Color cast correction method of panel
By using neural network models and transfer learning techniques, a color correction model for unknown panels can be quickly established, solving the problems of time-consuming and error-prone methods in existing methods, and achieving efficient and accurate panel color correction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-27
- Publication Date
- 2026-03-10
AI Technical Summary
Existing color calibration methods for display panels are time-consuming and have large errors, making it difficult to achieve high-precision color calibration in a short time.
A neural network model is used for color correction. Combined with transfer learning technology, a predictive model and an inverse model are built using a small number of RGB values. A correction model for unknown panels is quickly built, and brightness and color accuracy are ensured through white point correction.
Achieve high-precision panel color calibration in a short time, reduce the time for measuring panel data on the production line, and maintain high-precision color calibration results.
Smart Images
Figure CN121640900A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a color deviation correction method for a panel, and more particularly to a color deviation correction method for an organic light-emitting diode (OLED) panel. Background Technology
[0002] Display panels are an indispensable component of modern electronic products. Display panels include liquid crystal display (LCD) panels, organic light-emitting diode (OLED) panels, and quantum dot panels. LCD panels offer advantages such as thinness and low power consumption. OLED panels offer advantages such as high contrast and wide color gamut. Quantum dot panels offer advantages such as high saturation and wide color gamut. Display panels are widely used in televisions, computer monitors, mobile phones, tablets, automotive display systems, various digital signage, industrial control displays, and other fields, bringing a high-quality visual experience to modern life.
[0003] Because display panels have different display characteristics and expectations, the colors displayed on them often deviate from the intended colors. To achieve the desired visual effect, color calibration of the display panel is necessary to eliminate color shifts caused by its display characteristics and ensure that the displayed colors match the true colors. In addition to the initial calibration at the factory, the colors of the display panel will gradually shift over time, thus requiring periodic color calibration.
[0004] During color cast correction, the RGB values of the input panel are measured and calibrated to correspond to the XYZ and Lab values of the panel output. In previous related technologies, a common method was to use polynomial functions to calculate the transformation matrix and perform color calibration through iterative calculations. Another method is to create a look-up table (LUT) and use linear interpolation to achieve color calibration. Existing color cast correction methods are widely used in the field of color calibration; however, they also have some limitations. Existing methods require a significant amount of time and computation to obtain correction results with small errors. Summary of the Invention
[0005] This invention provides a method for correcting panel color shift, comprising converting target RGB values into XYZ values, converting XYZ values into RGB values using an inverse model, wherein the inverse model is generated based on neural network model training, and the panel under test displays pixels according to the RGB values.
[0006] This invention also provides a method for correcting panel color cast, comprising converting XYZ values into target Lab values based on the true white point of the panel under test, and converting the XYZ values into RGB values using an inverse model, wherein the inverse model is generated based on a neural network model trained on it. The method further comprises converting RGB values into predicted Lab values based on a prediction model and the true white point, wherein the prediction model is generated based on another neural network model trained on it, adjusting the RGB values based on the predicted Lab values and the target Lab values, and displaying pixels on the panel under test according to the RGB values. Attached Figure Description
[0007] Figure 1 is a neural network architecture diagram of a prediction model according to an embodiment of the present invention.
[0008] Figure 2 is a flowchart of the method for selecting 64 selected RGB data in an embodiment of the present invention.
[0009] Figure 3 is a flowchart of the prediction model transfer learning method in an embodiment of the present invention.
[0010] Figure 4 is a flowchart of the reverse model transfer learning method in an embodiment of the present invention.
[0011] Figure 5 is a flowchart of the white spot correction method in an embodiment of the present invention.
[0012] Figure 6 is a flowchart of the panel color correction method in an embodiment of the present invention.
[0013] The reference numerals in the attached figures are explained as follows:
[0014] 100: Predictive Model
[0015] 102: Normalized RGB values
[0016] 104, 106, 108: Hidden layers
[0017] 110: Predicted Lab value
[0018] 200: Method
[0019] RGB value of 202:273
[0020] 204:125 RGB values
[0021] 206:124 RGB values
[0022] 208:123 RGB values
[0023] 210:64 RGB values
[0024] 212: 64 RGB values used for transfer learning
[0025] 300: Method
[0026] S302, S304, S306, S308, S310, S312: Steps
[0027] 400: Method
[0028] S402, S404, S406, S408, S410, S412: Steps
[0029] 500: Methods
[0030] 502: Adjust RGB grayscale
[0031] 504: Predictive Model
[0032] 506: Lab value
[0033] 508: XYZ value
[0034] 510: x, y, Y values
[0035] 511: Steps
[0036] 512: True White Point (Xn, Yn, Zn)
[0037] 600: Method
[0038] 604, 606, 617, 620: Step 602: Target RGB value
[0039] 608: Target Lab value
[0040] 610: Step 612: Prediction Model
[0041] 614: Corrected Lab
[0042] 616: Error
[0043] 618: Adjusted RGB values Detailed Implementation
[0044] The RGB, XYZ, and Lab color spaces used in this embodiment of the invention are defined as follows. The RGB color space uses a combination of three basic colors—red (R), green (G), and blue (B)—to produce various colors. The R, G, and B values represent the grayscale values of red, green, and blue light, respectively, and these grayscale values can range from 0 to 255. When the R, G, and B values are all 0, the produced color is close to black; when the R, G, and B values are all 255, the produced color is close to white. The R, G, and B values are collectively referred to as RGB values, which can be provided to televisions, computer monitors, and electronic display devices to produce images. For example, the color of each pixel in a color image on a computer is generated by specific RGB values.
[0045] The XYZ color space was developed by the International Commission on Illumination (CIE) to describe how the human eye perceives light. The Y value represents luminance, the Z value is approximately equal to blue in the RGB model, and the X value is a mixture of the red, green, and blue curves. The X, Y, and Z values are collectively referred to as XYZ values, representing virtual reference stimulus values and are a mathematical description of color, unrelated to subjective human visual perception. XYZ values can be measured using a colorimeter or spectrophotometer and are used for color conversion and mapping between devices.
[0046] The Lab color space, also defined by the CIE, is used to mathematically mimic human visual perception. The L value represents the perceived lightness component, the a value represents the chromaticity components from green to red, and the b value represents the chromaticity components from blue to yellow. The L, a, and b values are collectively called Lab values, used to describe the color performance of a specific display panel. Broadly speaking, RGB is used for electronic displays, XYZ for color calculations and device mapping, while Lab is used to simulate the human eye's perception of a specific display panel.
[0047] The unique and innovative features of this invention include transfer learning, white point correction, a predictive model, and a reverse model, which together construct a neural network model color correction algorithm. For an unknown panel (the panel to be tested), a small number of color combinations can be selected to extract experimental color data. Based on a baseline model established with a large amount of prior data, transfer learning is implemented to quickly build a predictive model and a reverse model for the unknown panel. Using the predictive model to perform color compensation beforehand ensures brightness accuracy. White point correction, through the predictive model, ensures brightness accuracy and also verifies and ensures the accuracy of color compensation. The reverse model provides effective initial color compensation guesses, enabling a fast and accurate correction process.
[0048] This invention primarily utilizes artificial intelligence (AI) technology to establish a neural network (NN) model, replacing the previous method of using 3D lookup tables and 3D linear interpolation to find the corrected color, thus achieving more complex and high-dimensional nonlinear equations. This allows the model to learn more diverse and complex panel information. When the established model closely matches the panel's grayscale value information, the accuracy of panel correction can be significantly improved and the compensation time shortened. To avoid impacting the panel production line, this invention uses known panels to establish a pre-trained prediction model. When facing an unknown panel, N sets (N is a positive integer) of the most representative RGB values are selected from the panel's 19683 sets of RGB values, and the corresponding N sets of RGB values for the unknown panel are measured. Then, using a baseline model combined with transfer learning technology, a neural network model that closely approximates the unknown panel data is established in a short time. Since only N sets of panel RGB values are needed, the time for measuring panel data on the production line is significantly reduced. The 19683 sets of RGB values in this invention are one embodiment, and the invention is not limited thereto.
[0049] Figure 1 is a neural network architecture diagram of a prediction model 100 according to an embodiment of the present invention. The input value of the prediction model 100 is a normalized RGB value 102, and the output value is a predicted Lab value 110. The model passes through a hidden layer 104 containing 355 neurons, a hidden layer 106 containing 355 neurons, and a hidden layer 108 containing 155 neurons. The number of hidden layers and the number of neurons in each hidden layer of the prediction model 100 are merely examples, and the invention is not limited thereto. During training, the mean square error (MSE) is used as the loss function, and Adam (Adaptive Momentum) is used as the optimizer. The initial learning rate (LR) is set to 0.01, the learning rate decay rate is set to 0.7, cross-validation is used, the training lasts for 150 epochs, and the batch size is set to 256. Using the Lab value as the output value of the prediction model 100 provides better color accuracy. In some embodiments, the output value of the prediction model 100 can also be the predicted XYZ value. The input to the inverse model is normalized XYZ values, and the output is predicted RGB values. The model passes through three hidden layers (155 neurons each) in the first, second, and third layers. The number of hidden layers and neurons in each hidden layer are merely examples, and the invention is not limited to these. During training, MSE is used as the loss function and Adam is used as the optimizer. The initial learning rate is set to 0.001, the learning rate decay rate is set to 0.9, cross-validation is used, training is performed for 150 epochs, and the batch size is set to 128.
[0050] When dealing with unknown panels on a production line, transfer learning can be used to further fine-tune the neural network model. First, a baseline model is established using 273 RGB values and corresponding XYZ values from each of 15 known panels, thus creating 15 baseline models for the 15 known panels. Next, in one embodiment, the R, G, and B values are selected as [10, 70, 130, 180, 250], and different permutations and combinations are performed to obtain 125 different RGB combinations, which are the most representative grayscale RGB combinations. This invention can choose other R, G, and B values to achieve the same effect. On the production line, the XYZ values corresponding to each RGB value combination in the unknown panel are measured and compared. Next, the normalized XYZ values corresponding to each RGB value combination in the unknown panel are calculated. exact,i The corresponding normalized XYZ value in the known panel. approx,iA comparison is then made. In one embodiment, the Mean Absolute Error (MAE) is used as an error metric to measure the difference between two sets of normalized XYZ values, as follows:
[0051]
[0052] Where X approx,i Y approx,i Z approx,i These are the normalized XYZ values from the known panel, X... exact,i Y exact,i Z exact,i These are the normalized XYZ values in the unknown panel.
[0053] In another embodiment, the Mean Square Error (MSE) can also be used as an error metric to measure the difference between two sets of normalized XYZ values, as follows:
[0054]
[0055] In another embodiment, the root mean square error (RMSE) can also be used as an error metric to measure the difference between two sets of normalized XYZ values, as follows:
[0056]
[0057] Based on the comparison using MAE (Magnitude of Effect), the baseline model with the smallest difference from the normalized XYZ values of the known panel is selected as the chosen baseline model. Once this chosen baseline model is established, transfer learning can be performed on it. When dealing with an unknown panel, 125 of the most representative grayscale values (i.e., R, G, B values from [10, 70, 130, 180, 250]) can be selected from the 273 grayscale values of the known panel. Then, using the previously established chosen baseline model combined with transfer learning techniques, a predictive model that closely approximates the grayscale value information of the unknown panel and an inverse model that can guess the RGB values are built in a short time. The predictive and inverse models will retain the learning results of the neural network model on the known panel and apply them to the data of the unknown panel, saving time and cost while maintaining high accuracy in color correction.
[0058] In another embodiment, selected RGB values can also be chosen from the 125 representative grayscale RGB values for transfer learning of the selected benchmark model. Figure 2 is a flowchart of the method 200 for selecting 64 selected RGB values in an embodiment of the present invention. Method 200 first selects 125 representative RGB values 204 from the 273 training data 202, which are permutations of RGB values [10, 70, 130, 180, 250]. Then, the mean square error (MSE) of the 125 representative RGB values is calculated using the M benchmark models of the M known panels trained on the 273 data 202. MSE can be replaced by root mean square error (RMSE) or mean absolute error (MAE). The RGB value with the largest MSE is deleted to obtain 124 RGB values 206 with the smallest MSE among the M known panels. Then, the RGB value with the largest MSE is deleted again to obtain 123 RGB values 208 with the smallest MSE among the M known panels. This process continues until the 64 RGB values 210 with the smallest MSE among the remaining M known panels are obtained. These selected 64 RGB values 212 will be used for transfer learning on a selected baseline model of the unknown panels. The 64 RGB values are just one example; the invention is not limited to this, and in practice, there could be N RGB values, where N is a positive integer.
[0059] On the production line, the XYZ values corresponding to 64 RGB value combinations in an unknown panel are measured and compared. Next, the normalized XYZ value corresponding to each RGB combination in the unknown panel is calculated. exact,i The corresponding normalized XYZ value in the known panel. approx,i A comparison is then made. In one embodiment, the Mean Absolute Error (MAE) is used as an error metric to measure the difference between two sets of normalized XYZ values, as follows:
[0060]
[0061] In another embodiment, the Mean Square Error (MSE) can also be used as an error metric to measure the difference between two sets of normalized XYZ values, as follows:
[0062]
[0063] In another embodiment, the root mean square error (RMSE) can also be used as an error metric to measure the difference between two sets of normalized XYZ values, as follows:
[0064]
[0065] For detailed calculation methods of MAE, MSE, and RMSE, please refer to the previous text, which will not be repeated here.
[0066] Figure 3 is a flowchart of the prediction model transfer learning method 300 in an embodiment of the present invention. Method 300 includes steps S302 to S312. Any reasonable technical changes or adjustments to the steps fall within the scope of the present invention. Steps S302 to S312 are explained as follows:
[0067] Step S302: Measure the XYZ values corresponding to the N sets of RGB values of the unknown panel;
[0068] Step S304: Input N representative RGB values into M benchmark models to generate M×N predicted XYZ values;
[0069] Step S306: Calculate the M MSEs of the M baseline models based on the corresponding N XYZ values of the unknown panel and the M×N predicted XYZ values;
[0070] Step S308: Select the baseline model with the smallest MSE;
[0071] Step S310: Perform transfer learning based on the selected benchmark model; and
[0072] Step S312: Obtain the prediction model for the unknown panel.
[0073] In step S302, the XYZ values corresponding to the N sets of RGB values of the unknown panel are measured, where the N sets of RGB values are taken from the method 200 for selecting N selected RGB values in Figure 2. In one embodiment, N can be 64. In step S304, N representative RGB values are input into M benchmark models to generate M×N predicted XYZ values, with each benchmark model generating N predicted XYZ values. In one embodiment, M can be 15. In step S306, the M MSEs of the M benchmark models are calculated based on the corresponding N XYZ values of the unknown panel and the M×N predicted XYZ values. In one embodiment, MSE can be replaced by root mean square error (RMSE) or mean absolute error (MAE). The detailed calculation methods of MAE, MSE, and RMSE can be referred to above and will not be repeated here. In step S308, the benchmark model with the smallest MSE, MAE, or RMSE is selected. In step S310, a baseline model is selected and trained for transfer learning. The transfer learning data includes N representative RGB values and their corresponding XYZ values for the unknown panel. In one embodiment, before transfer learning, the Lab values predicted by the prediction model for the known panel are converted into corresponding normalized XYZ values (XYZ values corresponding to the RGB values (255, 255, 255) based on the reference white point) and compared with the corresponding normalized XYZ values of the unknown panel. In another embodiment, before transfer learning, the measured XYZ values of the unknown panel are converted into Lab values (XYZ values corresponding to the RGB values (255, 255, 255) based on the reference white point) and compared with the Lab values corresponding to the known panel. In step S312, the prediction model for the unknown panel is obtained after training. The input of the prediction model is RGB values, and the output is the Lab values corresponding to the unknown panel. In one embodiment, the input of the prediction model can be normalized RGB values.
[0074] Figure 4 is a flowchart of the reverse model transfer learning method 400 in an embodiment of the present invention. Method 400 includes steps S402 to S412. Any reasonable technical changes or adjustments to the steps fall within the scope of the present invention. Steps S402 to S412 are explained as follows:
[0075] Step S402: Measure the XYZ values corresponding to the N sets of RGB values of the unknown panel;
[0076] Step S404: Input the N XYZ values corresponding to the unknown panel into M benchmark inverse models to generate M×N predicted RGB values;
[0077] Step S406: Calculate the M MSEs of the M baseline inverse models based on the N representative RGB values and the M×N predicted RGB values;
[0078] Step S408: Select the selected baseline inverse model with the smallest MSE;
[0079] Step S410: Perform transfer learning based on the selected benchmark inverse model; and
[0080] Step S412: Obtain the inverse model of the unknown panel.
[0081] In step S402, the XYZ values corresponding to the N sets of RGB values of the unknown panel are measured, where the N sets of RGB values are taken from the method 200 for selecting N selected RGB values in Figure 2. In one embodiment, N can be 64. In step S404, the N XYZ values corresponding to the unknown panel are input into M benchmark inverse models to generate M×N predicted RGB values, with each benchmark model generating N predicted RGB values. In one embodiment, M can be 15. In step S406, the M MSEs of the M benchmark inverse models are calculated based on the N representative RGB values and the M×N predicted RGB values. In one embodiment, MSE can be replaced by root mean square error (RMSE) or mean absolute error (MAE). The detailed calculation methods of MAE, MSE, and RMSE can be referred to above and will not be repeated here. In step S408, the benchmark inverse model with the smallest MSE, MAE, or RMSE is selected. In step S410, a baseline inverse model is selected and trained for transfer learning. The transfer learning data includes N representative RGB values and their corresponding XYZ values for the unknown panel. In step S412, the training is completed to obtain the inverse model for the unknown panel. The input of the inverse model is the XYZ values corresponding to the unknown panel, and the output is the RGB values. In one embodiment, the input of the inverse model can be normalized XYZ values, and the output can also be normalized RGB values.
[0082] Figure 5 is a flowchart of the white point correction method 500 in this embodiment of the invention. Initially, the G value in the RGB values is set to the maximum G value (e.g., 255). The R and B values are adjusted to make the x and y values of the prediction model close to the x and y values of D65. Since the G value has the greatest impact on brightness, it is initially set to the maximum G value to provide the initial value for the true white point. The normalized XY values are the x and y values. D65 is the color of a light source with a color temperature of 6500K, and its (x, y) = (0.311, 0.329), which can be used as a reference white point. In Figure 5, adjusting the RGB grayscale 502 is the first step (i.e., fixing G to the maximum G value (e.g., 255) and adjusting the R and B values). The resulting RGB values are processed by the prediction model 504 to obtain the Lab value 506. The Lab value 506 is converted from the white point with RGB values of (255, 255, 255) to obtain the XYZ value 508. The XYZ value 508 can be used to calculate the predicted x, y, and Y values 510. The predicted x-value is compared with the x-value of D65 to generate a difference dx, and the predicted y-value is compared with the y-value of D65 to generate a difference dy. If both the difference dx and dy are less than the threshold th65 (e.g., 0.0015) (step 511), the true white point (Xn, Yn, Zn) 512 can be generated. If either the difference dx or the difference dy exceeds the threshold th65 (e.g., 0.0015) (step 511), the process returns to step 502 of adjusting the RGB grayscale. First, the G value is lowered, and then the R and B values are adjusted. This process is repeated until both the difference dx and dy are less than the threshold th65 (e.g., 0.0015) (step 511). The true white point (Xn, Yn, Zn) 512 of the unknown panel is obtained through the predicted Y value. In another embodiment, the predicted XYZ values can be directly compared with the XYZ values of D65 to find the true white points (Xn,Yn,Zn)512 of the unknown panel whose XYZ values all differ from another threshold. The true white points (Xn,Yn,Zn)512 can be used for conversion between the XYZ values and Lab values of the unknown panel.
[0083] Figure 6 is a flowchart of the panel color correction method 600 in an embodiment of the present invention. Any reasonable technical changes or adjustments to the procedures fall within the scope disclosed in this invention.
[0084] In step 604, the target RGB value 602 is first converted into the normalized XYZ value of the reference white point D65. The following formula is used:
[0085]
[0086] Where R srgb G srgb B srgb It is a normalized RGB value, and
[0087]
[0088] Next, in step 606, the normalized XYZ values are input into the pre-trained inverse model to obtain the guessed RGB values. The reason for using the inverse model to guess the RGB values is that the inverse model includes the display characteristics of the unknown panel. In one embodiment, if the accuracy of the inverse model 606 is high enough, the guessed RGB values can be directly used as the final result. In another embodiment, the guessed RGB values are all integers, so the guessed RGB values still need to be adjusted, and the inverse model 606 only provides an initial guessed RGB value. The normalized XYZ value 604 of the reference white point D65 can be converted into the target Lab value 608 through the reference white point D65 for comparison. In one embodiment, the normalized XYZ value 604 of the reference white point D65 is multiplied by the Y value of the real white point to generate the adjusted XYZ value. Then, the adjusted XYZ value is converted into the target Lab value 608 according to the XYZ value of the real white point. The conversion formula (1) is as follows:
[0089]
[0090] in,
[0091] The RGB values guessed by the inverse model 606 provide initial values. Then, the RGB values 610 are adjusted to be input into the pre-trained prediction model 612. The corrected Lab value 614 is obtained using formula (1) based on the XYZ values of the real white point. In some embodiments, after obtaining the predicted Lab value output by the prediction model 612, the predicted Lab value is first converted to an XYZ value using the original white point (R,G,B)=(255,255,255), and then the XYZ value is converted to the corrected Lab value 614 based on the XYZ values of the real white point using formula (1). This corrected Lab value 614 is used to compare with the target Lab value 608. After calculating the error (MSE, MAE, RMSE, or other ΔE00 error index) 616, if the error 616 is less than the preset error (e.g., 2) (step 617), the adjusted RGB value 618 is output. If the error 616 is greater than the preset error (e.g., 2) (step 617), return to the step of adjusting the RGB value 610, re-enter the prediction model 612 and obtain a new correction Lab value. Repeat this process until ΔE00 < the preset error (e.g., 2) or the number of attempts t reaches the threshold (step 617). Then output the adjusted RGB value 618 and display the pixels according to the adjusted RGB value (step 620).
[0092] The prediction model establishment method, reverse model establishment method, white point correction method, and color deviation correction method used in the embodiments of the present invention can all be implemented by any combination of software, firmware, or hardware.
[0093] In summary, this invention employs transfer learning to quickly build predictive and inverse models for unknown panels, which is more efficient and time-saving than existing methods. Before the color shift correction method, white point correction ensures the accuracy of brightness after color correction. An accurate predictive model ensures the accuracy of color compensation selection, and during the color correction process, a pre-trained inverse model provides effective initial color compensation guesses, enabling rapid correction.
[0094] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method of panel color wash correction, the method comprising: The method comprises: converting a target RGB value into an XYZ value; converting the XYZ value into an RGB value by using an inverse model, the inverse model being generated based on a neural network model training; and a display panel displays a pixel according to the RGB value.
2. The method of claim 1, wherein, The target RGB value and the XYZ value are both normalized.
3. The method of claim 1, wherein, The method further comprises: converting the XYZ value into a target Lab value according to a real white point of the display panel; converting the RGB value into a predicted Lab value according to a prediction model and the real white point, the prediction model being generated based on another neural network model training; and adjusting the RGB value according to the predicted Lab value and the target Lab value.
4. The method of claim 3, wherein, The method further comprises: correcting the real white point according to the prediction model to generate an XYZ value of the real white point; wherein converting the XYZ value into the target Lab value according to the real white point of the display panel comprises: converting the XYZ value into the target Lab value according to the XYZ value of the real white point of the display panel; and converting the RGB value into the predicted Lab value according to the prediction model and the XYZ value of the real white point comprises: converting the RGB value into the predicted Lab value according to the prediction model and the XYZ value of the real white point.
5. The method of claim 4, wherein, converting the XYZ value into the target Lab value according to the XYZ value of the real white point of the display panel comprises: generating an adjusted XYZ value according to the XYZ value and a Y value of the real white point; and converting the adjusted XYZ value into the target Lab value according to the XYZ value of the real white point.
6. The method of claim 4, wherein, correcting the real white point according to the prediction model to generate an XYZ value of the real white point comprises: setting an initial value of a G value to a maximum G value; adjusting R and B values so that an xy value corresponding to the predicted Lab value of the prediction model is close to an xy value of a reference white point; and outputting the XYZ value of the real white point of the display panel according to the xy value corresponding to the predicted Lab value.
7. The method of claim 6, wherein, The reference white point is determined according to a color of a light source with a color temperature of 6500K.
8. The method of claim 6, wherein, correcting the real white point according to the prediction model to generate an XYZ value of the real white point further comprises: selectively adjusting the G value according to the xy value corresponding to the Lab value and the xy value of the reference white point.
9. The method of claim 3, wherein, adjusting the RGB value according to the predicted Lab value and the target Lab value is based on whether a first error index of the predicted Lab value and the target Lab value is greater than a preset error.
10. The method of claim 9, wherein, The first error index is a mean square error, a root mean square error, or a mean absolute error.
11. The method of claim 3, wherein, A training process of the prediction model comprises: measuring an XYZ value corresponding to a representative RGB value of the display panel; calculating M second error indexes of M reference models according to the corresponding XYZ value of the display panel and predicted XYZ values of the M reference models; selecting a selected reference model having a smallest second error index from the M reference models; and According to the representative RGB values and corresponding Lab values of the to-be-tested panel, the selected reference model is trained to generate the prediction model. M is a positive integer.
12. The method of claim 11, wherein, The method further comprises: The representative RGB values are input into the M reference models to generate the predicted XYZ values.
13. The method of claim 11, wherein, The second error index is a mean square error, a root mean square error, or a mean absolute error.
14. The method of claim 11, wherein, The training process of the prediction model further comprises: The other neural network model is established, with RGB values as inputs and Lab values as outputs; The M reference models are trained according to data of M known panels for the other neural network model; and The representative RGB values are selected.
15. The method of claim 14, wherein, The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising:
16. The method of claim 15, wherein, The representative RGB values are selected by comprising:
17. The method of claim 1, wherein, The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising:
18. The method of claim 17, wherein, The representative RGB values are selected by comprising: The representative RGB values are selected by comprising:
19. The method of claim 17, wherein, The representative RGB values are selected by comprising:
20. The method of claim 17, wherein, The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising:
21. The method of claim 19, wherein, The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative RGB values are selected by comprising: The representative a fifth error index of the output RGB values corresponding to the XYZ values of the average distributed RGB values; and selecting values with the first N smallest fifth error indexes from the average distributed RGB values as the representative RGB values; wherein N is a positive integer.
22. The method of claim 20, wherein, The fifth error index is a mean square error, a root mean square error, or a mean absolute error.
23. A method of panel color wash correction, the method comprising: The method comprises: converting XYZ values into target Lab values according to a real white point of a to-be-tested panel; converting the XYZ values into RGB values using an inverse model, the inverse model being generated based on a neural network model training; converting the RGB values into predicted Lab values according to a prediction model and the real white point, the prediction model being generated based on another neural network model training; adjusting the RGB values according to the predicted Lab values and the target Lab values; and displaying a pixel by the to-be-tested panel according to the RGB values.
24. The method of claim 23, wherein, The method further comprises converting target RGB values into XYZ values.
25. The method of claim 23, wherein, The method further comprises: correcting the real white point according to the prediction model to generate XYZ values of the real white point; wherein converting the XYZ values into the target Lab values according to the real white point of the to-be-tested panel comprises: converting the XYZ values into the target Lab values according to XYZ values of the real white point of the to-be-tested panel; and converting the RGB values into the predicted Lab values according to the prediction model and the real white point comprises: converting the RGB values into the predicted Lab values according to the prediction model and XYZ values of the real white point.
26. The method of claim 25, wherein, converting the XYZ values into the target Lab values according to the XYZ values of the real white point of the to-be-tested panel comprises: generating adjusted XYZ values according to the XYZ values and a Y value of the real white point; and converting the adjusted XYZ values into the target Lab values according to the XYZ values of the real white point.
27. The method of claim 25, wherein, correcting the real white point according to the prediction model to generate XYZ values of the real white point comprises: setting an initial value of a G value to a maximum G value; adjusting R and B values so that xy values corresponding to the predicted Lab values of the prediction model are close to xy values of a reference white point; and outputting the XYZ values of the real white point of the to-be-tested panel according to the xy values corresponding to the predicted Lab values.
28. The method of claim 27, wherein, The reference white point is determined according to a color of a light source with a color temperature of 6500K.
29. The method of claim 27, wherein, correcting the real white point according to the prediction model to generate XYZ values of the real white point further comprises: selectively adjusting the G value according to the xy values corresponding to the Lab values and the xy values of the reference white point.
30. The method of claim 23, wherein, Adjusting the RGB values according to the predicted Lab values and the target Lab values is based on whether a first error index of the predicted Lab values and the target Lab values is greater than a preset error.
31. The method of claim 23, wherein, A training process of the prediction model comprises: measuring XYZ values corresponding to the representative RGB values of the to-be-tested panel; calculating M second error indicators of the M reference models according to the corresponding XYZ values of the to-be-tested panel and predicted XYZ values of the M reference models; taking out a selected reference model having a smallest second error indicator from the M reference models; and training the selected reference model according to the representative RGB values and the corresponding Lab values of the to-be-tested panel to generate the prediction model. wherein M is a positive integer.
32. The method of claim 31, wherein, The method further comprises: inputting the representative RGB values into the M reference models to generate the predicted XYZ values.
33. The method of claim 31, wherein, The training process of the prediction model further comprises: establishing the other neural network model, inputting RGB values and outputting Lab values; training the M reference models according to data of M known panels for the other neural network model; and selecting the representative RGB values.
34. The method of claim 33, wherein, The selecting of the representative RGB values comprises: selecting RGB values with average dispersion; calculating output Lab values of the M reference models according to the RGB values with average dispersion; calculating third error indicators of the output Lab values corresponding to the RGB values with average dispersion; and selecting values with top N smallest third error indicators from the RGB values with average dispersion as the representative RGB values, wherein N is a positive integer.
35. The method of claim 1, wherein, The training process of the reverse model comprises: measuring XYZ values corresponding to the representative RGB values of the to-be-tested panel; calculating M fourth error indicators of the M reference reverse models according to the representative RGB values and predicted RGB values of the M reference reverse models; taking out a selected reference reverse model having a smallest fourth error indicator from the M reference reverse models; and training the selected reference reverse model according to the corresponding XYZ values of the to-be-tested panel and the representative RGB values to generate the reverse model; wherein M is a positive integer.