TWI334928B - Mura detection method and system - Google Patents

Description

1334928 IX. INSTRUCTIONS: » [Technical Field of the Invention] The present invention is a method and system for detecting a flaw defect of a display using a 〇n machine vision, and adopts an advanced image processing and recognition technology. , can perform rapid detection of display spot defects. [Prior Art] The production process of the display is very complicated, and although most of the processes are completed in a clean room, it is still difficult to avoid some visual defects. The liquid crystal display is a type of display and is no exception. For liquid crystal displays, there are many types of visual defects, which can be classified into three types according to the area and shape of defects, including point defects, line defects, and surface defects. Surface defects can be further divided into block defects and plaques (called mura in English, which is derived from slang, meaning dirty, plaque, which is the most common type of surface defect). Point defects, line defects and block defects are mainly caused by electrical causes such as short circuit, open circuit or crystal damage of a thin film transistor (TFT) array in a liquid crystal display; the spot defects are generally due to uneven distribution of liquid crystal molecular materials or backlight Plate distortion and other non-electrical causes. Among all visual defects of liquid crystal displays, point defects, line defects, and block defects have high contrast and regular geometry, and thus are relatively easy to detect. Spot defects are the most difficult of all visual defects due to their low contrast, variable size, irregular shape, blurred edges and unfixed position. In addition, the overall liquid crystal display industry has so far not defined the flaws and the detection standards for the flaws, which have made it difficult to automatically detect the flaws. Therefore, so far, the detection of scar defects has been done by the skilled workers with the naked eye. SUMMARY OF THE INVENTION The object of the present invention is to provide a method and system for detecting scar defects in an image by using an automated image sensing display, and to directly detect scar defects on a display through advanced image processing and identification techniques. The present invention is a method for detecting a flaw defect of a display using an automated image, the method comprising the steps of: generating a display image to be tested; generating a polynomial curved surface that approximates a basic trend of the reference point by reference points and representing the reference points And subtracting the background image of the polynomial surface from the image of the display to be tested, and separating an image region whose identification error value exceeds a set threshold as a possible target of the spot defect is separated. The above-described method for detecting scar defects of a display using an automated image can include a mathematical step of Matheinat i ca 1 mcxrpho 1 ogy and a step of filtering out image noise; wherein the processing of mathematical morphology The steps are a dilation operation or an erosi〇n operation or a combination of binomial operations. In a preferred embodiment, the step of shaming the image of the display is: obtaining a plurality of towels, the average value of the image of the 51-segment display of the stomach is displayed; wherein the plurality of images are obtained The image of the display has a frequency greater than 10 frames per second and the number of acquisitions is 50 to 7 frames. In addition, the image of the display to be tested may be a side view image of the display to be tested; its side view angle is preferably between 30 and 60 degrees. In the preferred embodiment, the above-described side view image can be geometrically corrected using a bilinear interpolation method. In a preferred embodiment, the steps of 'generating a polynomial surface (four) model that approximates a reference point and can represent the basic trend of the reference points include: assuming that the grayscale value of each pixel in an image is a two-dimensional coordinate of the pixel The function, and the gray 14 value of all the pixels and its - dimensional coordinates constitute the spatial reference point _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ A polynomial surface background model that passes through the reference point and can represent the basic trend of the reference value. SUMMARY OF THE INVENTION The present invention provides a method for detecting scar defects in a display using an automated image, and further includes an identification mode step of a scar defect, the step comprising: • generating a plurality of input variables; performing a - fuzzy operation; and outputting - rotating variables. In the preferred embodiment, the input variables include contrast, area, edge parameters, positional parameters, grayscale uniformity and shape parameters; and the output variable is the level of the spot defect. The above fuzzy operation system divides the aforementioned input variables and output variables into a set number of dependent subsets, and rides a difficult subset to establish a corresponding function; the corresponding function may be a trigonometric function or a ladder number. In the preferred embodiment, the _ _ _ mode step further includes the step of inputting variables and output variables. Another object of the present invention is to provide a system for detecting scar defects in displays using automated imagery. The detection system comprises: an image generating device for generating an image of a plurality of caps to be tested; a button device for filtering out the noise of the zero image; and a surface fitting device for generating an approximation a polynomial surface background model that passes through the reference point and can represent the miscellaneous reference wire; and a separation device that subtracts the polynomial surface background model from the image of the noise removal by the rib, and sets the _ difference value to more than one setting The image area of the threshold value is separated as a possible target of the flaw defect. In a preferred embodiment, the filtering means is located between the image generating means and the surface fitting means; and after the separating means, further comprises - Mathema t i ca 1 morpho 1 〇 gy processing means. In a preferred embodiment, the detection system further includes a determining device for using a combination or combination of contrast, area, edge parameters, position parameters, gray level uniformity, and shape parameters of the target area as a system input variable. The level of the scar is used as the output variable of the system, and the aforementioned input variables and output variables are divided into a certain number of fuzzy subsets and a corresponding function is established for each fuzzy subset. In a preferred embodiment, the detecting system further comprises: a loading stage for '·-three purchases (four) tearing fines; and a hood. The stage, the three-finance platform and the _ generating device are arranged on the curved surface Stupid eight-prick (4) provides a method based on polynomial surface fitting technique, which includes: assuming a grayscale value of each pixel in an image, a function of the two-dimensional coordinates of the pixel, and all pixels The gray scale value and its two-dimensional coordinates constitute a set of spatial reference points distributed on the rectangular point; and the 70-degree polynomial performs surface fitting on the above spatial reference point, and obtains an approximation through the phase and the thief table The present invention is a method and system for detecting scar defects of a display using an automated image, and (4) advanced image processing and identification technology for directly detecting scar defects of the display. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The preferred embodiments of the present invention are described in detail below with reference to the drawings. The first figure is a schematic diagram of the overall structure of a method and system for detecting scar defects in a display using an automated image. The detection system for detecting the flaw defect of the display by the automatic image detection comprises: a three-axis precision positioning platform, a charge coupled device (CCD) camera 2, a liquid crystal display driving module 3, and a stage 4 , an image capture card 5 and a computer 6. The stage 4 is fixedly coupled to the three-axis precision positioning platform 1, and the 1334928 is transmitted through the interface of the electric lining material (4) liquid crystal display. The liquid crystal display 7 to be tested is placed on the stage 4 which can perform 1 & rotation in the vertical direction. The CCD camera 2 can be used as the image acquisition device of the liquid crystal display 7 to be tested and S1 is set on the three-reduction positioning platform , so that it can be moved under the control of the computer 6 according to the model of the liquid crystal display 7 to be tested. The corresponding position is used for image acquisition. In the preferred embodiment, the image pickup 5 is a standard analog or digital image capture card, such as a digital interface for a USB interface or a video card. The computer 6 is a master device having a quasi-counter-synchronization function, which includes devices such as a casing body, a cpu, a motherboard, a hard disk, and a display. The CCD camera 2 is connected to an input end of the image capture card 5. In the preferred embodiment, the computer 6 is provided with a system software for performing an automatic detection task, which can control the liquid crystal display driving module 3 and illuminate the liquid crystal display 7 to be tested, and cause the liquid crystal display 7 to be tested to display an initial face. When detecting the flaw defect, the stage 4 carrying the liquid crystal display 7 to be tested can be rotated by a certain angle under the control of the computer 6, so that the optical axis of the CCD camera 2 and the plane of the liquid crystal display 7 to be tested are clipped by 30 to 60 degrees. At this time, the CCD camera 2 can obtain a side view image from the side of the liquid crystal display 7 to be tested, and collect the overall image information of the liquid crystal display 7 to be tested by means of repeated sampling. Since the liquid crystal display 7 to be tested is composed of a liquid crystal display The drive module 3 is driven 'so that it can emit light itself.|All lighting equipment is required. To prevent external light interference, the three-dimensional positioning flat (XD Qing, the stage 4 11 and the liquid crystal to be tested) 7 filial piety is placed in a hood 8. The computer 6 can control the liquid crystal display driver module 3, so that the screen displayed by the liquid crystal display 7 to be tested is a grayscale image β. The second figure is an automatic image detecting display of the present invention. The method and system flow chart of the flaw defect, through which the voltage signal of the pixel on the liquid crystal display to be tested is generated, and an image acquisition step 801 is performed; in the step, the contrast of the flaw is low, In order to reduce the influence of the image noise and improve the product f of the image, the frequency of the square sampling of the sampling-receiving surface in the image capturing step 801 should generally be greater than 10 frames per second. The number of times of repeated sampling is generally 5 times to times. After the image acquisition step 801 is completed, an image chopping step 802 ': 3⁄4 occurs in the image on the image" (4) has no correlation, and has an average value. Zero random noise 'can be used · the average silk image of the random image generated under the same conditions. If the image is additive), the noise generated is as follows, then the image with noise is (4) = / (Caf At this time, the original image / (χ, β, where A / is the number of images) can be estimated. Obviously, such an estimate should have no deviation because E{g(Xiy)} = -~tf^ y) = f{x, y) <* After the image filtering step 802, the portion of the liquid crystal display to be tested is also separated from the image towel and the image is subjected to double-interpolation method_image U34928

The geometric correction is such that the liquid crystal display to be tested which becomes trapezoidal due to the CCD viewing angle in the image is restored to a standard rectangle. According to the bilinear interpolation method described above, the gray point value of the pixel point (9) in the corrected image can be calculated from the gray scale values of the four pixels around it. f 〇'»/)= [/ (/+1,y)-/(/5y)]*(/--〇+ [/(/,y+i). f(i _.. +[/( /+1, y+1)+/(/, j) -/(/+1, j) - f{i, j+1)] * (,·^ ^ 3 + ^ Force filtered image When performing flaw detection, due to liquid crystal display

The characteristics of the navigation line D (four) off, the gray level of the crystal to be examined varies greatly with the whole display, far away from the gray scale change of the flaw defect itself. Therefore, for the detection of shouting defects, the separation of the scar image is the most important and the most important part. It is difficult to perform the separation task of the scar image by using a conventional image separation method such as edge processing. Therefore, the present invention proposes a method of separating a scar image using a polynomial surface fitting.

Referring to step 8G3 of the second figure, in this step, the money assumes a grayscale value per pixel in the image / (as a function of the two-dimensional coordinate 该 of the pixel, and the grayscale value of all pixels and its two-dimensional The coordinates constitute a set of spatial reference 分布 distributed over the rectangular grid points. Next, the bivariate polynomial is subjected to the φ fitting of the above spatial reference points, that is, the estimator is approximated by the reference point and can represent the trend of the reference points. Polynomial surface: Set the XX rectangles in the known rectangular area (~, view=01,,~==1, the value of the function on the ~, assuming the polynomial to be sought is 13

, (==SSvV 1*0 7^0 where from ^"" (four) heart from (four) is a pending parameter 'it makes ~, /(χ, β the square of the difference between the values of the square grid points and the smallest two The minimum is reached by the power, that is, ~~)=||(~·||ν,)2=ηήη The parameter to be determined by the above formula is ~, and the corresponding surface is called the least squares polynomial surface. Using the multivariate function In the extreme value theory, let dl A/ 付 pay the algebraic equation of pxg order for {%} in the above formula: ,...,,-1) Solve the algebraic equation of this order and get %. Since the area of the scar defect is small 'and does not change the basic change trend of the image', the polynomial surface can be regarded as a background model without the flaw defect. The error value of the area identification without defects in the image of the liquid crystal display to be tested relative to the back model

pA qA ζ» -ΣΣ 1=0 y=〇 Very small' and the error value of the area where the spot defect is located and the background model is p*l zu - ΣΣ^ν *=0 7=0 is large. As step 804 of the second figure, using the background flip, the image of the liquid crystal display to be tested is subtracted from the background modality, and the error value exceeds the artificially set _ area as a sneak _ possible target from the background Separated from the model. The third figure is the result of the original image captured by the CCD after the above-mentioned multi-image averaging method and image correction processing steps, and there are five spot defects in the figure. The figure is the result of the processing of the background model lying in the third A_ going to the third B picture obtained by the polynomial surface fitting method for the third detail. After the above image processing, the mathematical morphology (Mathematics _hQlQgy) is used to analyze and process the shape and structure of the Newport hangar image. Analysis and processing of the image shape and architecture of the above-mentioned images are performed using dilation operations and erosion (er〇sl〇n) operations and both. The sigh collection is the input image, and the collection 5 is the structural element. If the collection Z is invaded by the collection B, it is expressed as: ^05 = {x:5 + xC^} - some independent noise points and burrs. The dilation operation is the inverse of the er〇si〇n operation and can be defined by the concept of the remainder set. The broken set β expansion is expressed as a lion, which is defined as: j10 5 = [4εΘ(-ΰ)] where, the remainder of J is expressed in the expansion system as a hole in the filled image smaller than the architectural element and at the edge of the image Small concave part. The third D picture is the result of mathematical morphology processing on the third C picture. As step 805 of the second figure, one or a combination of six parameters such as contrast, area, edge parameters, position parameters, gray level uniformity and shape parameters of the target area is used as the basis for the evaluation of the spot defect level. Wherein, the contrast parameter is defined as: c°(10)Σ peach-type neutral point; wherein /(,,y) and Qiu, _/) are the grayscale values of the possible target region and the background model at the image (^); ¢/ is the set of all the pixels in the target area # is the number of pixels in the target area. The area parameter is defined as: α = $ 1

(U)eU

U is the set of all the pixels in the target area: in the target area, the tester is to be tested, where L〇c〇tion = . And 3 are the coordinates and ordinates in the geometry of the crystal display; the cross tool of the i and the heart point is the horizontal seat b and the ordinate of the geometric center point of the moonlight target area, and has:

X

N Σ: Oj)^u ～ s and the ordinates and ordinates in the cross-station of the possible target (u) pixel points respectively ·- Edge parameter _ hetero-Pras equation definition: ^ Σ|8/(〇· )-/(/,7-1)-/(/-1,y)-/ay+1) where (/ is the set of all boundary points of the target area. The meaning of the ^ parameter is · coffee pe = In the max{ii, C} formula, 7? is the circularity of the target area, which describes the complexity of the object boundary, which can be expressed by the relationship between the area of the target area and the circumference. The mathematical expression of the circularity is: R = 4π·χTM corpse 2 where *5 is the area of the target area, /> is the perimeter of the target area. C is the squareness of the target area, using the ratio of the area of the target area to the area of the smallest circumscribed rectangle Description, where: S is the area of the target area 'Smer is the area of the smallest circumscribed rectangle of the target area. The gray level uniformity is described by the gray level standard deviation of the target area: uniformity = σ = i B (several, β one gas) Medium L. # %g/(IJ) ' is the grayscale average of all the pixels in the target area. In the identification mode step 8() 5, the present invention proposes to use a fuzzy set to be reasonable Na and the fuzzy logic shouting-fuzzy identification method. When designing the fuzzy identification method, we must first define the input variable and the input (four) quantity of the system. According to the better implementation of the present invention, the contrast of the possible target area , area, edge parameters, position parameters, gray level uniformity and shape parameters, etc. - or a combination of as a variable; the level of shouting defects as a rotation variable; where the lack of _ level - the age is unqualified ( N grade), qualified (p grade), good grade), the actual application can be adjusted according to the actual situation of the factory. Next, according to the experience of professional technicians, the above input variables and output variables are divided into a fixed number. Fuzzy subsets and corresponding functions for each fuzzy subset; in order to simplify the operation and improve the operating speed of the system, generally select triangles and trapezoids with simple shapes. For example, for the six input variables of the system, Make the following divisions: Divide the contrast into five levels, namely very low (VL), low (L), medium (8), high (8) and very high (VH); the area is divided into small (S), medium (M) and large (B) Level 3; the edge parameters are divided into low (L), towel (M), and high (8); the positional parameters are divided into three levels: edge (1), general (N), and center (6); gray level uniformity is divided into low (L), medium ( M) and high (8) three levels; shape parameters are divided into two levels: irregular (10)) and rule (R). The output variable of the system, mura defect #, is divided into three levels: mild mura defect (p), general mura defect (N), and severe mura defect (V). In the preferred embodiment, the corresponding functions corresponding to the input variables and output variables are represented by trigonometric functions and ladder functions, see Figure 4. The formulation of fuzzy if-then rules is the core problem of fuzzy identification methods. In accordance with a preferred embodiment of the present invention, the fuzzy if-then rules are based on expert experience. For example, the 'if-then rule' is as follows:

If (Const is H) and (Area is M) and (Edge is H) then (mura level is V), using fuzzy recognition method, the system can imitate the way of human identification and make full use of the experience and knowledge of the special 'completed pair The spot defect of the liquid crystal display is automatically recognized. Although the above embodiment has been described by taking a liquid crystal display as an example, the present invention is not limited to the detection of the liquid crystal display of the present invention, and the defect detection of the display using other display technologies can also be applied. The above is only the preferred embodiment of the present invention and is not intended to limit the scope of the present invention. Equivalent variations and modifications made to the invention are encompassed by the scope of the invention. 1334928 [Simple diagram of the diagram] The diagram of the overall architecture of a system using the automated image detection display for the flaw defect of the present invention is a flow of a method for applying the automatic flaw detection method to detect the flaw defect of the bribe Figure 3A ® is the result of the multi-image averaging method and image correction processing of the original image acquired by CCD. The ^® system is a non-stained mark produced by polynomial surface fitting method on the third eight figure. The background model of the defect The third C® system is the third eight figure minus the background result of the background model shown in the third figure. The third 1) the picture system is the fourth result of the mathematical morphology of the first picture. Figure system, the fuzzy identification system input variable and _ output variable corresponding to the function example [main component symbol description] 1 three-axis precision positioning platform 2 CCD camera 3 LCD display 1 § drive module 4 stage 20 1334928

5 Image Acquisition ‘Card 6 Computer 7 LCD Monitor to Be Tested 8 Hood 21

Claims (1)

1334928 Door Year/Day Correction Replacement Page X. Patent Application Range: ^ 1. A method for detecting a flaw defect in a display using an automated image, the method comprising the steps of: generating an image of a display to be tested, the display to be tested The image includes a plurality of pixels; generating a polynomial surface background model that approximates one of the basic trends of the reference points and can represent the reference points; The image of the display to be tested is subtracted from the background model of the polynomial surface, and the image region whose identification error value exceeds a set threshold value is taken as a possible target of the flaw defect and separated. 2. The method of claim 1, further comprising a processing step of a mathematical morphology. 3. The method described in claim 2, wherein the mathematical morphology processing step is an expansion (dilati〇n) operation, an invasive rhyme (_i〇n) operation, or a combination of expansion and invasion. . 4. The method of claim 1, further comprising the step of removing noise from the image. 5. The method of claim 1, wherein the step of generating an image of the display to be tested is to obtain an image of the plurality of displays to be tested. 6. The method of claim 5, wherein the step of removing the noise of the image is an average value, the average being an average of the images of the plurality of images to be tested. 22 September/Day Amendment Replacement Page 7. If the application is ambiguous, the cap—the frequency of the image to be tested is greater than 1 //sec. 8. The method of claim 5, wherein the number of images of the plurality of images to be tested is from 50 to 70. 9. as claimed in any of claims VIII to 8. The method of the display to be tested is a side view image of the display to be tested, wherein the angle of the side view image of the display to be tested is 30 to 60 degrees. 11. The method of claim 9, further comprising geometrically correcting the side image of the navigation H by a linear interpolation method. 12. The method of claim 2, wherein the step of generating the polynomial surface model that approximates the base trend of the reference points and representing the basic trend of the reference points comprises: hypothesis - each pixel in the image The gray scale values are functions of the two-dimensional coordinates of the pixel, and the gray scale values of all the pixels and their two-dimensional coordinates constitute a set of spatial reference points distributed on the rectangular grid points; and the space of the above is adopted by the -binary polynomial The reference point is surface-fitted, and the polynomial surface background model that approximates the reference point and can represent the basic trend of the reference points is obtained. 13. The method of claim 1, wherein the method further comprises a fuzzy identification step of the flaw defect. 14. The method of claim 13 wherein the fuzzy identification step of the flaw defect comprises: generating a plurality of input variables; performing a fuzzy operation; and rotating a round of out variables. 15. The method of claim 14, wherein the plurality of rounds of inclusions include contrast, area, edge parameters, positional parameters, ash P ratio uniformity, and shape parameters. 16. The method of claim 14, wherein the round-off variable is a grade of a scar defect. The method of claim 14, wherein the fuzzy operation is to divide the input variable and the output variable into a certain number of fuzzy subsets' and establish a corresponding function for each fuzzy subset. . The method of claim 17, wherein the function is a trigonometric function or a trapezoidal function. 19. If the application for the wire described in item 13 is applied, the fuzzy identification step of the towel missing includes further setting the plurality of input variables and the output variable. 20. A system for detecting a flaw defect of a display by using an automated image detection method for detecting a flaw defect on a display to be tested, the system comprising: an image acquisition device for generating an image of the plurality of displays to be tested, the plural The image of the display to be tested includes a plurality of pixels; a filtering device for filtering out the noise of the image; 24 1334928 is a year/day correction replacement page-surface fitting device for generating an approximate pass Referring to a polynomial surface background model of the basic trend of some reference points; and a sub-device for subtracting the above-mentioned polynomial surface background model from the image of the filtered noise removal, and a recognition error value exceeding An image area in which the value of the ceremonial value is set as a possible target of the smear defect. The system of claim 20, wherein the chopper device is disposed in the image generating device and Between surface fitting devices. 22. The system of claim 20, further comprising a mathematical morphology processing device, wherein the separation device is coupled to the mathematical morphology processing device. 23. The system of claim 20, further comprising a judging panel for contrasting, area, edge parameters, positional parameters, gray level uniformity, and shape parameters of the target area - or Combine as the wheeled variable of the system' where the level of the spot defect is the output variable of the system, divide the wheel and the output variable into a certain number of fuzzy subsets, and establish a corresponding one for each fuzzy subset. function. 2\ The system of claim 2, further comprising a carrier for carrying the display to be tested. 25. The system of claim 20, further comprising an r-axis platform for positioning the image generating device. 26. The system of claim 20, further comprising a shading 25 T) year q month/day correction replacement mask" the stage, the triaxial positioning platform and the image generating device are disposed Inside the hood. 27. A method for separating a scar image based on a polynomial surface fitting technique, the method comprising the steps of: assuming that a grayscale value of each pixel in an image is a function of a two-dimensional coordinate of the pixel, and all pixels The gray scale value and its two-dimensional coordinates constitute a set of spatial reference points distributed on the rectangular grid points; and the surface fitting points of the above spatial reference points are fitted by a bivariate polynomial, and the reference 4 is approximated by the reference point and can represent the A polynomial surface of the basic Zhao potential of some reference points. "Year of the month / 曰 correction replacement page 1334928 XI, schema: First picture 27 1334928 f~~**—-··— -— · ^ Corrected replacement page I "· ' .·_ ---I I 28 1334928 1334928