JP6208426B2 - Automatic unevenness detection apparatus and automatic unevenness detection method for flat panel display - Google Patents

Description

  The present invention relates to a flat panel display automatic unevenness detection apparatus and automatic unevenness for automatically detecting luminance and color non-uniformity (unevenness) generated in a flat panel display manufacturing process by performing arithmetic processing based on image data. It relates to a detection method.

  In recent years, various types of flat panel displays have been developed and commercialized. One of the defects that degrade the image quality of such flat panel displays is luminance and color non-uniformity (unevenness).

  Conventionally, inspection of unevenness at a production site has been visually performed by an inspector. However, since much time, inspection personnel, and cost are required, automation of inspection is desired, and some automation is progressing.

  As an example of such an automation method, a technique for photographing an emitted flat panel with a camera and detecting unevenness by image processing has been developed and is used in an inspection process (for example, non-patented). Reference 1).

Tetsuya Otani et al., “Image quality inspection algorithm for flat panel displays”, Yokogawa technique, vol. 47 No. 3 (2003)

However, the prior art has the following problems.
Such automatic inspection is still insufficient in performance and cannot cope with various defects other than specific point defects and line defects.

The general flow of defect inspection by conventional image processing is roughly divided into the following three parts.
(1) Pre-processing unit (noise removal by filters, pixel structure removal, background image prediction removal, etc.)
(2) Defect candidate extraction section (candidates are extracted by calculating feature values related to defects, binarizing, and labeling)
(3) Judgment part (determining and displaying defective parts with judgment rules based on experience)

  In each image processing part, a plurality of different processes are adopted depending on the type of defect to be detected and the shooting environment, etc., but in many cases, the process is generally complicated, and detection performance can be obtained stably. I could not say.

  Here, the problem which arises in a pre-processing part is demonstrated concretely. Flat panel display defects often appear as subtle changes in brightness (brightness). However, since the flat panel display is a panel having a fixed pixel structure, the structure of the pixel unit is reflected in an image captured by the camera as light and dark, and as it is, it may hinder detection of a defect of the flat panel display.

  In order to avoid the problem, it is necessary to make the pixel structure invisible by applying a filter that passes a predetermined band, and to extract only the light / dark information displayed in each pixel of the display. However, by applying the filter, the characteristic of the defect that should be detected is blurred, and a new problem that subsequent defect detection processing becomes difficult occurs.

  Further, it is necessary to finely adjust the characteristics of the band-pass filter and the subsequent defect detection processing algorithm depending on the type of defect to be detected and the type of panel, and this causes a problem that the number of work processes increases.

  Further, the background image is predicted by a low-pass filter or the like having a large filter size and subtracted from the input image, thereby emphasizing the uneven portion, separating it from noise, and detecting the unevenness by binarizing with a certain threshold value. It is possible. However, in such a detection, unless the filter size is increased to realize a sufficient low-pass effect, the unevenness detection performance is affected. For this reason, the burden of hardware increases and the problem that the processing time of software becomes long also arises, and the influence on the cost increase of the flat panel display which is a final product arises.

  The present invention has been made to solve the above-described problems, and in the inspection process of a flat panel display, labor saving of the inspection process can be achieved when automatically detecting a non-uniformity (unevenness) region by image processing. An object of the present invention is to obtain a flat panel display automatic unevenness detecting apparatus and automatic unevenness detecting method capable of increasing the speed and reducing the cost of a flat panel display as a final product.

An automatic unevenness detection apparatus for a flat panel display according to the present invention captures an image of a light emission state of a flat panel display captured by a camera as an input image, and performs preprocessing on the input image. By performing image processing by a defect candidate extraction unit that performs extraction processing of the unevenness candidate region on the output image and a determination unit that performs authenticity determination of the unevenness candidate region extracted by the defect candidate extraction unit, An automatic unevenness detection device for a flat panel display that determines the presence or absence of a defective uneven region, wherein the preprocessing unit is captured by the camera without removing the influence of the pixel structure of the flat panel display from the input image Input using a Gaussian filter configured as a size to remove camera noise in the image Camera noise is removed from the image, and outputs the defect candidate extracting unit in the subsequent stage, the defect candidate extracting unit, a process of calculating the maximum and minimum values for the output image of the pre-processing unit in the target blocks of a predetermined size This is executed while moving the block of interest for each pixel, and a local median value = (maximum value + minimum value) / 2 is calculated for each pixel, and the output image and median value of the preprocessing unit are calculated in pixel units. Comparing and binarizing adaptively using the median value calculated locally for each pixel so that the output pixel is 1 (white) when the output pixel is larger than the median, and 0 (black) otherwise. By doing so, an adaptive binarized image that becomes a non-uniformity candidate region is generated, and black / white reversal processing is executed on the adaptive binarized image after expansion / contraction processing, or contraction / expansion processing is executed after black / white reversal processing After removing isolated points and connecting regions, By extracting the frequency, labeling the extracted closed regions, further refine unevenness candidate region, and outputs to the subsequent stage.

Also, the flat panel display automatic unevenness detection method according to the present invention according to the present invention is a preprocessing that takes an image of a light emission state of a flat panel display captured by a camera as an input image and performs preprocessing on the input image. Image processing by the defect candidate extraction unit that performs the extraction process of the unevenness candidate region on the output image of the image processing unit and the preprocessing unit, and the determination unit that determines the authenticity of the unevenness candidate region extracted by the defect candidate extraction unit Thus, an automatic unevenness detection method for a flat panel display that determines the presence or absence of uneven areas with spot-like defects, without removing the influence of the pixel structure of the flat panel display from the input image in the preprocessing unit Gauss configured as a size to remove camera noise contained in images taken by the camera Using down filter camera noise removal from the input image, and outputting the defect candidate extracting unit in the subsequent stage, the defect candidate extracting unit, the maximum value for the output image of the pre-processing unit in the target blocks of a predetermined size And a process of calculating the minimum value while moving the block of interest for each pixel, calculating a local median value = (maximum value + minimum value) / 2 for each pixel, The center calculated locally for each pixel so that the output image and the median are compared in units of pixels, and the output pixel is 1 (white) when the output pixel is larger than the median, and 0 (black) otherwise. Adaptively binarizing using the values to generate an adaptive binarized image that becomes a non-uniformity candidate region, and executing black and white inversion processing on the adaptive binarized image after expansion / contraction processing; Or shrink / After performing isolated point removal and region linking by executing a stretching process, a step of extracting a white closed region, and a step of further narrowing down the non-uniformity candidate region by labeling the extracted closed region and outputting to the subsequent stage Are provided.

  According to the present invention, the processing for removing the influence of the pixel structure of the flat panel display from the input image and the processing for removing the background noise are eliminated, so that the preprocessing unit is largely cut, and the defect candidate extracting unit By adopting an adaptive binarization method that actively uses the pixel structure, a flat panel display that can save labor and speed up the inspection process and reduce the cost of the final flat panel display. An automatic unevenness detection apparatus and an automatic unevenness detection method can be obtained.

1 is an overall configuration diagram of an automatic unevenness detection apparatus for a flat panel display according to Embodiment 1 of the present invention. It is the flowchart which showed the series of processes by the image process part in Embodiment 1 of this invention. It is the flowchart which showed the series of processes by the image process part in Embodiment 1 of this invention. It is explanatory drawing which showed the specific process image by the image process part in Embodiment 1 of this invention. It is the flowchart which showed the series of processes by the conventional image processing part.

Hereinafter, preferred embodiments of an automatic unevenness detection apparatus and automatic unevenness detection method for a flat panel display according to the present invention will be described with reference to the drawings.
The invention of the present application is characterized by having the following means in a method for capturing a camera-captured image of a flat panel display that emits light in an arbitrary pattern into a computer and automatically detecting unevenness by image processing. To do.
(Feature 1) Means for locally detecting the maximum value and minimum value and calculating the median value from the detected maximum value and minimum value as a threshold value (Feature 2) Adaptive to the input image using the median value as a threshold value To binarize

Furthermore, it can have the following technical features.
(Characteristic 3) By performing the closing process and the opening process on the adaptive binarized image, the subtle changes are emphasized, the non-uniformity (unevenness) part is visualized, the closed region is labeled, Means for determining unevenness by threshold processing

Embodiment 1 FIG.
FIG. 1 is an overall configuration diagram of an automatic unevenness detection apparatus for a flat panel display according to Embodiment 1 of the present invention. The automatic unevenness detection apparatus for a flat panel display according to the first embodiment shown in FIG. 1 includes a camera 10, an image processing unit 20, and a display unit 30, and is a flat panel display (hereinafter referred to as “inspection target”). Automatic unevenness detection of “inspection panel 1” is performed.

  The image processing unit 20 performs image processing on the light emission state of the inspection panel 1 captured by the camera 10, thereby specifying a region where unevenness has occurred and causing the display unit 30 to display the specified location. Can do. The image processing unit 20 includes a preprocessing unit 21, a defect candidate extraction unit 22, and a determination unit 23.

The pre-processing unit 21 is a part that performs correction processing on an image captured by the camera 10 in order to perform processing by the defect candidate extraction unit 22 at the subsequent stage. Specific examples of processing include the following. Can be mentioned.
-Enlargement / reduction processing-Geometric correction-Shading correction-Noise removal (smoothing filter, median filter, etc.)
・ Background image prediction removal

Further, the defect candidate extraction unit 22 is a part that extracts a candidate region of a defect portion corresponding to unevenness from the image that has passed through the preprocessing unit 21, and specific examples of processing include the following. Can be mentioned.
・ Enhancement processing ・ Edge detection (Laplacian filter, Sobel filter, etc.)
-First feature value calculation-Binarization-Isolated point removal-Expansion / contraction processing-Labeling

Furthermore, the determination unit 23 is a part that finally determines whether or not the candidate area extracted by the defect candidate extraction unit 22 is uneven. Specific examples of processing include the following. .
・ Second feature amount calculation ・ Identification (threshold judgment, classification, etc.)

  Next, the automatic unevenness detection method according to the first embodiment will be specifically described with reference to a flowchart and an explanatory diagram. 2 and 3 are flowcharts showing a series of processes performed by the image processing unit 20 according to the first embodiment of the present invention. FIG. 4 is an explanatory diagram showing a specific processed image by the image processing unit 20 according to Embodiment 1 of the present invention. Further, FIG. 5 is a flowchart showing a series of processes by the conventional image processing unit.

  First, the automatic unevenness detection method according to the first embodiment will be specifically described with reference to the flowchart of FIG. 2 and the explanatory diagram of FIG. In step S <b> 201, the image of the inspection panel 1 photographed by the camera 10 is input to the preprocessing unit 21 as an input image (see FIG. 4A). Although not shown in detail, the preprocessing unit 21 generates an image after camera noise removal is performed on the input image. Here, since the purpose is to remove minute noise from a camera-captured image, it can be configured with a simple Gaussian filter or the like having a small filter size.

  Next, in step S <b> 211 and step S <b> 212, the defect candidate extraction unit 22 performs maximum value detection processing and minimum value detection processing on the image after camera noise removal generated by the preprocessing unit 21.

  In the maximum value detection process and the minimum value detection process, the defect candidate extraction unit 22 detects the maximum value and the minimum value from the pixel data in the block unit of the predetermined size, and moves the block of the predetermined size in the pixel unit. Local maximum and minimum values are calculated sequentially.

Further, in step S213, the defect candidate extraction unit 22 calculates the median value of each pixel by the following equation (1) based on the maximum value detected in step S211 and the minimum value detected in step S212.
Median = (maximum value + minimum value) / 2 (1)

  Here, since the operation of dividing by 2 can be realized by bit shift, the circuit elements do not increase.

  Next, in step S214, the defect candidate extraction unit 22 compares each pixel of the image after the camera noise removal generated by the preprocessing unit 21 with the median calculated in step S213. Then, an adaptive binarized image binarized into white (1) and black (0) is generated (see FIG. 4B). More specifically, in each pixel, the input pixel larger than the median value is binarized to 1 (white), and the other input image is binarized to 0 (black), thereby generating an adaptive binarized image. .

  As can be seen from a comparison between the adaptive binarized image shown in FIG. 4B and the enlarged image of the input image shown in FIG. 4C, white and black in the defective portion and the non-defective portion of the panel. There is a difference in the distribution of. That is, in the adaptive binarized image, there are many black portions corresponding to portions having spot-like defects in the enlarged image.

  In order to further emphasize such distribution, in the next step S215, the defect candidate extraction unit 22 performs a closing process (expands and then contracts the white portion) to generate an image after the closing process. (Refer FIG.4 (d)). By performing such a closing process, white portions are connected and a gap between black portions is visible.

  Further, in step S216, the defect candidate extraction unit 22 performs black and white reversal processing to generate a black and white reversal image in which a portion corresponding to the defect portion is identified as a white closed region (see FIG. 4E). In step S217, the defect candidate extraction unit 22 extracts a white closed region from the black-and-white inverted image, and in subsequent step S218, performs a labeling process on the extracted closed region to generate a defect candidate image (FIG. 4 (f)).

  In the flowchart shown in FIG. 2, a case has been described in which a defect candidate image is finally generated by executing a closing process on an adaptive binarized image. However, the present invention is not limited to such a process, and the same effect can be obtained by performing an opening process instead of the closing process. This series of processes is shown in the flowchart of FIG. Equivalent to.

  The flowchart of FIG. 3 is different from the flowchart of FIG. 2 in that instead of the closing process in step S215 and the black and white inversion process in step S216, the black and white inversion process in step S315 and the opening process in step S316 are used. .

  That is, when performing the opening process, the defect candidate extraction unit 22 first performs the black and white reversal process in step S315 so that the portion corresponding to the defect part is identified as a white closed region. . After that, in step S316, the defect candidate extraction unit 22 performs an opening process (expands the white part and then expands it), and generates an image after the opening process.

  By performing such an opening process, an image equivalent to the case where the previous closing process is performed can be obtained.

  Finally, in step S221, the determination unit 23 specifies a defect part and its attribute with a determination rule based on experience with respect to the image generated in step S218 of FIG. 2 or FIG. To display.

  More specifically, the determination unit 23 calculates parameters such as a brightness value, a contrast ratio, and a SEMU value within each region of the non-uniformity (unevenness) region candidate specified in step S218 (here, “ “SEMU” is defined as a measurement unit of unevenness in SEMI (Semiconductor Equipment and Materials International) and corresponds to an abbreviation of “SEMI MURA”). Furthermore, the determination unit 23 can display the defect candidate image generated in step S218 on the display unit 30 and display each parameter obtained by the calculation in association with each candidate region.

  As a result, the inspector can make a true / false determination of the non-uniformity (unevenness) region by making a comprehensive determination based on the display content. Further, by setting in advance an evaluation value for determining whether or not unevenness has occurred from each parameter based on the experience of the inspector, the determination unit 23 sets each parameter obtained by calculation as each evaluation value. It is also possible to perform automatic unevenness determination by comparison.

  FIG. 5 is a flowchart showing a series of processing by the conventional image processing unit, and a comparison with the flowchart showing the series of processing by the image processing unit of the first embodiment shown in FIG. 2 or 3 will be described. As shown in FIG. 5, the pre-processing unit 21 in the prior art performs a filtering process that passes a predetermined band because the pixel unit structure of a flat panel display having a fixed pixel structure is reflected as light and dark in an image captured by a camera. The pixel structure has been removed (see step S502 in FIG. 5).

  Further, background noise is removed by performing background image prediction using a low-pass filter or the like having a large filter size and subtracting from the input image (see steps S503 and S504 in FIG. 5).

  However, as described above as the problem to be solved by the invention, the filtering in step S502 causes the characteristics of the defect to be detected to be blurred, and the subsequent defect detection process becomes difficult. There was a problem. Furthermore, depending on the type of defect to be detected and the type of panel, it is necessary to finely adjust the characteristics of the band-pass filter and the subsequent defect detection processing algorithm, which further increases the number of work steps.

  Further, in the filter processing in step S503, unless the filter size is increased to realize a sufficient low-pass effect, the unevenness detection performance is affected. For this reason, the burden of hardware increases and the problem that the processing time of software becomes long also arises, and the influence on the cost increase of the flat panel display which is a final product arises.

  In contrast, the image processing unit 20 according to the first embodiment does not perform pixel structure removal processing, background image prediction processing, and difference image processing by the preprocessing unit 21, and the defect candidate extraction unit 22 performs panel processing. The feature of the pixel portion is positively extracted, and the feature of the defective portion is extracted with simple logic and hardware by adaptive binarization processing. Furthermore, the feature of the defective part is extracted with higher accuracy by simple logic and hardware by the opening / closing process and the labeling process.

  As a result, the original defect can be extracted with high accuracy at high speed, and the labor saving and speeding up of the inspection process can be realized, thereby reducing the cost of the flat panel display as the final product. The automatic unevenness detection method according to the first embodiment is particularly effective for inspection of spot-like defects or defects having a certain area.

  As described above, according to the first embodiment, the defect candidate region can be extracted by actively using the pixel structure of the panel without performing the pixel structure removal process by the pre-processing. As a result, it is possible to obtain a flat panel display automatic unevenness detecting apparatus and automatic unevenness detecting method capable of realizing labor saving and high speed of the inspection process and achieving cost reduction of the flat panel display as the final product. .

  DESCRIPTION OF SYMBOLS 1 Inspection panel, 10 cameras, 20 Image processing part, 21 Pre-processing part, 22 Defect candidate extraction part, 23 Determination part, 30 Display part

Claims (3)

A pre-processing unit that captures an image of a light emission state of a flat panel display captured by a camera as an input image, performs pre-processing on the input image, and performs extraction processing of unevenness candidate areas on the output image of the pre-processing unit. By performing image processing by a defect candidate extraction unit to be performed and a determination unit that performs authenticity determination of the unevenness candidate region extracted by the defect candidate extraction unit, it is determined whether or not there is a nonuniformity region having a spot-like defect An automatic unevenness detection device for a flat panel display,
The pre-processing unit is a Gaussian configured as a size for removing camera noise included in an image captured by the camera without removing the influence of the pixel structure of the flat panel display from the input image. The camera noise is removed from the input image using a filter, and output to the defect candidate extraction unit in the subsequent stage,
The defect candidate extraction unit
The process of calculating the maximum value and the minimum value in the block of interest having a predetermined size for the output image of the preprocessing unit performs while moving the block of interest for each pixel, the local median for each pixel = (Maximum value + Minimum value) / 2
The output image of the pre-processing unit and the median value are compared on a pixel-by-pixel basis. When the output image is larger than the median value, 1 (white), otherwise 0 (black). By adaptively binarizing using the median value calculated locally, an adaptive binarized image serving as the unevenness candidate region is generated,
A black and white reversal process is performed on the adaptive binarized image after the expansion / contraction process, or an isolated point is removed and regions are connected by executing a contraction / expansion process after the black / white reversal process, and then a white closed region Extract
An automatic unevenness detection apparatus for a flat panel display that further narrows down the unevenness candidate areas by labeling the extracted closed areas and outputs them to a subsequent stage. In the automatic nonuniformity detection apparatus of the flat panel display of Claim 1,
The determination unit calculates a parameter based on a luminance value or a contrast ratio for each of the unevenness candidate regions calculated by the defect candidate extraction unit, and compares the calculated value with a predetermined evaluation value. Automatic flatness detection device for flat panel display that performs true / false judgment. A pre-processing unit that captures an image of a light emission state of a flat panel display captured by a camera as an input image, performs pre-processing on the input image, and performs extraction processing of unevenness candidate areas on the output image of the pre-processing unit. By performing image processing by a defect candidate extraction unit to be performed and a determination unit that performs authenticity determination of the unevenness candidate region extracted by the defect candidate extraction unit, it is determined whether or not there is a nonuniformity region having a spot-like defect An automatic unevenness detection method for a flat panel display,
Gaussian configured as a size for removing camera noise included in the image captured by the camera without removing the influence of the pixel structure of the flat panel display from the input image in the pre-processing unit. Removing the camera noise from the input image using a filter, and outputting to the defect candidate extraction unit at a later stage;
In the defect candidate extraction unit,
The process of calculating the maximum value and the minimum value in the block of interest having a predetermined size for the output image of the preprocessing unit performs while moving the block of interest for each pixel, the local median for each pixel = (Maximum value + minimum value) / 2 calculating step;
The output image of the pre-processing unit and the median value are compared on a pixel-by-pixel basis. When the output image is larger than the median value, 1 (white), otherwise 0 (black). Generating an adaptive binarized image that becomes the unevenness candidate area by adaptively binarizing using the median value calculated locally;
A black and white reversal process is performed on the adaptive binarized image after the expansion / contraction process, or an isolated point is removed and regions are connected by executing a contraction / expansion process after the black / white reversal process, and then a white closed region Extracting the
An automatic unevenness detection method for a flat panel display, comprising: further narrowing down the unevenness candidate regions by labeling the extracted closed regions and outputting the subsequent candidate regions.

Images