of emission spectra to obtain the composition of different principal components.

[0028] The above method, optionally, includes S2 specifically:

[0029] S21. Based on the RGB three-primary-color mixing mechanism of display light emission color information, retaining three principal components;

[0030] S22. Based on the data in S13, obtain the principal component decomposition matrix and reconstruction matrix of the display emission spectrum;

[0031] S23. Based on the principal component reconstruction matrix, perform inverse principal component transformation to obtain the principal component reconstruction spectrum. Compare the average relative error of the spectral coverage area between the reconstructed spectrum and the original spectrum to evaluate the reconstruction accuracy of the three principal components of the display emission spectrum.

[0032] The above method, optionally, includes S3 specifically:

[0033] S31. The monitor generates three-factor, eight-level experimental samples, removes the S1 feature color block, sets the RGB three-driving values ​​to be located in the central color block of the screen, and generates the set RGB space color block.

[0034] S32. Select the spectrophotometer measuring instrument and measure the color patch generated on the current display. The corresponding spectral radiance value, with a spectral interval of 10nm and a spectral width of 380nm-780nm, is multiplied by the principal component transformation decomposition matrix to obtain the three principal components of the display emission spectrum. The color driving value and the spectral principal component data pairs are obtained to construct the dataset.

[0035] S33. Using the device-dependent color space color driving value RGB as the input space and the principal components of the display emission spectrum in the device-independent color space as the output space, establish a support vector machine regression model.

[0036] The above method, optionally, includes S4 specifically:

[0037] S41. Divide the dataset into a training set, a validation set, and a test set;

[0038] S42. Select the optimal model hyperparameters based on grid search. The selected hyperparameters are Gamma, penalty coefficient C, and bias. ;

[0039] S43. Based on the accuracy of the validation set, select the optimal parameter combination and construct a support vector machine model to represent color-driven values ​​to spectral principal components.

[0040] S44. Encapsulate a spectral principal component representation model based on support vector machines.

[0041] The above method, optionally, includes S5 specifically:

[0042] Extract color driving values ​​from the color image displayed on the monitor. After prediction by the spectral principal component characterization model of the support vector machine, the output is the predicted value of the three principal components of the emission spectrum of the color-driven value display.

[0043] The above method, optionally, includes S6 specifically:

[0044] By combining the predicted values ​​of the three principal components and the spectral principal component reconstruction matrix, an inverse principal component transformation is performed to obtain the emission spectrum prediction values ​​corresponding to the RGB color driving values ​​of the display, and then a spectral reconstruction based on principal components is performed.

[0045] The above method, optionally, includes S7 specifically:

[0046] S71. Calculate the average relative error between the predicted spectral values ​​and the test spectra based on the test set;

[0047] S72. After converting the reconstructed spectral values ​​and test spectral values ​​to the CIE1931XYZ color space using the color matching function, a color difference-based accuracy assessment is performed.

[0048] The above methods, optionally, include, but are not limited to, the following methods for evaluating color difference: CIELAB, CIEDE2000, and CIELUV.

[0049] As can be seen from the above technical solution, compared with the prior art, the present invention provides a display spectral characterization method based on support vector machine, which has the following beneficial effects:

[0050] By utilizing support vector machine regression and principal component dimensionality reduction methods, compared with current mainstream display spectral characterization methods, this approach demonstrates good prediction accuracy, robustness, and low redundancy, providing an effective method for display spectral characterization. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0052] Figure 1 This is a flowchart of the RBF support vector machine-based display spectral characterization process disclosed in this invention;

[0053] Figure 2 A diagram showing the experiments disclosed in this invention;

[0054] Figure 3 This is a spectral distribution diagram of the principal component characteristic color patches disclosed in this invention;

[0055] Figure 4 This is a diagram showing the variance contribution rate of each principal component after principal component transformation as disclosed in this invention.

[0056] Figure 5 This is a 3D scatter plot of the experimental RGB driving value distribution disclosed in this invention;

[0057] Figure 6 This is a diagram of the support vector machine structure disclosed in this invention;

[0058] Figure 7 This is the average spectral error diagram disclosed in this invention;

[0059] Figure 8 This is the color difference analysis diagram disclosed in this invention;

[0060] Figure 9 This invention discloses the CIE 1931 XYZ color matching function. Detailed Implementation

[0061] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0062] In this application, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0063] Reference Figure 1 As shown, this invention discloses a method for spectral characterization of a display, comprising the following steps:

[0064] S1. Perform principal component analysis on the radiance of the display's emission spectrum to obtain the orthogonal composition of the display's emission spectrum;

[0065] S2. Based on the RGB three-primary-color mixing mechanism of the display color information, determine the number of principal components, conduct principal component transformation matrix calibration experiments, and obtain the decomposition matrix and reconstruction matrix.

[0066] S3. Based on the experiment, construct data pairs of display color driving values ​​and principal components of emission spectrum, and establish a theoretical model for the regression between support vector machine color driving values ​​and principal components of display emission spectrum.

[0067] S4. Design parameter optimization algorithm, determine the theoretical model parameters of S3, and obtain the prediction model;

[0068] S5. Input the color driving value into the spectral principal component prediction model to obtain the spectral emission principal component prediction value corresponding to the driving value;

[0069] S6. Further combine the principal component transformation and reconstruction to perform principal component inverse transformation, and perform principal component-based display emission spectrum reconstruction.

[0070] S7. Evaluate the accuracy of the reconstructed emission spectrum.

[0071] Furthermore, S1 specifically includes:

[0072] S11. Define the characteristic color patches for principal component analysis, including primary color patches and neutral color patches. The driving value for the primary color patches is... The driving value of neutral color blocks ,in They are The values ​​of each element in the sample color patch are 29 in total.

[0073] S12, such as Figure 2 As shown, select the measurement environment: The measurement environment should be in a dark room, the monitor should be placed on an optical platform, and the measurement distance should be four times the width of the screen;

[0074] The monitor displays a sample color patch. In a dark room, a spectrophotometer is used to measure the emission spectrum of the sample color patch on the monitor. The spectrophotometer is positioned at the center of the displayed color patch, with an observation angle of 1°. The monitor displays the sample color patch. The spectrophotometer is used to measure the emission spectrum of the sample color patch on the monitor in a dark room. Figure 3 As shown;

[0075] S13. Perform principal component analysis on the 29 sets of emission spectra measured. Change the display color blocks in sequence to obtain 29 sets of emission spectra. Perform principal component analysis on the 29 sets of emission spectra, calculate the covariance matrix of the data matrix, and then obtain the eigenvalues ​​and eigenvectors of the covariance matrix.

[0076] Furthermore, S2 specifically includes:

[0077] S21. Based on the RGB three-primary-color mixing mechanism of the light emitted by the display, the three principal components are retained;

[0078] S22. Based on the data in S13, the principal component decomposition matrix and reconstruction matrix of the display emission spectrum are obtained. Through the three-primary-color mixing mechanism, the matrix composed of the eigenvectors corresponding to the three preferred features with the largest eigenvalues ​​(i.e., the largest variance) is selected to realize principal component analysis. The relationship is shown in formula (1):

[0079] (1)

[0080] for The matrix composed of the three eigenvectors, and for The dataset is a matrix composed of several eigenvalues, with the eigenvalues ​​located on the main diagonal and the remaining positions set to 0. To reduce the dataset from n dimensions to k dimensions, we need to find the eigenvectors corresponding to the three largest eigenvalues. The matrix formed by these three eigenvectors... This is the matrix we need, i.e., the principal component decomposition matrix of the display's emission spectrum. Its transpose is the principal component reconstruction matrix. The cumulative variance contribution rate is calculated, and the total contribution rate is verified to be greater than 99%, thus verifying the feasibility of the display's emission spectrum based on the color mixing principle with three principal components. Figure 4 As shown. And retain the decomposition matrix and transformation matrix;

[0081] S23. Based on the principal component reconstruction matrix, perform inverse principal component transformation to obtain the principal component reconstructed spectrum. Compare the average relative error of the spectral coverage area between the reconstructed spectrum and the original spectrum with the color difference between the reconstructed spectrum and the original spectrum, as shown in Table 1.

[0082]

[0083] Table 1

[0084] Furthermore, S3 specifically includes:

[0085] S31. The monitor generates a three-factor, eight-level experimental sample, removes the S1 characteristic color block, sets the RGB three-driving values ​​to be located in the central color block of the screen, and generates the set RGB space color block. The RGB driving values ​​are as follows: Figure 5 As shown;

[0086] S32. Select the spectrophotometer measuring instrument and measure the color patch generated on the current display. The corresponding spectral radiance value, with a spectral interval of 10nm and a spectral width of 380nm-780nm, is multiplied by the principal component transformation decomposition matrix to obtain the three principal components of the display emission spectrum. The color driving value and the spectral principal component data pairs are obtained to construct the dataset.

[0087] S33. Using the device-dependent color space color driving value RGB as the input space and the principal components of the display emission spectrum in the device-independent color space as the output space, modeling is performed based on the support vector machine regression model.

[0088] Furthermore, a three-factor, eight-level experimental sample was designed. The RGB color driving value was three factors, each with eight levels. The length and width of the color block were 1 / 5 of the height of the monitor, but not less than 5cm. The color block was placed in the center of the monitor and the driving value data was recorded.

[0089] Select a Konica Minolta CS2000 spectrophotometer to measure the color patches generated by the current display. The corresponding spectral radiance is in the range of 380nm-780nm, with an interval of 10nm.

[0090] Multiply the measured spectrum by the principal component decomposition matrix to obtain the three principal components of the display emission spectrum, and obtain the principal component data corresponding to all driving values, thereby obtaining the color driving value and spectral principal component data pair;

[0091] Using the device-dependent color space (RGB) color driving values ​​as the input space and the device-independent color space (PCI) display emission spectrum principal components as the output space, modeling is performed based on a support vector machine regression model, as follows: Figure 6 As shown;

[0092] The model between the device-dependent color space color driving value RGB and the principal components of the display emission spectrum in the device-independent color space can be described by formula (2):

[0093] (2)

[0094] In the formula, This represents the device-dependent color space color driver value (RGB). Indicates the principal components of the emission spectrum of a device-independent color space display; Indicates from RGB to Color space conversion model.

[0095] The model uses radial basis function (RBF) kernel function. By modeling nonlinear regression and classification as convex optimization problems, and finding the optimal hyperplane through supervised learning, the nonlinear problem is transformed into a support vector machine model for linear problems.

[0096] The objective function for support vector machine regression is shown in equation (3):

[0097] (3)

[0098] In the formula, the first term is Insensitive loss function The first term is the regression hyperplane coefficient; the second term is the penalty term, where C is the penalty coefficient. , Defined as slack variables, representing the upward slack and downward slack.

[0099] set up For support vector regression in the transformation from device-dependent color space to device-independent color space -Bandwidth, taking into account - To improve bandwidth robustness, add a penalty term for loss contribution. Establish... The support vector machine regression model has an allowable error band of [-]. ].

[0100] The constraints for each sample observation are shown in Equation (4):

[0101] (4)

[0102] In the formula, This represents the maximum deviation between the function and the predicted value. The constrained optimization problem is reformulated into a dual problem using Lagrange multipliers, and quadratic programming is used to determine each constraint. The above problem, when transformed into a dual problem, also needs to satisfy the KKT conditions as shown in formula (5):

[0103] (5)

[0104] In the formula, , , , (i=1, 2, ..., N) are all Lagrange multipliers. Centered on the boundary, upper and lower boundaries were constructed as follows: and The "pipeline".

[0105] Mapping the device-specific color space values ​​(RGB) to a high-dimensional space, the resulting feature vector is... Then the regression prediction value of the spectral characteristic tristimulus is given by formula (6):

[0106] (6)

[0107] In the formula, n This indicates the number of support vectors.

[0108] Therefore, the main parameters that the support vector machine model adjusts are kernel, Gamma, penalty coefficient C, and bias. (epsilon). Kernel represents the kernel function;

[0109] The RBF kernel function was chosen, and Gamma represents the influence of a single training sample. A smaller penalty coefficient C results in a smoother decision surface, allowing for greater tolerance and stronger generalization ability; a larger C penalizes slack variables, aiming for them to be close to zero, leading to high accuracy on the training set but weaker generalization ability. Bias This indicates the tolerance of SVR for the deviation between the predicted value and the label of the sample. The larger the value, the fewer the number of support vectors; The smaller the value, the better. By dividing the dataset into training and validation sets, the optimal model hyperparameters are selected based on grid search, specifically for the three hyperparameters mentioned above: Gamma, penalty coefficient C, and bias. Based on the accuracy of the validation set, the optimal parameter combination is selected to realize the construction of a support vector machine model from color-driven values ​​to spectral principal component representation.

[0110] Furthermore, S4 specifically includes:

[0111] S41. Divide the dataset into a training set, a validation set, and a test set;

[0112] S42. Select the optimal model hyperparameters based on grid search. The selected hyperparameters are Gamma, penalty coefficient C, and bias. ;

[0113] S43. Based on the accuracy of the validation set, select the optimal parameter combination and construct a support vector machine model to represent color-driven values ​​to spectral principal components.

[0114] S44. Encapsulate a spectral principal component representation model based on support vector machines.

[0115] Furthermore, S5 specifically includes:

[0116] The color driving values ​​(RGB) of the color image displayed on the monitor are extracted, and then predicted by the spectral principal component prediction model of the support vector machine. The output is the predicted value of the three principal components of the emission spectrum of the monitor for the color driving value.

[0117] Furthermore, S6 specifically includes:

[0118] By combining the predicted values ​​of the three principal components and the spectral principal component reconstruction matrix, an inverse principal component transformation is performed to obtain the emission spectrum prediction values ​​corresponding to the RGB color driving values ​​of the display, and then a spectral reconstruction based on principal components is performed.

[0119] Furthermore, the S7 specifically includes:

[0120] S71. Calculate the average relative error between the predicted spectral values ​​and the test spectra based on the test set, such as... Figure 7 As shown;

[0121] S72. After converting the reconstructed spectral values ​​and test spectral values ​​to the CIE1931XYZ color space using the color matching function, an evaluation based on color difference is performed.

[0122] Furthermore, color difference assessments include: CIELAB, CIEDE2000, and CIELUV.

[0123] After converting the reconstructed and test spectral values ​​to the CIE 1931 XYZ color space using a color matching function, an evaluation based on color difference is performed. For example... Figure 8 The results include color difference evaluation using CIELAB, CIEDE2000, and CIELUV. The predicted and tested spectral values ​​are converted to the XYZ color space using the Color Matching Function (CMF) before evaluation. The conversion formula for the tristimulus values ​​of the light source color spectrum is shown in formula (7):

[0124] (7)

[0125] In the formula, k is the normalization coefficient. , , For color matching functions, It refers to the spectral power distribution of the light source. The CIE specifies that the Y stimulus value of the light source is 100. , , These are the spectral tristimulus values ​​in the XYZ color space; where CIE1931 , , 10nm interval color matching function as follows Figure 9 The specific matrix data is shown in Table 2.

[0126] Table 2

[0127]

[0128] CIELUV color difference calculation requires the use of CIE. Color space is used as the standard. This is called metric brightness. , This is called the metric system of color. The conversion is shown in formula (8):

[0129] (8)

[0130] In the formula, , , These are the tristimulus values ​​of the object. , , The measured tristimulus values ​​for the white peak color patch. and Here are the chromaticity coordinates of the color sample. and The chromaticity coordinates of the white peak color block. It's about brightness. and These are the chromaticity coordinates of the color block to be converted.

[0131] CIE Color difference between two chromaticity values ​​in a color space The color difference is as shown in formula (9):

[0132] =[( ) 2+( ) 2+( ) 2] 1 / 2 (9)

[0133] CIELAB color difference and CIEDE2000 color difference calculations require the use of CIE Labs. Color space is used as the standard. (CIELab) The color space is shown in formula (10):

[0134] (10)

[0135] in,

[0136] =( )1 / 2

[0137] =

[0138] In the formula, Represents brightness. Representing the weight from green to red, It represents the range from blue to yellow. and These are psychological chroma and psychological hue angle.

[0139] Color difference between two chromaticity values ​​in the CIELAB color space The color difference is as shown in formula (11):

[0140] =[( ) 2+( ) 2+( ) 2] 1 / 2 (11)

[0141] The CIEDE2000 color difference formula is shown in equation (12):

[0142] = (12)

[0143] The meaning of each parameter and the specific calculation process are as follows:

[0144] Furthermore, calculations are performed in the CIELAB color space. , , , , Equivalent.

[0145] Further calculations , , , h is calculated using the formula shown in (13):

[0146] =

[0147] a =

[0148] =

[0149] =( ) 1 / 2

[0150] = (13)

[0151] Where, G = 0.5 × This refers to the 'a' in the CIE 1976 LAB color space. The axis adjustment factor is a function of chroma. Here... It involves calculating the two colors. The arithmetic mean.

[0152] Further calculations , , These represent the difference in lightness, chroma, and hue, respectively, and are calculated using formula (14):

[0153]

[0154]

[0155] =2× ×sin( (14)

[0156] In the formula, the subscripts s and b represent the standard color and sample color in a pair of colors for which the color difference is to be calculated, respectively.

[0157] Further calculations and And RT and RC. and This is called the weighting function, which defines the length of the semi-axis of the ellipse and allows for individual adjustments based on different regions within the CIELAB color space to correct for the uniformity of the space. Its definition is shown in formula (15):

[0158]

[0159] ×

[0160] × ×T (15)

[0161] Where T = 1 - 0.17 × cos( )+0.24×cos( )+0.32×cos( -0.20×cos(4 According to CIELAB's wide-capacity diagram, the wide-capacity ellipse in the blue area does not point to the origin. To rotate the ellipse, we introduce... function:

[0162]

[0163] in The rotation angle is determined by the hue. Based on the rotation angle of the chroma transformation.

[0164] Going further, the choice Parameter factor These are correction factors related to the conditions of use; they are factors that affect the perception of chromatic aberration. Under the standard observation conditions given by the CIE... If the conditions are not met, these values ​​must be determined according to the industrial color difference evaluation conditions. The standard conditions specified by CIE are: Light source, illuminance 1000 lx, greater than 4 The field of view and color uniformity are achieved with CIELAB color difference between 0 and 5. The background is a medium-brightness gray (L=50).

[0165] Furthermore, by substituting all the parameters into formula (11), the color difference can be calculated.

[0166] This invention, based on the RGB three-primary-color mixing mechanism of display emitted color information, first analyzes that the display's emission spectrum consists of three principal components; secondly, it proposes a display spectral characterization model based on radial basis function vector machines. The model takes the display's color driving values ​​as input and increases its dimensionality through a Gaussian kernel function, aiming at spectral principal component regression. Bandwidth was used to construct the objective function for support vector regression; principal component analysis was then applied to reconstruct the display's emission spectrum, achieving spectral characterization of the display. Experimental samples were generated using selected display devices and spectral radiance measurement instruments to construct a dataset; these samples were divided into training and test sets, and the model was trained using the training set. The emission spectrum was predicted using the test set, and the algorithm's accuracy was evaluated based on both spectral and chromatic differences. Experiments verified the feasibility and accuracy of the method.

[0167] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for system or system embodiments, since they are basically similar to method embodiments, the description is relatively simple, and relevant parts can be referred to the descriptions in the method embodiments. The systems and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0168] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A display spectral featureization method based on support vector machines, characterized in that, Includes the following steps: S1. Perform principal component analysis on the radiance of the display's emission spectrum to obtain the orthogonal composition of the display's emission spectrum; S2. Based on the RGB three-primary-color mixing mechanism of the display color information, determine the number of principal components, conduct principal component transformation matrix calibration experiments, and obtain the decomposition matrix and reconstruction matrix. S3. Based on the experiment, construct data pairs of display color driving values ​​and principal components of emission spectrum, and establish a theoretical model for the regression between support vector machine color driving values ​​and principal components of display emission spectrum. S4. Design parameter optimization algorithm, determine the theoretical model parameters of S3, and obtain the prediction model; S5. Input the color driving value into the spectral principal component prediction model to obtain the spectral emission principal component prediction value corresponding to the driving value; S6. Further combine the principal component reconstruction matrix to perform principal component inverse transformation, and perform principal component-based display emission spectrum reconstruction. S7. Evaluate the accuracy of the reconstructed emission spectrum; Step S1 specifically includes: S11. Define the characteristic color patches for principal component analysis, including primary color patches and neutral color patches. The driving value for the primary color patches is... The driving value of neutral color blocks ,in They are The values ​​of each element in the sample color block are N, where N is a positive integer. S12. The display shows a sample color patch, and a spectrophotometer is used in a dark room to measure the emission spectrum of the sample color patch on the display. S13. Perform principal component analysis on the measured N sets of emission spectra to obtain the composition of different principal components; Step S4 specifically includes: S41. Divide the dataset into a training set, a validation set, and a test set; S42. Select the optimal model hyperparameters based on grid search. The selected hyperparameters are Gamma, penalty coefficient C, and bias. ; S43. Based on the accuracy of the validation set, select the optimal parameter combination and construct a support vector machine model to represent color-driven values ​​to spectral principal components. S44. Encapsulate a spectral principal component representation model based on support vector machines.

2. The display spectral featureization method based on support vector machine according to claim 1, characterized in that, Step S2 specifically includes: S21. Based on the RGB three-primary-color mixing mechanism of display light emission color information, retaining three principal components; S22. Based on the data in S13, obtain the principal component decomposition matrix and reconstruction matrix of the display emission spectrum; S23. Based on the principal component reconstruction matrix, perform inverse principal component transformation to obtain the principal component reconstruction spectrum. Compare the average relative error of the spectral coverage area between the reconstructed spectrum and the original spectrum to evaluate the reconstruction accuracy of the three principal components of the display emission spectrum.

3. The display spectral feature generation method based on support vector machine according to claim 1, characterized in that, Step S3 specifically includes: S31. The monitor generates three-factor, eight-level experimental samples, removes the S1 feature color block, sets the RGB three-driving values ​​to be located in the central color block of the screen, and generates the set RGB space color block. S32. Select the spectrophotometer measuring instrument and measure the color patch generated on the current display. The corresponding spectral radiance value, with a spectral interval of 10nm and a spectral width of 380nm-780nm, is multiplied by the principal component transformation decomposition matrix to obtain the three principal components of the display emission spectrum. The color driving value and the spectral principal component data pairs are obtained to construct the dataset. S33. Using the device-dependent color space color driving value RGB as the input space and the principal components of the display emission spectrum in the device-independent color space as the output space, establish a support vector machine regression model.

4. The display spectral feature generation method based on support vector machine according to claim 1, characterized in that, Step S5 specifically includes: Extract color driving values ​​from the color image displayed on the monitor. After prediction by the spectral principal component prediction model of the support vector machine, the output is the predicted value of the three principal components of the emission spectrum of the color driving value display.

5. The display spectral characterization method based on support vector machine according to claim 1, characterized in that, Step S6 specifically includes: By combining the predicted values ​​of the three principal components and the spectral principal component decomposition matrix, an inverse principal component transformation is performed to obtain the predicted emission spectrum values ​​corresponding to the RGB color driving values ​​of the display, and then a spectral reconstruction based on principal components is performed.

6. The display spectral feature generation method based on support vector machine according to claim 1, characterized in that, Step S7 specifically includes: S71. Calculate the average relative error between the predicted spectral values ​​and the test spectra based on the test set; S72. After converting the reconstructed spectral values ​​and test spectral values ​​to the CIE1931XYZ color space using the color matching function, a color difference-based accuracy assessment is performed.

7. The display spectral characterization method based on support vector machine according to claim 6, characterized in that, Color difference assessment includes, but is not limited to: CIELAB, CIEDE2000, and CIELUV.

Images