Detection of unevenness defects in the master panel of a flat panel display during manufacturing

Abstract

To detect mura defects in a master panel during fabrication.SOLUTION: A method includes: preparing a combined image from image data of a master panel; enhancing the quality of the combined image, including removing artifacts from the combined image; filtering the enhanced quality combined image to detect local mura defects, the local mura defects including at least one structured pattern of defined geometric shapes; applying different candidate patterns to the filtered combined image; selecting one of the candidate patterns as a defect detection pattern, the defect detection pattern being the closest to the structured pattern of defined geometric shapes of the detected local mura defects; and displaying at least a portion of the defect detection pattern on a display, together with the quality-enhanced combined image.SELECTED DRAWING: Figure 1

Description

Background Art

[0001] Flat panel displays such as liquid crystal displays (LCDs) are used in many electronic devices such as mobile phones, televisions, and computer monitors. Flat panel displays typically form a thin film transistor (TFT) array on a master panel (or substrate) (the TFT array corresponds to a plurality of flat panel displays), and ultimately, the master panel is manufactured by separating it into individual flat panel displays. In the manufacturing line, a large number of master panels are processed daily. Flat panel displays are manufactured in a plurality of process steps. In each step, chemical and mechanical surface modifications occur, and some of these modifications are regarded as defects including non-uniformity defects. To ensure product quality (panel yield), the surface of the master panel must be repeatedly inspected between process steps, for example, in pixel units. The inspection can be performed using an inspection system such as the Keysight 88000 HS-100 series array test system sold by Keysight Technologies, Inc. The inspection generates data in the form of a matrix corresponding to the pixels on the master panel. The data matrix usually contains noise from a plurality of sources both inside the environment and the inspection hardware. This noise should be suppressed in order to give the user a clear image of the measurement data and enable automatic detection of panel defects.

[0002]

[0003] ​​​​​​​​​​​In particular, mura defects may appear in the master panel before separation into individual flat panel displays. For example, in the case of LCDs, mura-type defects are generally caused by process deficiencies related to cell assembly, and these defects affect the transmission of light through the flat panel display and are generally unpleasant to the viewer. An inspection system can be placed in situ to provide measurements of the flat panel display that yield information about the defects, and these measurements can be introduced at the individual manufacturing steps. For example, the Keysight 88000 HS-100 series array test system can generate single-pixel resolution charge maps of flat panel displays that can be used to identify defects. Given a large dataset, a particular defect pattern can be identified as originating from a particular production facility. However, the resulting gigapixel dataset poses difficulties for visual inspection and annotation. Therefore, there is a need for efficient automated systems and methods for detecting, classifying, and quantifying a particular type of defect, including mura defects, based on, for example, Fourier transforms. Automated detection and quality control reduce labor costs and provide further standardization. Exemplary embodiments are best understood by reading the following detailed description in conjunction with the figures of the accompanying drawings. It is emphasized that the various features are not necessarily drawn to scale.

[0004] . In practice, the dimensions can be arbitrarily increased or decreased for clarity of discussion. Applicable Wherever applicable and wherever actually useful, like reference numerals refer to like elements .

Brief Description of the Drawings

[0005]

Figure 1

Figure 2A

Figure 2B

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

[0006] In the following detailed description, for purposes of explanation and not limitation, exemplary embodiments are disclosed that set forth specific details in order to provide a thorough understanding of one embodiment of the present teachings. However, it will be apparent to those skilled in the art having the benefit of this disclosure that other embodiments according to the present teachings may depart from the specific details disclosed herein and still be within the scope of the appended claims. Moreover, descriptions of well-known devices and methods may be omitted so as not to obscure the description of the exemplary embodiments. Such methods and devices are clearly within the scope of the present teachings. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting. Defined terms are to be construed as adding to the technical and scientific meaning of these defined terms as generally understood and accepted in the technical field of the present teachings.

[0007] ​​​​​​​​​

[0008] The terms "a" and "the" as used in this specification and the appended claims include both the singular and the plural referents unless the context clearly dictates otherwise. Thus, for example, "device" includes one device and a plurality of devices. As used in this specification and the appended claims, the terms "substantially" or "substantially" in addition to their ordinary meaning refer to being within an acceptable limit or degree. As used in this specification and the appended claims, the term "generally" in addition to its ordinary meaning refers to being within an acceptable limit or amount for those skilled in the art. For example, "generally the same" means that a person skilled in the art would consider the plurality of items being compared to be the same. Relative terms such as "above", "below", "top", "bottom", etc. can be used to describe the relationship between various elements as shown in the accompanying drawings. These relative terms are intended to encompass various orientations of the elements in addition to the orientation shown in the drawings.

[0009] For example, if the drawing is inverted with respect to the display within the drawing, an element described as "above" another element, for example, would here be "below" that element. Similarly, if the device is rotated by 90 degrees with respect to the display within the drawing, an element described as "above" or "below" another element would here be "adjacent" to that other element, provided that "adjacent" means in contact with that other element or having one or more layers, materials, structures, etc. between the elements. position

[0010] In general, according to an exemplary embodiment, a non-uniformity defect in the master panel during manufacturing is detected A method of outputting, wherein the master panel includes a plurality of flat screen displays, A method is provided. The method includes preparing a composite image from the image data of the master panel and improving the quality of the composite image, including removing artifacts from the composite image and filtering the improved quality composite image to detect local non-uniformity defects wherein the local non-uniformity defects include at least one structural pattern of a specified geometric shape applying a plurality of different candidate patterns to the filtered composite image and selecting one of the plurality of candidate patterns as a defect detection pattern wherein this defect detection pattern is the defect detection pattern closest to the structural pattern of the specified geometric shape of the detected local non-uniformity defect and displaying at least a part of the defect detection patterns on the display together with the improved quality composite image to indicate the position of the detected local non-uniformity defect in the structural pattern of the specified geometric shape. Filtering the improved quality composite image to detect local non-uniformity defects may include filtering out the relevant spatial frequencies corresponding to the length scale of the detected local non-uniformity defect to provide a first filtered image and superimposing the first filtered image on a set of templates each corresponding to a specified geometric shape. can include.

[0011] FIG. 1 is a simplified block diagram of a defect detection system for detecting non-uniformity defects in a master panel including a plurality of flat panel displays during manufacturing according to an exemplary embodiment.

[0012] Referring to FIG. 1, the defect detection system 100 controls the operation of the scan head 120 ​​​​, including a controller 110 that processes the image data provided by the scan head 120 is provided. The image processing includes, for example, detecting unevenness defects by the method mentioned in FIGS. 2A and 2B . The controller 110 includes a processor 112, a memory 114, and an interface (I / F) 116 . The defect detection system 100 further includes a display 130 that displays the images and data / the results of image processing provided via the I / F 116 (including the display interface). The display 130 can incorporate a graphical user interface (GUI) that allows the user to interact with the controller 110 using the information displayed on the display 130 together with a user input device (not shown) such as a keyboard, a mouse, a touch pad, and / or a touch sensitive screen. Without departing from the scope of the present teachings, any other suitable means for providing input and receiving output information can be incorporated into the defect detection system 100. Without departing from the scope of the present teachings, the scan head 120 can include various types of scanning devices that collect data from the master panel 105 being scanned to provide the corresponding image data . For example, the scan head 120 can be an electrical contact scanner that collects data by making electrical contact with the master panel 105. The address signal and the data signal indicating the uppermost surface of the master panel 105 are conveyed through the metal probes of the head. More specifically, the electrical signals from the scan head 120 are respectively connected to the pixels on the master panel 105 in a predetermined sequence

[0013] ​ Provides pixel signals (image data). The scan head 120 can be an electro-optical scanner and can also be an optical contact scanner that functions similarly. In an alternative configuration, the scan head 120 provides data indicating the master panel 105 being scanned, for example, a camera (multiple in some cases), a scanning sensor (such as a charge-coupled device (CCD)), or data can be provided through a non-contact scanning method that includes the use of laser excitation . Also, in an alternative configuration, the scan head 120 includes a laser source, a lens, and electrical contact through a probe needle that excites the photoconductive current in the TFT cells of the master panel 105 that can be part of the scan head 120 .

[0014] The master panel 105 includes pixels arranged in an array corresponding to a flat panel display. As mentioned above, the scan head 120 can be electrically connected to the pixels on the master panel 105 to obtain image data . In one embodiment , the scan head 120 is movable and can be moved substantially parallel to the master panel 105 as indicated by the arrow 122 . This enables the scan head 120 to obtain image data from different parts of the master panel 105, that is, even if the master panel 105 is larger than the scan area of the scan head 120 , it is possible to obtain image data. The movement of the scan head 120 can be performed manually or automated under the control of the controller 110 . Using the scan head 120, the entire master panel 105 or a portion of the master panel can be scanned. ​ Any part (or parts) of the top panel 105 can be scanned. For example , the scanning by the scan head 120 and the unevenness detection by the controller 110 are performed on the selected part and / or the randomly identified part of the master panel 105, so that the presence and nature of the unevenness defect can be sampled without scanning the entire master panel 105.

[0015] The processor 112 of the controller 110 is programmed to execute the defect detection process according to various embodiments such as the method steps described with reference to FIGS. 2A and 2B below. The processor 112 can also be programmed to perform quality assurance of the master panel 105 and / or the flat panel display after detecting a defect regarding the presence and extent of non-uniformity, as described below. The processor 112 can be implemented by one or more computer processors, application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs) or combinations thereof using software, firmware, hardwired logic circuits, or combinations thereof. In one embodiment, the processor 112 can be, for example, a digital signal processor (DSP). The computer processor can be constructed from any combination of hardware architecture, firmware architecture or software architecture in particular, and can include a memory dedicated to the computer processor (e.g., non-volatile memory) that stores executable software / firmware executable code that enables the computer processor to perform various functions. In one embodiment, the computer processor can include, for example, a central processing unit (CPU) that executes the operation of the system.

[0016] The memory 114 stores at least a part of the image data obtained using the scan head 120 and the processing result from the processor 112. The memory 114 is applied by the processor 112 when processing the image data to identify the unevenness defect as described below. ​​​​​It can also be a database that stores various types of templates, structural patterns of specified geometric shapes, and candidate patterns for potential defect detection patterns. Memory 114 can implement, for example, random access memory (RAM) and read-only memory (ROM) of any number, type, and combination, and can also store, for example, computer programs and software algorithms executable by processor 11 , and can also store, for example, computer programs and software algorithms executable by processor 112 (and / or other components), as well as image data and / or test and measurement data streams , and can also store various types of information such as templates and patterns (described above). Various types of ROM and RAM can include any number, type, and combination of computer-readable storage media such as disk drives, disk storage , electrically programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), CD, DVD, universal serial bus (USB) drives, etc. These are tangible and non-transitory storage media (as compared to, for example, temporary propagation signals). and can store various types of information such as templates and patterns (described above). Various types of ROM and RAM can include any number, type, and combination of computer-readable storage media such as disk drives, disk storage , electrically programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), CD, DVD, universal serial bus (USB) drives, etc. These are tangible and non-transitory storage media (as compared to, for example, temporary propagation signals). , electrically programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), CD, DVD, universal serial bus (USB) drives, etc. These are tangible and non-transitory storage media (as compared to, for example, temporary propagation signals). , electrically programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), CD, DVD, universal serial bus (USB) drives, etc. These are tangible and non-transitory storage media (as compared to, for example, temporary propagation signals). , electrically programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), CD, DVD, universal serial bus (USB) drives, etc. These are tangible and non-transitory storage media (as compared to, for example, temporary propagation signals). , electrically programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), CD, DVD, universal serial bus (USB) drives, etc. These are tangible and non-transitory storage media (as compared to, for example, temporary propagation signals). , electrically programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), CD, DVD, universal serial bus (USB) drives, etc. These are tangible and non-transitory storage media (as compared to, for example, temporary propagation signals). , electrically programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), CD, DVD, universal serial bus (USB) drives, etc. These are tangible and non-transitory storage media (as compared to, for example, temporary propagation signals).

[0017] FIG. 2A is a flowchart showing an automated method for detecting non-uniformity defects in a master panel during manufacturing according to a representative embodiment.

[0018] Referring to FIG. 2A, in block S211, a composite image of the master panel (e.g., master panel 105) is obtained, where the master panel can be, for example, a glass substrate (e.g., mother glass substrate (MGS)) including a plurality of flat panel displays, although the master panel can be made of other materials without departing from the scope of the present teachings. ​​​​ The substrate can include a composite (also available). Obtaining a composite image can be done, for example, by scanning a master panel using a scan head (e.g., scan head 120) to obtain a plurality of partial images of the master panel respectively. The scan head can include one or more imaging devices that obtain image data of the master panel when the master panel passes through a defect detection system during the manufacturing process. In one embodiment, the image includes pixel signals received from a matrix of pixels on the master panel in response to the scan. That is, the scan signals from the scan head can be respectively coupled to the pixels on the master panel in a predetermined sequence to provide pixel signals for obtaining a plurality of images. The plurality of images include corresponding image data and at least a portion of it can be stored in a memory (e.g., memory 114). Figure 3 is a screenshot of an exemplary composite image obtained according to a representative embodiment. Referring to Figure 3, the composite image 300 of the master panel can be larger than, for example, 10000 pixels × 10000 pixels. As shown, the composite image 300 includes images of a plurality of flat panel displays that are not yet separated from the master panel during the manufacturing process. The composite image of the master panel has artifacts removed or suppressed and is flattened. Although the images of the flat panel displays are shown by a rectangular 8×6 array, the number of flat panel display images per composite image can be changed without departing from the scope of the present teachings. The substrate can include a composite (also available). Obtaining a composite image can be done, for example, by scanning a master panel using a scan head (e.g., scan head 120) to obtain a plurality of partial images of the master panel respectively. The scan head can include one or more imaging devices that obtain image data of the master panel when the master panel passes through a defect detection system during the manufacturing process. In one embodiment, the image includes pixel signals received from a matrix of pixels on the master panel in response to the scan. That is, the scan signals from the scan head can be respectively coupled to the pixels on the master panel in a predetermined sequence to provide pixel signals for obtaining a plurality of images. The plurality of images include corresponding image data and at least a portion of it can be stored in a memory (e.g., memory 114). Figure 3 is a screenshot of an exemplary composite image obtained according to a representative embodiment. Referring to Figure 3, the composite image 300 of the master panel can be larger than, for example, 10000 pixels × 10000 pixels. As shown, the composite image 300 includes images of a plurality of flat panel displays that are not yet separated from the master panel during the manufacturing process. The composite image of the master panel has artifacts removed or suppressed and is flattened. Although the images of the flat panel displays are shown by a rectangular 8×6 array, the number of flat panel display images per composite image can be changed without departing from the scope of the present teachings. The substrate can include a composite (also available). Obtaining a composite image can be done, for example, by scanning a master panel using a scan head (e.g., scan head 120) to obtain a plurality of partial images of the master panel respectively. The scan head can include one or more imaging devices that obtain image data of the master panel when the master panel passes through a defect detection system during the manufacturing process. In one embodiment, the image includes pixel signals received from a matrix of pixels on the master panel in response to the scan. That is, the scan signals from the scan head can be respectively coupled to the pixels on the master panel in a predetermined sequence to provide pixel signals for obtaining a plurality of images. The plurality of images include corresponding image data and at least a portion of it can be stored in a memory (e.g., memory 114). Figure 3 is a screenshot of an exemplary composite image obtained according to a representative embodiment. Referring to Figure 3, the composite image 300 of the master panel can be larger than, for example, 10000 pixels × 10000 pixels. As shown, the composite image 300 includes images of a plurality of flat panel displays that are not yet separated from the master panel during the manufacturing process. The composite image of the master panel has artifacts removed or suppressed and is flattened. Although the images of the flat panel displays are shown by a rectangular 8×6 array, the number of flat panel display images per composite image can be changed without departing from the scope of the present teachings. The substrate can include a composite (also available). Obtaining a composite image can be done, for example, by scanning a master panel using a scan head (e.g., scan head 120) to obtain a plurality of partial images of the master panel respectively.

[0019] Figure 3 is a screenshot of an exemplary composite image obtained according to a representative embodiment. Referring to Figure 3, the composite image 300 of the master panel can be larger than, for example, 10000 pixels × 10000 pixels. As shown, the composite image 300 includes images of a plurality of flat panel displays that are not yet separated from the master panel during the manufacturing process. The composite image of the master panel has artifacts removed or suppressed and is flattened. Although the images of the flat panel displays are shown by a rectangular 8×6 array, the number of flat panel display images per composite image can be changed without departing from the scope of the present teachings. The substrate can include a composite (also available). Obtaining a composite image can be done, for example, by scanning a master panel using a scan head (e.g., scan head 120) to obtain a plurality of partial images of the master panel respectively. The scan head can include one or more imaging devices that obtain image data of the master panel when the master panel passes through a defect detection system during the manufacturing process. In one embodiment, the image includes pixel signals received from a matrix of pixels on the master panel in response to the scan. That is, the scan signals from the scan head can be respectively coupled to the pixels on the master panel in a predetermined sequence to provide pixel signals for obtaining a plurality of images. The plurality of images include corresponding image data and at least a portion of it can be stored in a memory (e.g., memory 114). Figure 3 is a screenshot of an exemplary composite image obtained according to a representative embodiment. Referring to Figure 3, the composite image 300 of the master panel can be larger than, for example, 10000 pixels × 10000 pixels. As shown, the composite image 300 includes images of a plurality of flat panel displays that are not yet separated from the master panel during the manufacturing process. The composite image of the master panel has artifacts removed or suppressed and is flattened. Although the images of the flat panel displays are shown by a rectangular 8×6 array, the number of flat panel display images per composite image can be changed without departing from the scope of the present teachings. It is possible. For convenience, only a part of the first row and the second row of the composite image is labeled. That is, the first row includes flat panel display images 311 to 318, and the second row includes flat panel display images 321 to 323. The composite image 300 shows examples of non-uniformity defects such as ring-shaped non-uniformity defects 331 and 332 of the ring array partially shown in flat panel display images 312, 313, 316, and 317 (however, the rings of the ring array also appear similarly in other flat panel display images). The ring may be a trace of a suction cup used to fix the master panel during the manufacturing process. Another example of a non-uniformity defect is a spot array including typical spot-shaped non-uniformity defects 341, 342, and 343 corresponding to flat panel display images 321, 322, and 323 respectively (however, although not labeled, further spot-shaped non-uniformity defects of the spot array also appear similarly in other flat panel display images). The spot-shaped non-uniformity defect may be the result of, for example, a handling device used to manipulate and / or move the master panel.

[0020] Providing a composite image by synthesizing image data of a plurality of images (for example, those stored in the memory 114) involves downsampling the plurality of images to reduce the storage requirements for storing the image data, removing measurement artifacts, configuring the downsampled images into a two-dimensional pattern, synthesizing the two-dimensional pattern into a single larger image, suppressing processing artifacts, and a single larger ​​​​​​​​​​​The downsampled image of the image is corrected for contrast and background level and can include providing a composite image of the master panel. Thus, the composite image appears to be homogeneous. Regions of the composite image from the actual data (as regions contrasted with unknown regions) can be marked in a separate binary matrix. The binary matrix regions are used, for example, to weight the filter response resulting from filtering the composite image as described below for block S213.

[0021] By removing and suppressing artifacts, the quality of the composite image is improved by performing a cleaning up process on the image data before and after synthesizing the image data into a single larger image. For example, assuming that the scan head 120 is an electrical contact scanner as discussed above, before synthesizing the two-dimensional pattern into a single larger image, measuring artifacts from the image data in individual frames (flat panel display images) can include removing noise originating from the electronic amplifier components, removing white noise using a low-pass filter, locally flattening the image, and suppressing completely vertical and completely horizontal spatial frequencies. Removing noise originating from the electronic amplifier components can include removing drift by subtracting a gliding average, removing gain variations by normalizing the signal strength, removing cross talk between amplifier channels, and applying a narrowband spatial frequency filter to remove It can include removing measurement artifacts. Removing measurement artifacts can further include, for example, removing lines caused by laser annealing and other manufacturing processes. After synthesizing the two-dimensional patterns into a single larger image, suppressing artifacts can include, for example, at the corners of the stored image data of individual frames, removing signal overshoot and undershoot by, for example, local flattening of the image signal, which is performed before correcting for contrast and background level. For example, it can further include removing lines caused by laser annealing and other manufacturing processes. After synthesizing the two-dimensional patterns into a single larger image, suppressing artifacts can include, for example, at the corners of the stored image data of individual frames, removing signal overshoot and undershoot by, for example, local flattening of the image signal, which is performed before correcting for contrast and background level. This can include removing signal overshoot and undershoot, and is performed before correcting for contrast and background level.

[0022] Figure 4 shows screenshots of exemplary composite images before and after downsampling, cleaning, and composing image data according to an exemplary embodiment. Comparing the mother substrate image 401 with the composite image 402, it is clear that the composite image 402 is flattened. Also, the signal overshoot and undershoot that appear as distinct spots (e.g., representative distinct spots 411 and 421) at some of the corners of the individual flat panel display images within the mother substrate image 401 do not appear in the improved quality composite image 402. In block S212, local non-uniformity defects are detected by filtering the improved quality composite image. Figure 2B is a flowchart showing an automated method of filtering the improved quality composite image to detect local non-uniformity defects as shown by block S212 according to an exemplary embodiment. Comparing the mother substrate image 401 with the composite image 402, it is clear that the composite image 402 is flattened. Also, in some of the corners of the individual flat panel display images within the mother substrate image 401, the signal overshoot and undershoot that appear as distinct spots (e.g., representative distinct spots 411 and 421) do not appear in the improved quality composite image 402.

[0023] In block S212, local non-uniformity defects are detected by filtering the improved quality composite image. Figure 2B is a flowchart showing an automated method of filtering the improved quality composite image to detect local non-uniformity defects as shown by block S212 according to an exemplary embodiment. As shown by block S212 according to an exemplary embodiment, to detect local non-uniformity defects, an automated method of filtering the improved quality composite image is shown in the flowchart. Various modifications of the method can be incorporated without departing from the scope of the present teachings. .

[0024] The detected local non-uniformity defect includes at least one structural pattern of the corresponding defined geometry. For example, the structural pattern of the defined geometry includes ring-shaped non-uniformity defects 331 and 332 and representative spot-shaped non-uniformity defects 341, 342, and 343, etc. in the improved quality composite image shown in FIG. 3, and can include ring and / or spot structural patterns on the improved quality composite image. Therefore, filtering the improved quality composite image to detect local non-uniformity defects can include using a defect pattern specific template for detecting the structural pattern of the defined geometry. Generally, filtering the improved quality composite image can include extracting a relevant spatial frequency corresponding to the length scale (e.g., characteristic length) of the local non-uniformity defect to provide a first filtered image, and then superimposing this first filtered image with a set of defect pattern specific templates corresponding to the defined geometries (e.g., ring-shaped non-uniformity defects and spot-shaped non-uniformity defects). FIG. 2B shows an exemplary process shown by block S212 of FIG. 2A for filtering the improved quality composite image. Referring to FIG. 2B, according to one embodiment, the improved quality composite image from block S211 is filtered using a kernel that selects (or extracts) relevant spatial frequencies in block S221, compiling a histogram of the kernel-filtered composite image, and this filtered image can include the following steps: superimposing the first filtered image with a set of defect pattern specific templates corresponding to the defined geometries (e.g., ring-shaped non-uniformity defects and spot-shaped non-uniformity defects).

[0025] FIG. 2B shows an exemplary process shown by block S212 of FIG. 2A for filtering the improved quality composite image. Referring to FIG. 2B, according to one embodiment, the improved quality composite image from block S211 is filtered using a kernel that selects (or extracts) relevant spatial frequencies in block S221, thereby compiling a histogram of the kernel-filtered composite image, and this filtered image can include the following steps: ​The synthesized image that has been filtered is, in block S222, the pixel signal of this filtered synthesized image is renormalized in pixel signal units as a probability that is not random noise by comparing it at least partially with a histogram. The histogram can include an empirical histogram fitted to a Gaussian distribution, where the difference between the empirical histogram and the Gaussian fitting is large for high-intensity signals. The probability that a pixel is background (random noise) is calculated by comparing the corresponding pixel signal with both the empirical histogram and the Gaussian fitting. The kernel for extracting relevant spatial frequencies can include a simple Gaussian function having a width based on a visual inspection of a predetermined characteristic length scale. For example, if time is a limiting factor, the selected relevant spatial frequencies can be downsampled to reduce processing time. Unrelated spatial frequencies (e.g., overly high frequencies, overly low frequencies) can be discarded. The pixel signal is renormalized in pixel signal units as a probability that is not random noise by comparing it at least partially with a histogram. The histogram can include an empirical histogram fitted to a Gaussian distribution, where the difference between the empirical histogram and the Gaussian fitting is large for high-intensity signals. The probability that a pixel is background (random noise) is calculated by comparing the corresponding pixel signal with both the empirical histogram and the Gaussian fitting. The difference between the empirical histogram and the Gaussian fitting is large for high-intensity signals. The probability that a pixel is background (random noise) is calculated by comparing the corresponding pixel signal with both the empirical histogram and the Gaussian fitting. The kernel for extracting relevant spatial frequencies can include a simple Gaussian function having a width based on a visual inspection of a predetermined characteristic length scale. The kernel for extracting relevant spatial frequencies can include a simple Gaussian function having a width based on a visual inspection of a predetermined characteristic length scale. For example, if time is a limiting factor, the selected relevant spatial frequencies can be downsampled to reduce processing time. Unrelated spatial frequencies (e.g., overly high frequencies, overly low frequencies) can be discarded. Unrelated spatial frequencies (e.g., overly high frequencies, overly low frequencies) can be discarded.

[0026] FIG. 5 shows screenshots of exemplary synthesized images before and after filtering using a kernel according to a representative embodiment. Comparing the synthesized image 501 of improved quality with the kernel-filtered synthesized image 502, it is clear that the frequency characteristics of the signal of interest are emphasized. This is shown by the improved contrast of the mottle defect and similar signals against the background of the kernel-filtered synthesized image 502. Also, high spatial frequencies are discarded by filtering using the kernel as described above, which is shown by the scales of the vertical and horizontal axes of the synthesized images 501 and 502. Comparing the synthesized image 501 of improved quality with the kernel-filtered synthesized image 502, it is clear that the frequency characteristics of the signal of interest are emphasized. Comparing the synthesized image 501 of improved quality with the kernel-filtered synthesized image 502, it is clear that the frequency characteristics of the signal of interest are emphasized. This is shown by the improved contrast of the mottle defect and similar signals against the background of the kernel-filtered synthesized image 502. This is shown by the improved contrast of the mottle defect and similar signals against the background of the kernel-filtered synthesized image 502. Also, high spatial frequencies are discarded by filtering using the kernel as described above. Also, high spatial frequencies are discarded by filtering using the kernel as described above, which is shown by the scales of the vertical and horizontal axes of the synthesized images 501 and 502. is shown. That is, these vertical and horizontal axes indicate that the image resolution is decreasing and are relatively short. Since high frequencies are removed, the remaining low frequencies are larger and can thus be evenly and well represented by fewer pixels. For efficiency reasons, the image is downscaled in this way. The composite image 501 can correspond to the flattened composite image 402 in FIG. 4 discussed above.

[0027] FIG. 6 shows enlarged portions of screenshots of exemplary composite images in FIG. 5 before and after filtering with a kernel according to an exemplary embodiment. Each of the enlarged portions 501' and 502' corresponds to the location of the ring-shaped non-uniformity 530 of the composite image 501 of improved quality and the kernel-filtered composite image 502, respectively. The ring-shaped non-uniformity 530 can correspond to one of the ring-shaped non-uniformities 331 or 332 in FIG. 3 for the sake of explanation. A Gaussian filter 505 showing the kernel filter ring executed on the composite image 501 is also shown. The structure of interest in the illustrated example has a "line width" of about 10 pixels to 20 pixels. Thus, although the filter width of the Gaussian filter 505 is selected to have a similar size (e.g., 20 pixels × 20 pixels) for superposition, other similar filter widths can be incorporated without departing from the scope of the present teachings. Comparing the enlarged portion 501' of the composite image 501 of improved quality with the enlarged portion 502' of the kernel-filtered composite image 502, it can be seen that signals from neighboring pixels are being combined and averaged. This Thus, individual pixels are prevented from having an impact on subsequent steps in the non-uniformity defect detection process that is above the necessary level. It is assumed that there is a significant random background response from feature detection and that the area where the real signal exists is small. The absolute intensity of the signal is less important than whether the signal is genuine, as opposed to a signal that is, for example, derived from a very large background variation.

[0028] FIG. 7 shows screenshots of exemplary synthetic images before and after re-normalizing the kernel-filtered synthetic image in block S222 according to an exemplary embodiment. The kernel-filtered synthetic image 701 shows a significant amount of variation in the signal, as indicated by spots throughout the synthetic image. As shown, the signal intensity of the background variation approximates the real signal intensity, making it more difficult to detect the real signal. On the other hand, in contrast, the re-normalized synthetic image 702 removes the background signal and sets the real signal to a uniform signal intensity. This is clearly contrasted with the substantially uniform and homogeneous background of the re-normalized synthetic image 702 by potential non-uniformity defects (bright areas). In particular, the kernel-filtered synthetic image 701 in FIG. 7 can correspond to the kernel-filtered synthetic image 502 in FIG. 5 discussed above. Also, the re-normalized synthetic image 702 can be the first filtered image described above.

[0029] Next, in block S223, the synthetic image of improved quality is filtered to localize Detecting unevenness defects involves superimposing the re-normalized composite image (first filtered image) with a set of defect pattern-specific templates corresponding to respective specified geometric shapes (e.g., rings and spots) to identify defect patterns in the re-normalized filtered composite image. The defect patterns do not have to exactly match the shape and / or size of the templates and multiple iterations can be performed if necessary. In block S224, the result of superimposing the re-normalized composite image with the set of defect pattern-specific templates is weighted by the number of pixel signals that are actual data, as opposed to pixel signals of 0 corresponding to pixels in areas where image data was not acquired. In block S225, the weighted result is smoothed using a kernel to provide a filtered composite image. Smoothing includes penalizing isolated signals and forcing signals to match a predetermined characteristic length scale, e.g., penalizing longer and shorter characteristic lengths.

[0030] FIG. 8 shows a screenshot of an exemplary re-normalized composite image superimposed with a defect pattern-specific template according to a representative embodiment. Referring to FIG. 8, the superimposed image 801 shows a superimposition using a template corresponding to a ring-shaped unevenness defect, such as the exemplary ring template 811. The superimposed image 802 shows a superimposition using a template corresponding to a spot-shaped unevenness defect, such as the exemplary spot template 812. The ring template 811 and the spot template 812 are, for example, software It can be implemented as an algorithm or an applet. In the illustrated embodiment, for example, The ring-shaped non-uniformity defect is significantly larger than the spot-shaped non-uniformity defect. For example, the ring template - 811 can be 200 pixels × 200 pixels, and the spot template - can only be 20 pixels × 20 pixels, but other sizes and relative sizes can be incorporated without departing from the scope of the present disclosure. Of course, for the ring - shaped non-uniformity defect and / or the size and aspect ratio of the spot-shaped non-uniformity defect, if there is uncertainty present, multiple ring templates and / or spot templates can be used. Also, various embodiments are not limited to ring-shaped non-uniformity defects and spot-shaped non-uniformity defects. That is, if necessary, templates for any various shapes that are expected to appear as non-uniformity defects in the re-normalized composite image can be created.

[0031] The superimposition executed in block S223 detects the cross-correlation between each of the ring template 811 and the potential non-uniformity defect, and the cross-correlation between each of the spot template 812 and the potential non-uniformity defect in the re-normalized composite image 702 of FIG. 7. When the cross-correlation shows sufficient similarity in shape, the initial pattern of the non-uniformity defect is detected. As described above, the result of the superimposition is weighted by the number of pixel signals that are actual data, and this weighted result is smoothed using a kernel to provide a filtered composite image including the detected patterns of the ring-shaped non-uniformity defect and the spot-shaped non-uniformity defect. Effectively smoothing penalizes isolated signals (e.g., spurious responses) Impose a penalty on "i" and force a signal that matches the associated length scale.

[0032] Figure 9 shows an exemplary smoothed filtered composite image screen shot showing a non-uniformity defect template according to a representative embodiment. In particular, composite image 90 1 shows a template corresponding to a ring-shaped non-uniformity defect, and composite image 902 shows a spot -shaped non-uniformity defect template. As discussed above with respect to block S225 in FIG. 2B, smoothing includes imposing a penalty on isolated signals and forcing (or enhancing) a signal that matches a predetermined characteristic length scale such as the exemplary characteristic length scale 915 shown in FIG. 9. The characteristic length scale 915 is, for example, slightly smaller than the spot template 812, whereby spot-shaped non-uniformity defects are enhanced along with ring-shaped non-uniformity defects, while signals having different characteristic lengths are not (at least to the same extent) enhanced.

[0033] Referring back to FIG. 2A, at block S213, different candidate patterns are created and applied to the filtered composite image. The candidate patterns each have a two-dimensional coordinate system sized differently to attempt to indicate the positioning of the detected local non-uniformity defect. Each of the candidate patterns can include the horizontal and vertical periodicity of the detected local non-uniformity defect, as well as the horizontal and vertical positioning, thereby creating a rectangular pattern. Of course, patterns having shapes other than rectangular can be incorporated without departing from the scope of the present teachings, depending, for example, on the shape of the substrate and / or the shape of the scan head. It is possible. For example, the candidate pattern can be circular or hexagonal. The horizontal and vertical positioning of each can be changed, for example, by a search algorithm, and this may be referred to as a scan offset. That is, the scan offset corresponds to the positioning of the candidate pattern with respect to the position of each display on the glass panel. Also, at least one of the candidate patterns can be rotated by introducing a rotation angle parameter in order to fit well with the orientation of one or more of the detected local non-uniformities among the detected local non-uniformities.

[0034] The user can estimate a reasonable range of horizontal and vertical periodicities to create candidate patterns. Alternatively, as discussed above, the cross-correlation between the template and the potential non-uniformity can be analyzed in terms of the maximum value applied to obtain a reasonable horizontal and vertical periodicity. Also, an exhaustive scan of all possible horizontal and vertical periodicity values can be performed, and the results can be used to constrain the range of reasonable horizontal and vertical periodicities even in multiple subsequent detection trials.

[0035] The candidate pattern is scanned across a data matrix (array) corresponding to the filtered composite image showing the non-uniformity template. The local data at the intersection of the candidate pattern is excised, and the signal intensity at the pixel corresponding to the excised local data is measured. To allow for some error margin, a finite-size region corresponding to the allowable error range margin is excised around the intersection of the candidate pattern, and the composite signal within that region is ​​​​​​​The maximum value (as opposed to the average) is measured. For a data matrix having many rows and / or columns In the case of, there is no chance to contribute to the signal because it accidentally exists at a position where data is not stored , the points within the data matrix should not be penalized. On the other hand, only the 2×2 positions If it exists within the data matrix, all positions must be at least partially visible , otherwise there is no defined solution to the location problem.

[0036] FIG. 10 shows a screen shot of an exemplary filtered composite image showing a ring-shaped non-uniformity defect template together with a plurality of different candidate patterns according to an exemplary embodiment . In particular, a first candidate pattern 1030, a second candidate pattern 1040 and a third candidate pattern 1050 of different sizes and dimensions are shown on a composite image 901 having a template corresponding to the ring-shaped non-uniformity defect. (Candidate patterns of different sizes and dimensions are similarly applicable to the composite image 902 in FIG. 9 having a template corresponding to the spot-shaped non-uniformity defect, and thus this discussion is equally applicable to the composite image 902 is understood. ) Also, double-headed arrows 1041 and 1042 indicate the horizontal and vertical scan offsets of the second candidate pattern 104 0, and double-headed arrows 1051 and 1052 indicate the horizontal and vertical scan offsets of the third candidate pattern 1050. The first candidate pattern 1030 will also have horizontal and vertical scan offsets, but is not explicitly shown in FIG. 10 . The first candidate pattern 1030, the second candidate pattern 1040 and the third candidate pattern 1050 are arranged in the horizontal and vertical directions (i.e., ), ), and the third candidate pattern 1050 are arranged in the horizontal and vertical directions (i.e., Although not rotated, one or more of the first candidate pattern 1030, the second candidate pattern 1040, and the third candidate pattern 1050 can be rotated so as to fit well into the pattern of the template corresponding to the ring-shaped unevenness defect. At the corners of the first candidate pattern 10 30, the second candidate pattern 1040, and the third candidate pattern 1050 signals are detected, and these signals are part of the total filter response from the composite image 901. As discussed above, some finite-sized regions corresponding to the error tolerance margin are cut out around the intersections (e.g., corners) of the first candidate pattern 1030, the second candidate pattern 1040, and the third candidate pattern 10 50, and within the region of the remaining portions of the first candidate pattern 1030, the second candidate pattern 1040, and the third candidate pattern 1050, ( in contrast to the average composite signal) the maximum composite signal is measured. This signal is quantized to pixel values .

[0037] In one embodiment, the signals detected at the corners of the first candidate pattern 1030, the second candidate pattern 1040, and the third candidate pattern 1050 are the maximum signals obtained within the tolerance region. For example, FIG. 11 shows an exemplary filtered composite image with a candidate pattern having a tolerance region and a ring-shaped unevenness defect template according to a representative embodiment. Referring to FIG. 11, although the tolerance region is shown for convenience on the corner of the first candidate pattern 1030, it is understood that other candidates turns (e.g., the second candidate pattern 1040 and the third candidate pattern 1050) can also include a tolerance region. In particular, although the tolerance region is shown for convenience on the corner of the first candidate pattern 1030, it is understood that other candidates turns (e.g., the second candidate pattern 1040 and the third candidate pattern 1050) can also include a tolerance region. In particular, the four Each of the corners has tolerance range regions 1031, 1032, 1033 and 1034. The tolerance range regions 1031, 1032, 1033 and 1034 are not to scale. This is because the actual tolerance range reflects horizontal and / or vertical periodicity uncertainties and is thus more likely to be smaller in scaled depictions. In one embodiment, each of the tolerance range regions 1031, 1032, 1033 and 1034 is centered at the corresponding corner (intersection) of the first candidate pattern 1030. The tolerance range regions 1031, 1032, 1033 and 1034 allow for fitting of a slightly rotated data matrix without the need to rotate the candidate pattern 1030. This speeds up the process of creating and selecting candidate patterns and ultimately helps

[0038] In block S214, one of the candidate patterns is selected as the defect detection pattern, where the defect detection pattern is the candidate pattern that most closely resembles the defined geometric shape structure of the detected local non-uniformity defect. The defect detection pattern can be automatically selected by determining which candidate turn provides the highest intensity signal of the total filter response from the filtered synthetic image. For example, selecting the defect detection pattern can include summing the filter responses at regularly spaced lattice points for each of the candidate patterns. In that case, the candidate pattern Capture the positioning of unevenness defects.

[0039] The selected defect detection pattern can be fine-tuned by measuring the median signal at the horizontal and vertical (X, Y ) coordinates for each row and column of pixels in the filtered composite image, for example, to provide robustness against outliers. Optionally , rotation can be determined to provide the positioning of the selected defect detection pattern. For example, the rotation can be inferred by linear fitting in polar coordinates. This too is optional, and linear fitting can be used to infer the bunch periodicity so that rows and columns that are not visible in the filter composite image can be inferred. Additionally, a scan offset can be selected in block S214, where the selected scan offset most closely corresponds to the positioning of the detected local unevenness defect.

[0040] FIG. 12 shows an exemplary screenshot of a filtered composite image showing the selected defect detection pattern according to an exemplary embodiment. In particular, composite image 1201 shows the selected defect detection pattern for spot-like unevenness defects, and composite image 1202 shows the selected defect detection pattern for ring-like unevenness defects. In composite image 1201, the hash symbols indicate the selected defect detection pattern for the unevenness spots, and the circles indicate the detection of spot-like unevenness defects. Of course, these detections include some margin of error, so these detections are assumed to be arranged in a regular pattern, i.e., the hash symbols are at the actual positions. FIG. 1 As shown in 2, the hash symbol substantially aligns with the actual spot-like unevenness defect. Synthetic In the composite image 1202, there are four sets of hash symbols, and each of these sets shows a selected defect detection pattern for the corresponding ring-like unevenness defect. Each part of the ring-like unevenness defect is visible in the composite image 1 202 surrounded by the selected defect detection pattern.

[0041] In block S215, the detected local unevenness defect is quantified using the (fine-tuned) defect detection pattern. Parameters such as diameter and eccentricity, as well as statistical parameters such as mean, median, maximum and minimum signal intensities, as well as noise level, and other parameters can be extracted from the area marked as an unevenness defect.

[0042] In block S216, at least a part of the defect detection pattern is displayed on a display (e.g., display 130) together with the filtered composite image. The displayed part of the defect detection pattern indicates the position of the detected local unevenness defect in a structural pattern of a prescribed geometric shape. The detected local unevenness defect can be useful for a plurality of purposes. For example, the master panel can accept or reject based on the degree to which a local unevenness defect is detected. Also, using the detected local unevenness defect, the manufacturing apparatus that causes the unevenness defect can be identified, whereby the manufacturing apparatus can be adjusted or tuned or replaced.

[0043] According to various embodiments, non-uniformity defects on a master panel and / or a flat panel display included in the master panel can be automatically detected, classified, and / or quantified. As a result of the automatic non-uniformity defect detection process according to various embodiments, throughput is increased, labor costs are reduced, and the results are standardized as compared to manual inspection. This process also provides flexibility, enabling the user to add target detection patterns to be detected as needed. Furthermore, by localizing and quantifying the detection patterns, it becomes possible to perform a stepwise scoring of the panel quality, which is superior to a simple pass / fail evaluation. The stepwise scoring of the panel quality can be incorporated, for example, into a quality assurance determination that can be used as a criterion for accepting and rejecting panels based in part on the presence and degree of non-uniformity defects. Also, based on the detected pattern type and localization, it becomes possible to determine which processing apparatus in the manufacturing line has caused a particular non-uniformity defect. Therefore, as described above, improvements can be brought to the production line by adjusting, regulating, or replacing the apparatus that has caused the non-uniformity defect, thereby increasing the yield. Factors that can be considered in the stepwise scoring of panel quality can include, respectively, the number, size, depth / intensity, position, and / or shape of non-uniformity defects. For example, the size can be determined by the measurement area, diameter, and / or length of the non-uniformity defect, and the depth / intensity can be determined by the sharpness measured. Generally, a panel having a larger number of larger and more intense defects is scored lower, and vice versa. The position of the non-uniformity defect can be determined based on where it appears on the master panel and / or the flat panel display, and whether the non-uniformity defects are periodically arranged or isolated. The shape includes, for example, circular, linear, and dot. The effects of position and size on panel quality can be varied to suit any particular situation or the specific design requirements of various embodiments, as will be apparent to those skilled in the art. The position and shape can each also indicate the cause of the non-uniformity defect, identifying the equipment on the manufacturing line that caused the non-uniformity defect, how the equipment can be adjusted or tuned, and ultimately whether the equipment needs to be replaced. As described above, the factors are automatically determined and scored by the processor 112, or are manually displayed and / or interfaced to the display 130 by the user.

[0044] Although the present invention has been described and illustrated in detail in the drawings and the foregoing description, such description and illustration are to be considered illustrative or exemplary and not restrictive. The present invention is not limited to the disclosed embodiments. Those skilled in the art will, upon consideration of the drawings, disclosure, and the scope of the appended claims, understand and be able to make other variations of the disclosed embodiments. In the claims, the word "comprising" does not exclude other elements or steps, and the indefinite article "a" or "an" means that there are plural instances.

[0045] Those skilled in the art, when implementing the invention according to the claims, will understand and be able to make other variations of the disclosed embodiments by considering the drawings, the disclosure, and the scope of the appended claims. In the claims, the word "comprising" does not exclude other elements or steps, and the indefinite article "a" or "an" means that there are plural instances. This is not excluded. If a certain means is listed in a plurality of dependent claims that are different from each other, it is not shown that these means cannot be used advantageously in combination. Let it be assumed that.

[0046] Representative embodiments have been disclosed herein, but many variations are possible according to the present teachings, and those skilled in the art will recognize that they remain within the scope of the appended claims. Accordingly, the present invention is not limited except as defined by the scope of the appended claims. Note that the description of the claims in the original application at the time of filing was as follows. Claim 1: A method for detecting non-uniformity defects in a master panel (105) during manufacturing, wherein the master panel (105) includes a plurality of flat screen displays, the method comprising: preparing a composite image (1201) from the image data of the master panel (105); improving the quality of the composite image (1201), including removing artifacts from the composite image (1201). Filtering the synthetic image (402) of the improved quality to detect local non-uniformity defects, wherein the local non-uniformity defects include at least one structural pattern of a specified geometric shape, Applying a plurality of different candidate patterns to the filtered synthetic image (1201), Selecting one of the plurality of candidate patterns as a defect detection pattern, wherein the defect detection pattern closest to the structural pattern of the specified geometric shape of the detected local non-uniformity defect is selected from among the plurality of candidate patterns, Displaying at least a portion of the defect detection pattern on a display (130) together with the synthetic image (1201) of the improved quality to indicate the position of the detected local non-uniformity defect in the structural pattern of the specified geometric shape, A method comprising. Claim 2: Filtering the synthetic image (402) of the improved quality to detect local non-uniformity defects, Filtering out a relevant spatial frequency corresponding to the length scale of the detected local non-uniformity defect to provide a first filtered image, Overlaying the first filtered image with a set of templates corresponding to the specified geometric shapes respectively, The method according to claim 1, comprising. Claim 3: The method according to claim 1, wherein the specified geometric shape includes spots and / or a plurality of rings. Claim 4: A method for detecting non-uniformity defects in a flat screen display before separation from a master panel (105) during a manufacturing process, Obtaining a plurality of images of the master panel (105) and storing image data corresponding to at least some of the plurality of images, wherein the image data includes pixel signals received from a matrix of pixels on the master panel (105), Synthesizing the stored image data of the plurality of images to provide a synthetic image (1201), Filtering the synthetic image (1201) using at least one defect pattern specific template to detect a structural pattern of a specified geometric shape corresponding to a non-uniformity defect to detect the non-uniformity defect, Creating candidate patterns having two-dimensional coordinate systems dimensioned differently from each other to attempt to indicate the positioning of the detected local non-uniformity defect, Selecting a defective detection pattern from among the created candidate patterns and a scan offset that most closely corresponds to the positioning of the detected non-uniformity defect, wherein the selected defective detection pattern provides the signal with the highest intensity among the total filter (505) responses from the filtered composite image (1201), Fine-tuning the selected defective detection pattern, Quantifying the detected non-uniformity defect using the fine-tuned defective detection pattern, A method comprising. Claim 5: ​ The method according to claim 4, wherein each of the candidate patterns includes the horizontal and vertical periodicities of the detected local non-uniformity defect and horizontal and vertical scan offsets, whereby a rectangular pattern is created. Claim 6: The method according to claim 5, wherein at least one of the candidate patterns is rotated by introducing a rotation angle parameter. Claim 7: Obtaining the plurality of images of the master panel (105) includes scanning the master panel (105) using a scan head (120), and electrical or optical signals from the scan head (120) are respectively coupled to the pixels on the master panel (105) in a predetermined sequence to provide the pixel signals. The method according to claim 4. Claim 8: The method according to claim 4, wherein the master panel (105) includes a glass substrate. Claim 9: Providing the composite image (1201) by synthesizing the stored image data of the plurality of images includes Downsampling the plurality of images to reduce storage requirements for storing the image data, Configuring the downsampled plurality of images into a two-dimensional pattern and synthesizing the two-dimensional pattern into a single larger image, Correcting the downsampled plurality of images of the single larger image for contrast and background level to provide the composite image (1201) that appears to be homogeneous, Marking in a separate binary matrix a region (1201) of the composite image consisting of real data as a region contrasting with an unknown region, wherein the binary matrix region is used to weight the filter (505) response resulting from filtering the composite image (1201), The method according to claim 4, comprising. Claim 10: Further comprising cleaning up the composite image (1201) before filtering the composite image (1201) to detect the unevenness defect, and cleaning up the composite image (1201) comprises removing measurement artifacts from the stored image data before synthesizing the two-dimensional pattern into the single larger image, and suppressing artifacts in individual frames of the single larger image before correcting for contrast and background level, The method according to claim 9, comprising.

Description of Symbols

[0047] 100 Defect Detection System 105 Master Panel 110 Controller 112 Processor 114 Memory 116 Interface (I / F) 120 Scan Head 130 Display

Claims

1. A method for detecting non-uniformity defects in a flat screen display before separation from a master panel (105) during a manufacturing process, comprising: obtaining a plurality of images of the master panel (105) and storing image data corresponding to at least some of the plurality of images, the image data including pixel signals received from a matrix of pixels on the master panel (105); combining the stored image data of the plurality of images to provide a combined image; filtering the combined image using at least one defect pattern specific template to detect a structural pattern of a defined geometry corresponding to the non-uniformity defect and detecting the non-uniformity defect; creating candidate patterns each having a two-dimensional coordinate system dimensioned differently to attempt to indicate the positioning of the detected local non-uniformity defect; selecting a defect detection pattern from among the created candidate patterns and a scan offset that most closely corresponds to the positioning of the detected non-uniformity defect, the selected defect detection pattern providing the signal of the highest intensity among the total filter (505) responses from the filtered combined image; fine-tuning the selected defect detection pattern; quantifying the detected non-uniformity defect using the fine-tuned defect detection pattern; performing quality assurance on the master panel (105) based on the quantified detected local non-uniformity defect; filtering the combined image with a kernel that selects a relevant spatial frequency; compiling a histogram of the filtered combined image and re-normalizing the pixel signals of the filtered combined image in pixel signal units as a probability of not being noise by comparing the pixel signals of the filtered combined image with the histogram at least partially; A method including the above steps.

2. After filtering using the at least one defect pattern specific template, superimposing the re-normalized combined image with a set of templates corresponding to the defined geometry; weighting the superimposed result by the number of pixel signals that are real data as opposed to pixel signals of 0 corresponding to pixels in regions where no image data was acquired; Smoothing the weighted result using a kernel to provide a filtered composite image, wherein the smoothing includes imposing a penalty on isolated signals and forcing signals that match a predetermined characteristic length scale (915), and providing a filtered composite image The method according to claim 1, comprising.

3. The method according to claim 1, wherein the kernel for extracting the associated spatial frequency includes a simple Gaussian function having a width based on a visual inspection of the predetermined characteristic length scale (915).

4. The method according to claim 1, wherein the selected associated spatial frequency is downsampled to reduce processing time.

5. The histogram includes an empirical histogram fitted to a Gaussian distribution, and the difference between the histogram and the Gaussian distribution is large for high-intensity signals. The method according to claim 1, wherein the probability that a pixel is a background is calculated by comparing the corresponding pixel signal with both the histogram and the Gaussian distribution.

6. Creating the candidate pattern includes Determining a range of a two-dimensional coordinate system sized to be plausibly different, Scanning the two-dimensional coordinate system across the smoothed filtered composite image, removing local pixel signals at each intersection of the two-dimensional coordinate system, and measuring the composite signal intensity in the removed local pixel signals within an allowable range region at each intersection of each of the candidate patterns. The method according to claim 5, comprising.

7. Selecting the defect detection pattern from the created candidate patterns includes Summing the filter (505) responses at regularly spaced lattice points for each of the candidate patterns, Selecting the candidate pattern having the highest summed filter (505) response as the best candidate pattern. Including Each of the lattice points of the candidate pattern marks the location of a high-intensity filter (505) signal, and thus captures the positioning of the detected non-uniformity defect. The method according to claim 6.

8. The method according to claim 1, wherein each of the candidate patterns includes the horizontal and vertical periodicities of the detected local non-uniformity defect and horizontal and vertical scan offsets, whereby a rectangular pattern is created.

9. The method according to claim 1, wherein at least one of the candidate patterns is rotated by the introduction of a rotation angle parameter.

10. Obtaining the plurality of images of the master panel (105) includes scanning the master panel (105) using a scan head (120), and the electrical or optical signals from the scan head (120) are respectively coupled to the pixels on the master panel (105) in a predetermined sequence to provide the pixel signals. The method according to claim 1.

Images