TWI494549B - A luminance inspecting method for backlight modules based on multiple kernel support vector regression and apparatus thereof - Google Patents

Description

Backlight module luminance detection method and detector using multi-core support vector regression

The invention relates to a backlight module luminance detection method, in particular to a backlight module luminance detection method using multi-core support vector regression.

Today's display technology has become significantly different from earlier methods such as cathode ray tubes. Especially after the LED diode and liquid crystal display technology have made important progress, the popularity of flat panel displays has been greatly improved.

However, there is a problem in the application of the liquid crystal panel as a flat panel display. Since the liquid crystal panel needs to change the color displayed after receiving the light through the internal backlight module, the correctness of the brightness reflection can be regarded as a backlight mode. An important indicator of group quality performance. However, in terms of production line quality control, there are still many insurmountable obstacles to the detection of the brightness of all backlight modules on the production line.

The conventional backlight module detection method is repeated by a luminance colorimeter. The ground moves on each detection position on the backlight module to obtain the luminance value of the detection target. Therefore, the detection method of the conventional technology appears to be in vain, and the detection speed of the prior art is much lower than the manufacturing speed of the front end, so that the conventional knowledge The detection technology cannot be properly applied to the production line of the backlight module to test all products.

A recent method for detecting the brightness of a backlight module is to use a light sensing device to correspond to a backlight module, and the prior art uses the luminance measurement to measure the actual luminance of each detection position on the backlight module, and then optically The sensing device detects the grayscale values at the detection positions, and inputs the data into a neural network system for system training, and then estimates the luminance of the backlight module to be measured according to the internal data of the system.

However, the detection mode of the neural network system of the prior art has a long training time, and the importance of all the training materials is regarded as equal, and the weight value of the training data cannot be adjusted according to the judgment of the examiner. The training and detection process of the neural network can not be flexible and adjustable. Therefore, it is difficult for the conventional technology to accurately reflect the true luminance. Therefore, how to quickly and accurately verify the luminance of each detection position on the backlight module, It has always been a problem that the industry is looking forward to.

The object of the present invention is to provide a backlight module luminance detection method and a detector using multi-core support vector regression, which simultaneously captures a plurality of pixel points of a plurality of detection positions on a backlight module to obtain gray scale values of respective pixel points. And through a series of calibration steps and the amount of luminance colorimeter The training samples of each detection position are established, and according to the training samples, a plurality of core functions for calculating the luminance values are generated by using the Holstein transformation, and the prediction of the luminance is established by modifying the weight variation values of the plurality of core functions. The model can complete the brightness detection of the entire backlight module by one shot in the detection mode.

According to an embodiment of the method aspect of the present invention, a backlight module luminance detection method using multi-core support vector regression includes the following steps: defining a plurality of backlight modules, wherein each backlight module includes a plurality of detection positions, each The detection location further includes a plurality of pixel points, and each pixel point has a corresponding one of the grayscale values and a darkfield value. The backlight modules of the selected part are a first group, and the backlight modules of the other part are a second group. The first group is placed in the plurality of mold cavities, and one of the backlight modules corresponds to a cavity. A camera unit is used to capture each backlight module of the first group, and grayscale values and dark field values corresponding to the respective pixel points are captured. Dark field correction is performed using gray scale values and dark field values corresponding to respective pixel points of the first group. The result of the dark field correction is subjected to flat field correction, and a gray scale correction value corresponding to one of the respective pixel points in the first group is obtained. A first average of one of the grayscale correction values corresponding to each of the detected positions is calculated. One of the actual detection values of each detection position is measured. A first average value corresponding to each detected position and an actual luminance value are established as a training sample. A prediction model is generated by using a Hawkes transform to generate at least one core function corresponding to those training samples, and inputting those training samples and core functions to a multi-core support vector regression system. The first group is taken out, and the second group is placed in those cavities, and one of the backlight modules corresponds to a cavity. Using the camera to capture the backlights of the second group The module selects the grayscale value and the dark field value corresponding to each pixel. Dark field correction is performed using gray scale values and dark field values corresponding to respective pixel points of the second group. The result of the dark field correction is subjected to the flat field correction, and the gray scale correction value corresponding to one of the respective pixel points in the second group is obtained. A second average of one of the grayscale correction values corresponding to each of the detected positions in the second group is calculated. The predicted luminance values are calculated using one of the second average values, the core function, and the prediction model in the second group.

It can be seen from the foregoing embodiment that after the first average value and the actual luminance value of each detection position in the first group are respectively obtained, a plurality of sets of training samples corresponding to the respective detection positions may be output, and the fire samples are transmitted according to the training samples. The transformation generates a core function, and the aforementioned training samples and the core function are input into the multi-core support vector regression system to generate a prediction model, and then the second average value corresponding to each detection position in the second group is used to calculate the prediction. Brightness value.

In the foregoing method aspect, the backlight module luminance detecting method using the multi-core support vector regression, wherein the imaging device can be a photosensitive coupling element. The backlight module luminance detection method using multi-core support vector regression, wherein the actual luminance value can be measured by a luminance colorimeter. In the embodiment of the foregoing method aspect, the core function is a binary one-time equation, and it has a slope of a positive number. The foregoing prediction model may include a two-Lagrangian multiplier, a weighting variable, and a constant, wherein the weighting variable corresponds to the core function, and the value of the weighting variable provides a detector as a core function for adjusting its corresponding The proportion. The core model can be greatly improved by using the prediction model generated by the multi-core support vector regression system. The efficiency of the luminance detection of the optical module, and when there are multiple sets of core functions and combinations of prediction models coexist, the detector can also adjust the ratio of the weight variables corresponding to each core function according to the actual situation, which is higher Flexibility and accuracy of detection.

According to an embodiment of the structural aspect of the present invention, a backlight module luminance detector using multi-core support vector regression includes an imaging device, a measuring device, a dark field correcting module, and a flat field correcting mode. A group, a training data module, a core function generation module, a multi-core support vector regression system, and a luminance prediction module. The camera device is configured to capture a plurality of backlight modules. The backlight module has a plurality of detection positions, and each detection position has a plurality of pixel points. The camera device captures each detection position on each backlight module, and captures each detection position. One of the pixel points corresponds to a grayscale value and a dark field value. The measuring device is configured to measure an actual luminance value at each of the detected positions. The dark field correction module signal is connected to the camera device and the dark field is corrected for each gray scale value and each dark field value captured by the camera device. The flat field correction module signal is connected to the dark field correction module, and the correction result of the dark field correction module is extracted to generate gray scale correction values corresponding to the respective pixel points, and then the respective gray scale corrections included in the respective detection positions are calculated. One of the first average or a second average. The training data module signal is connected to the flat field correction module and the measuring device, and outputs respective average values corresponding to the respective detection positions and respective actual luminance values as a training sample. The core function generates a module signal connection training data module, which captures the training samples in the training data module, and generates at least one core function corresponding to those training samples according to the training samples. Multi-core branch The aid vector regression system signal connects the training data module and the core function generation module to respectively retrieve the training samples and the core functions, and generate a prediction model. The luminance prediction module system is connected to the multi-core support vector regression system and the flat field correction module, and obtains the prediction model, the core function, and the second average values from the foregoing two, respectively, and according to the prediction model, the core function, and those The two average values calculate the predicted luminance values corresponding to the respective detected positions.

According to the implementation of the foregoing structural aspect, the detector can pre-shoot the backlight module in the training mode to obtain the grayscale value and the dark field value corresponding to each pixel point on the backlight module, and pass the dark field correction module. The flat field correction module and the training data module obtain a first average value of the plurality of gray scale correction values included in each detection position, and the actual luminance values of the respective detection positions are measured by the measuring device to obtain corresponding detection positions. After training the sample, and in the subsequent detection process, the brightness detection of the backlight module is completed in one shot.

In the foregoing structural aspect, a backlight module luminance detector using multi-core support vector regression, wherein the imaging device can be a photosensitive coupling element. The aforementioned backlight module luminance detector using multi-core support vector regression, wherein the actual luminance value can be measured by a luminance colorimeter. In the embodiment of the foregoing structural aspect, the core function is a binary one-time equation, and it has a slope of a positive number. The foregoing prediction model may include a two-Lagrangian multiplier, a weighting variable, and a constant, wherein the weighting variable corresponds to the core function, and the value of the weighting variable provides a detector as a core function for adjusting its corresponding The proportion. With core functions with multiple cores The prediction model generated by the support vector regression system can greatly improve the efficiency of the luminance detection of the backlight module, and when there are multiple sets of core functions and a combination of prediction models coexist, the detector can also adjust the corresponding according to the actual situation. The ratio of the weight variables of each core function has both high flexibility and accuracy of detection.

100‧‧‧Backlight module luminance detection method using multi-core support vector regression

A01~A16‧‧‧Steps

i ‧‧‧Backlight module

c ‧‧‧Detection location

j ‧‧‧ pixels

I _j ‧‧‧ gray scale value

G1‧‧‧First Group

G2‧‧‧ second group

M ‧‧‧ core functions

w _m ‧‧‧weight variable

K _m ‧‧‧ core function

b ‧‧‧ constant

‧‧‧set

g ‧‧‧Slope

p ‧‧‧ intercept

x _i , x _j ‧‧‧Operator

ρ , θ ‧‧‧ polar coordinates

q ‧‧‧ cavity

‧‧‧set

D _j ‧‧‧ dark field value

‧‧‧First average

f _j ‧‧‧ flat field correction function

‧‧‧ actual luminance value

‧‧‧second average

‧‧‧Predicted luminance values

l ‧‧‧Number of backlight modules

α _i , α _i ^* ‧‧‧Lagrangian multiplier

200‧‧‧Backlight module luminance detector using multi-core support vector regression

210‧‧‧ camera

220‧‧‧Measurement device

230‧‧‧ Dark Field Correction Module

240‧‧‧ Flat Field Correction Module

250‧‧‧ Training Data Module

260‧‧‧ core function generation module

270‧‧‧Multicore Support Vector Regression System

280‧‧‧luminance prediction module

1A and 1B are flowcharts showing the steps of the backlight module luminance detection method using multi-core support vector regression according to an embodiment of the method of the present invention.

FIG. 2 is a schematic diagram showing a backlight module and a cavity thereof of the backlight module luminance detection method using multi-core support vector regression in FIG. 1 .

FIG. 3 is a schematic diagram showing a backlight module and an imaging device of a backlight module luminance detection method using multi-core support vector regression in FIG. 1 .

Fig. 4 is a schematic diagram showing the luminance measurement of the backlight module luminance detection method using multi-core support vector regression in Fig. 1.

FIG. 5 is a schematic diagram showing the core function of the backlight module luminance detection method using multi-core support vector regression in FIG. 1 .

Fig. 6 is a diagram showing the relationship between the Cartesian coordinates and the core function of the backlight module luminance detection method using multi-core support vector regression in Fig. 1.

Figure 7 is a diagram showing the polar coordinates and core functions of the backlight module luminance detection method using multi-core support vector regression in Figure 1.

Figure 8 is a diagram showing the use of a multi-core according to an embodiment of the present invention. A block diagram of the backlight module luminance detector of the heart support vector regression.

1A and 1B are flowcharts showing the steps of the backlight module luminance detecting method 100 using multi-core support vector regression according to an embodiment of the method of the present invention. FIG. 2 is a schematic diagram showing a backlight module and a cavity thereof of the backlight module luminance detecting method 100 using multi-core support vector regression in FIG. 1 . Please refer to Figure 1 and Figure 2 together. The method includes steps A01 to A16, wherein step A01 defines a plurality of backlight modules i , and i N , each backlight module i each includes a plurality of detection positions c , and c N , each detection position c further includes a plurality of pixel points j , and j N , each pixel point j corresponds to a gray scale value I _j and a dark field value D _j . As shown in step A02, a portion of the backlight module i is selected as a first group G1, and another portion of the backlight module i is a second group G2. Step A03 is to place each backlight module i included in the first group G1 in a cavity q , and q N , one of the backlight modules i corresponds to a cavity q .

FIG. 3 is a schematic diagram showing the backlight module i and the imaging device 210 of the backlight module luminance detecting method 100 using multi-core support vector regression in FIG. 1 . With reference to FIG. 3, step A04 based imaging using an imaging device each of the backlight module 210 i of the first group G1, and retrieve the corresponding grayscale value of each pixel j I _j and D _j dark field value, detail For example, the display image of each detection position c on the backlight module i is composed of a plurality of pixel points j included therein, and the image capturing device 210 captures each pixel point j and simultaneously captures the gray scale values I _j and a dark field value D _j , further, a set of internal pixel points j of each detection position c may Express, for example Indicates that the backlight module i numbered 2 is placed in the cavity q numbered 1, and the backlight module i numbered 2 (i.e., the i value here is 2) is the third detection position c inside. A collection of all pixel points j that are included.

Next, step A05 extracts each gray scale value I _j and dark field value D _j captured through step A04, and performs dark field correction accordingly. Step A06 performs flat field correction on the result of the dark field correction in step A05, and generates gray scale correction values corresponding to the respective pixel points j . Step A07 is to take a first average of one of the grayscale correction values included in each detection position c . And can be expressed by the following formula (1), wherein Express The number of pixel points j included, f _j is a flat field correction function, and the result of dark field correction in steps A04 to A05 is max( I _j - D _j , 0) shown in (1), Some programs such as dark field correction and flat field correction are well-known techniques in the related art, and therefore will not be described in detail here.

FIG. 4 is a schematic diagram showing the luminance measurement of the backlight module luminance detecting method 100 using multi-core support vector regression in FIG. 1 . Referring to FIG. 4, step A08 measures an actual luminance value of each detection position c by a measuring device 220. . Step A09 is to establish a first average value corresponding to each of the foregoing detection positions c . Actual luminance value Become a training sample. Step A10 is to generate at least one core function K _m corresponding to the training samples by Hooby conversion, and input the training samples and the core function K _m to a multi-core support vector regression system 270 to generate a prediction model. The formula is expressed as follows (2), Where l is the number of backlight modules, α _i and α _i ^* are Lagrangian multipliers, M is the number of core functions, w _m is the weight variable, K _m is the core function, b is a constant, in the above formula Predicting a luminance value for one of the samples to be tested in the second group G2, one of which is a predicted luminance value Corresponding to the second average value of a detection position c in the second group G2 The foregoing prediction model includes a Lagrangian multiplier α _i and α _i ^* , a weight variable w _{m ,} and a constant b . In the multi-core support vector regression system 270, a plurality of training samples and a core function K _m are It is used to solve various parameters included in the above prediction model, and the solution of the prediction model and internal parameters here is used by the conventional multi-core vector support device, so it will not be described in detail here.

FIG. 5 is a schematic diagram showing the core function K _m of the backlight module luminance detecting method 100 using multi-core support vector regression in FIG. 1 . For a detailed description of the core function K _m generated by Holstein conversion in the present embodiment, please refer to FIG. 5, ,then L represents a set of the i th backlight module q cavity within respective detected position c of the training samples, for example, A set of training samples representing the first detection position c on the 80 backlight modules i measured inside the first cavity q . In Fig. 5, the horizontal axis represents the first average value corresponding to each detection position c The vertical axis indicates the corresponding first average Actual luminance value Therefore, each point in Figure 5 represents a training sample.

FIG 6 illustrates a system using a first map of the multi-core support vector regression luminance of the backlight module Cartesian coordinate detection function 100 of the core K _m diagram. FIG 7 illustrates the use of a first line of FIG plurality of core support vector regression method for detecting the luminance of the backlight module and the polar core 100 of the function K _m diagram. Please refer to Figure 6 and Figure 7. As shown in Figure 5, each training sample is represented by a point on the Cartesian coordinate in Figure 6 (for ease of understanding, only three training samples are shown in the figure). As an illustration), and the set of all training samples is Where x _i and y _i correspond to the first average in Figure 5, respectively Actual luminance value In this embodiment, the core function K _{m is} obtained by the Hawker transform, which converts the full linear function of each point in FIG. 6 into the polar coordinates of the seventh figure, and the straight line functions satisfy ρ = x _i cos θ + y _i sin θ , as can be seen from general mathematical knowledge, in the case of the same set of x _i and y _i , all the ( ρ , θ ) coordinate systems satisfying the condition in Fig. 7 form a sine wave function, therefore, Through all the straight line functions of the three training samples in Fig. 6, the system is divided into three groups and is transformed into three sine wave functions in Fig. 7.

In the case of the Cartesian coordinates of Fig. 6, all the straight lines passing ( x _i , y _i ) satisfy ρ = x _i co co θ + y _i sin θ , so these straight lines can be expressed by the following formula (3), 6 figure as an example, i =1, 2, 3, It can be seen from equation (3) that these straight lines can be regarded as a plurality of binary one-time equations, and the slope and the intercept are affected by ρ and θ . It can be clearly observed in Fig. 6 that three points are The collinearity may occur, so when a particular binary equation passes through three points at the same time, it can be known that this ( ρ , θ ) is in the polar coordinates of Figure 7, and must all pass three sine wave functions, that is, ( ρ , θ ) is the intersection of three sine wave functions. Referring again to Figure 5, using this characteristic of the Hawker transformation, when more training samples are passed by the same binary equation, the corresponding two The meta-first equation is a good description of the first average of multiple training samples. Actual luminance value The relationship, that is to say, this one or a plurality of binary one-time equations has a high reference value, and can be selected as the core function K _m required for detecting the luminance, and can be expressed by the equation (4).

K _m ( x _i , x _j )=( g .< x _i , x _j 〉+ p ) (4); where g is the slope, p is the intercept, and x _i and x _j are the operations of the kernel function K _m Yuan, which corresponds to the first average And a second average It should be noted that, as shown in Figures 5 and 6, the first average value in multiple training samples under reasonable natural conditions is known. Actual luminance value Both are positively correlated, so in equation (3), only the slope g is considered to be positive.

In step A11, the first group G1 located in each cavity q is taken out, and those backlight modules i of the second group G2 are placed inside each cavity q , and one backlight module i corresponds to one cavity q . Steps A12 to A15 repeatedly perform the same processing procedure as step A04 to step A07 for the second group G2, and obtain one of the grayscale correction values included in each detection position c corresponding to the second group G2. Two average . Step A16 is to substitute the prediction model generated by the foregoing step A10 and the core function K _m into the equation (2) and obtain the predicted luminance value. .

Overall, the predicted luminance values inside the second group G2 are started. Before the calculation, the foregoing steps A01 to A09 are used to establish a plurality of training samples, and the step A10 generates at least one or a plurality of core functions K _m by using the training samples, and inputs those training samples. the core functions of up to K _m and the core support vector regression prediction model generating system 270, and then the step A11 to step A15 for obtaining a second average value of each pixel j of the second group G2 internal In order to obtain the predicted luminance value in step A16. .

In addition, the detector can automatically set the binary one-time equation as the core function K _m when the binary function equation is passed through a certain number of training samples in the core function generation module 260, for example, only selecting When more than 10 training samples are passed at the same time, the kernel function K _{m is} selected. At this time, the multi-core support vector regression system 270 generates a prediction model according to the selected core function K _m and those training samples.

It is worth mentioning that, in addition to the foregoing process, the plurality of core functions K _m provided by the implementation of the method aspect are calculated by the foregoing process, and the weight variable w _m calculated in the foregoing formula (2) is additionally The specific gravity value of each core function K _m can be adjusted. Further, as one of the training samples known in FIG. 5 is preferred, when one or more core functions K _m pass through the training sample when the coordinate position, the examiner can be raised so this core function or a plurality of weights corresponding K _m of variable weight w _m, and vice versa when the coordinates of a known training samples is poor or less consistent with common sense, detection Zheyi corresponding to the core and so can be cut right function _m K variables relevant to the weight w _m, or may be variable by adjusting the weights W _m K _m that directly corresponds to the core functions are excluded outer luminance detecting the selection condition.

Therefore, the foregoing embodiment may use the shooting of each backlight module to obtain a first average value corresponding to each detection position, and compare the actual luminance values of the respective detection positions to establish training samples, and then use the training samples to adopt Holmes The transformation generates one or a plurality of qualified core functions, and the prediction model is generated according to the training samples and the core function by the multi-core support vector regression system, and the weight variable of the plurality of core functions can be adjusted by the detector. Further, the detection result is closer to the actual actual luminance value.

The image pickup device 210 used in the foregoing embodiment may be a photosensitive coupling element, and is not limited thereto, and in addition, the actual luminance value is measured. The measuring device 220 can be a luminance colorimeter, and is not limited thereto.

FIG. 8 is a block diagram showing the structure of a backlight module luminance detector 200 using multi-core support vector regression according to an embodiment of the present invention. Referring to FIG. 2 and FIG. 8 together, the backlight module luminance detector 200 using multi-core support vector regression includes an imaging device 210, a measuring device 220, a dark field correction module 230, and a flat field correction mode. a group 240, a training data module 250, a multi-core support vector regression system 270, and a luminance prediction module 280, wherein the camera device 210 is configured to capture a plurality of backlight modules i , and i N , wherein each backlight module i is placed in a corresponding one of the cavities q , and q N , and the backlight module i has a plurality of detection positions c , and c N , each detection position c has a complex pixel point j , and j N , wherein each pixel point j corresponds to a gray scale value I _j and a dark field value D _j . In detail, the display image of each detection position c on the backlight module i is composed of a plurality of pixel points included therein. when j is constituted, with reference to FIG. 2 and FIG. 3, each of the imaging apparatus 210 imaging backlight module i, while capturing each pixel grayscale value j I _j D _j and the dark-field values, further, each of the detection The set of internal pixel points j of position c can Express, for example Indicates that the backlight module i numbered 2 is placed in the cavity q numbered 1, and the backlight module i numbered 2 (i.e., the i value here is 2) is the third detection position c inside. A collection of all pixel points j that are included. Referring to FIG. 4 again, the measuring device 220 is configured to measure an actual luminance value at each detecting position c . . The dark field correction module 230 signals the camera device 210, and the dark field corrects the gray scale values I _j and the respective dark field values D _j captured by the camera device 210. Referring to the formula (1), the flat field correction module 240 is connected to the dark field correction module 230 and calculates the complex gray corresponding to the pixel points j by extracting the dark field correction result output by the dark field correction module 230. The order correction value, and the gray scale correction values are averaged according to the corresponding detection position c to become a first average value , refer to the formula (1), where Express The number of pixel points j included, f _j is a flat field correction function, and the dark field correction and flat field correction used herein are well-known techniques in the related art, and will not be described in detail here. Referring to FIG. 5, the training data module 250 is connected to the flat field correction module 240 and the measuring device 220, and outputs respective first average values corresponding to the respective detection positions c . And the actual luminance values Is a training sample, so in Figure 5, the first average of each training sample Actual luminance value It can be represented by a dot. The core function module 260 signals the training data module 250 and retrieves the training samples inside the training data module 250, and then generates at least one core function corresponding to those training samples according to the training samples according to the Hoech transform. K _m , and the core function K _m is a binary one-order equation with a positive slope. Here, the generation and characteristics of the core function K _m are the same as those of the foregoing method aspect of the present invention, and therefore will not be described. The multi-core support vector regression system 270 signals the training data module 250 and the core function module 260 and separately retrieves the training samples and the core functions, and thereby generates one of the prediction models for the luminance prediction use, where the prediction model is Including the Lagrangian multipliers α _i and α _i ^* , the weight variable w _m and the constant b , the prediction model is generated in the same manner as the above-described method aspect embodiment of the present invention, and is used by a conventional multi-core vector support machine. Therefore, it is not detailed. The luminance prediction module 280 is connected to the multi-core support vector regression system 270 and the flat field correction module 240. The luminance prediction module 280 has a prediction mode. Here, in the non-predicted state, the flat field correction module 240撷The dark field correction result output by the dark field correction module 230 is taken, and the flat field correction result corresponding to each detection position is output, and the flat field correction results are averaged to become a first average value. When the luminance prediction module 280 is in the prediction mode, the flat field correction module 240 performs the same steps as described above, but the output is a second average. At this time, the luminance prediction module 280 retrieves the prediction model, the core function K _{m ,} and the second average from the multi-core support vector regression system 270 and the flat field correction module 240, respectively. And based on the prediction model, the core function K _m and those second averages Calculating a predicted luminance value corresponding to one of the respective detection positions c , here predicts the luminance value The calculation is the same as the foregoing embodiment of the method aspect of the present invention.

According to one embodiment of the structural aspect of the present invention, the detector can first establish a training sample by using the camera device and the dark field correction module, the flat field correction module, and the training data module before performing the luminance detection of the backlight module. The core function generation module generates a core function, and uses the multi-core support vector regression system to generate a prediction model based on the training sample and the core function, thereby, in the subsequent detection, the luminance prediction mode can be performed in one shot. The group obtains the predicted luminance values corresponding to the respective second average values through the core function and the prediction model.

In addition, the imaging device 210 used in the foregoing embodiment may be a photosensitive coupling element, and is not limited thereto, and in addition, the actual luminance value is measured. The measuring device 220 can be a luminance colorimeter, and is not limited thereto. The kernel function K _m is calculated by the foregoing process, and the weight variable w _m in the above formula (2) can adjust the specific gravity value of each core function K _m , and the adjustment is based on the method described in the present invention. In the case of the embodiment, it will not be repeated.

It can be seen from the above embodiments that the present invention has the following advantages: First, the multi-core support vector regression detection method can improve the conventional detection method, and each test needs to reciprocate on each detection position on the backlight module to detect each time. The time-consuming shortcoming of luminance, the invention can complete the luminance detection of a plurality of backlight modules in one shot after establishing the luminance prediction model The measurement effectively improves the efficiency of detection. Secondly, compared with the conventional neural network system, the invention requires shorter training time and is flexible and suitable for use in the production line, and the detection result is also More accurate than the previous method, it can be applied to the quality control of the backlight module production line. Thirdly, the present invention can automatically select a plurality of core functions according to the conditions set by the user to calculate the luminance values of the respective detection positions, and allow the user to adjust the core function according to the gold sample of the production line. The weighting variable makes the predicted luminance value more accurate to the actual situation.

Although the present invention has been disclosed in the above embodiments, it is not intended to limit the present invention, and the present invention can be modified and modified without departing from the spirit and scope of the present invention. The scope is subject to the definition of the scope of the patent application attached.

210‧‧‧ camera

q ‧‧‧ cavity

i ‧‧‧Backlight module

c ‧‧‧Detection location

Claims (14)

A backlight module luminance detection method using multi-core support vector regression includes the following steps: defining a plurality of backlight modules, each of the backlight modules including a plurality of detection positions, each of the detection positions including a plurality of pixel points, each of the pixel points having a corresponding a grayscale value and a dark field value; the backlight modules of the selected portion are a first group, and the backlight modules of the other portion are a second group; the first group is placed in a complex mode In the hole, one of the backlight modules corresponds to the cavity; the backlight module of the first group is captured by a camera device, and the grayscale value corresponding to each pixel point and the dark field are captured a dark field correction using the grayscale value corresponding to each pixel of the first group and the dark field value; performing flat field correction on the result of the dark field correction, and obtaining the corresponding corresponding in the first group a grayscale correction value of each of the pixels; calculating a first average value of the grayscale correction values corresponding to each of the detection positions; measuring an actual luminance value of each of the detection locations; establishing corresponding to each detection The first level of position And the actual luminance value is a training sample; generating at least one core function corresponding to the training samples by using the Hawker transformation, and inputting the training samples and the core function to a multi-core support vector regression system to generate one Predictive model; Taking out the first group, and placing the second group in the cavity, one of the backlight modules corresponding to the cavity; using the camera to capture the backlight module of the second group, And capturing the grayscale value corresponding to each pixel point and the dark field value; performing dark field correction by using the grayscale value corresponding to each pixel point of the second group and the dark field value; Performing a field correction on the result of the field correction, and obtaining a grayscale correction value corresponding to one of the pixels in the second group; calculating the grayscale correction values corresponding to each of the detection locations in the second group a second average value; and calculating, by using each of the second average values in the second group, the core function, and the prediction model, a predicted luminance value of each of the detected positions in the second group. The backlight module luminance detecting method using the multi-core support vector regression of claim 1, wherein the imaging device is a photosensitive coupling element. The backlight module luminance detection method using multi-core support vector regression of claim 1, wherein the actual luminance value is measured by a luminance colorimeter. The backlight module luminance detection method using multi-core support vector regression of claim 1, wherein the core function is a binary one-time equation. The backlight module luminance detection method using multi-core support vector regression of claim 4, the binary one-time equation has a slope, and the slope is a positive number. The backlight module luminance detection method using multi-core support vector regression of claim 1, wherein the prediction model includes two Lagrangian multipliers, a weight variable, and a constant. The backlight module luminance detection method using multi-core support vector regression of claim 6, wherein the weight variable corresponds to a core function, and the weight variable can be regulated by a detector. A backlight module luminance detector using multi-core support vector regression includes: a camera device for capturing a plurality of backlight modules, wherein each of the backlight modules includes a plurality of detection positions, each of the detection positions including a plurality of pixel points, The camera device captures each of the detection positions of each of the backlight modules, and captures one grayscale value and one dark field value corresponding to each pixel of each detection position; a measuring device, which measures each Detecting an actual luminance value of a position; a dark field correction module, wherein the signal is connected to the camera device and the dark field corrects each grayscale value and each dark field value captured by the camera device; a flat field correction module The signal is connected to the dark field correction module, and the flat field correction module captures the correction result of the dark field correction module and generates corresponding Each of the grayscale correction values of the pixel points, and then calculating a first average value or a second average value corresponding to each of the grayscale correction values in each of the detection positions; a training data module, and the signal connection The flat field correction module and the measuring device output a first average value corresponding to each of the detection positions and each of the actual luminance values as a training sample; a core function generating module, wherein the signal is connected to the training data The module captures the training samples and generates at least one core function corresponding to the training samples by using a Holker conversion; a multi-core support vector regression system, wherein the signal is connected to the training data module and the core function Generating a module and separately extracting the training samples and the core function, and generating a prediction model; and a luminance prediction module, wherein the signal is connected to the multi-core support vector regression system and the flat field correction module respectively The prediction model and the second average values are used to calculate a predicted luminance value of each of the detection positions in each of the backlight modules according to the prediction model and the second average values. The backlight module luminance detector of claim 8 which utilizes multi-core support vector regression, wherein the camera device is a photosensitive coupling element. A backlight module luminance detector using multi-core support vector regression of claim 8, wherein the actual luminance value is measured by a luminance colorimeter. A backlight module luminance detector using multi-core support vector regression of claim 8, wherein the core function is a binary one-time equation. A backlight module luminance detector using multi-core support vector regression of claim 11 has a slope and the slope is a positive number. A backlight module luminance detector using multi-core support vector regression of claim 8, wherein the prediction model includes two Lagrangian multipliers, a weight variable, and a constant. The backlight module luminance detector of claim 13 using multi-core support vector regression, wherein the weight variable corresponds to a core function, and the weight variable can be regulated by a detector.