KR20050052034A - Method for detecting shaped mura in a light-related plate element for a flat panel - Google Patents

Abstract

According to the present invention, an image obtained from a light-related plate element for a flat panel is linear (lineality) existing in a background having nonlinearity by a direction projection subband feature using a directional filter bank (DFB) and a line projection technique. ) Provides a method for efficient automatic detection of stains.

The linear spot detection method is a defect detection method of a flat panel light related plate element for obtaining an image from a flat panel light related plate element and detecting defective linear spots through image processing. A median filtering step of filtering by a median value so as to preserve the sharpness of the boundary while reducing the influence of noise caused by pixels having a large difference in brightness from the surroundings; In order to apply the directional filter bank (DFB), the zero padding step of zero-inserting the original image because the image should be a square of multiples of eight; A directional filter bank (DFB) processing step of separating the directional band components of the input image into eight directional bands by using the directional filter bank; A projection step of projecting the entire area of the DFB result so that the linear component becomes more pronounced; And a linear spot discrimination step of determining the position and width of the linear spot at positions and intervals of two peaks having a different sign and a large absolute value from the projection result data.

Description

Method for detecting shaped mura in a light-related plate element for a flat panel

The present invention relates to a method for detecting linear spots of light related plate elements for flat panels, and more particularly, to automatically detect linear spots in detecting defects through image processing by acquiring an image from the light related plate elements for flat panels. It relates to a method that can be detected.

Currently, flat panels, such as LCDs and PDPs, are constructed by stacking / incorporating light-related plate elements such as glass substrates or optical pattern plates. appear. In other words, various defects such as uneven color, dust, cracks, scratches, and other defects such as point defects, predecessors, and certain types of linear stains such as block shapes, and sublinear stains with undefined edges may appear. do.

In order to discriminate the defects of the optically related plate element, inspection and discrimination are usually carried out by visual inspection, which not only lowers the productivity and accuracy of the inspection, but also consumes a lot of cost and lacks reproducibility in detecting the defects. Therefore, in order to obtain objective and cost-effective effects, it is necessary to introduce an automatic inspection system.

In this way, the automatic inspection system is capable of reproducible processing in real time using advanced computer vision technology and high-speed image processing algorithms, thereby improving productivity as well as efficient quality control. However, there are some difficulties in using image processing by acquiring an image at a high resolution. First, the luminance level of the panel itself appears nonlinearly over the entire surface, and the luminance according to the activation time even after a signal is applied to the panel. The distribution will change. Second, each model has different image characteristics according to the position of the back light and the internal structure of the diffusion sheet. Due to these difficulties, it is difficult to apply image segmentation techniques such as general edge operators.

 Accordingly, the present invention is to solve this problem, and relates to an image processing algorithm for automatic inspection of semi-finished or finished products for flat panel displays (FPD), a high level of inspection process that relies on the existing visual inspection Its main purpose is to enable efficient quality control as well as productivity by reproducible processing in real time using computer vision technology and high-speed image processing algorithms.

To this end, the present invention, the linear image existing in the background having a non-linearity through the direction subband feature and line projection technique using the directional filter bank (DFB) obtained from the light-related plate element for flat panel (Linearity) Provides a method for efficiently and automatically detecting spots.

In order to achieve the above object, a linear spot detection method of a flat panel light related plate element according to an embodiment of the present invention detects defective linear spots through image processing by acquiring an image from the flat panel light related plate element. In the defect detection method of flat panel optical related plate element for filtering, filtering by the median value to preserve the sharpness of the boundary while reducing the influence of noise caused by the pixels with much difference in brightness from the surroundings in the input image A median filtering step; In order to apply the directional filter bank (DFB), the zero padding step of zero-inserting the original image because the image should be a square of multiples of eight; A directional filter bank (DFB) processing step of separating the directional band components of the input image into eight directional bands by using the directional filter bank; A projection step of projecting the entire area of the DFB result so that the linear component becomes more pronounced; And a linear spot discrimination step of determining the position and width of the linear spot at positions and intervals of two peaks having different absolute values from the data of the projection result.

Flat panel light-related plate elements that can be inspected include TFT-LCD finished products, LCD semi-finished products such as BLU and diffuser plates, organic EL, and PDP, and can also be applied to ITO glass and optical film series. The image related to the flat panel product is not only difficult to acquire, but the acquired image is high resolution and has non-linear luminance change in the image. Therefore, in the present invention, the use of pre-processing and post-processing techniques for image improvement, and high-speed inspection algorithm for real-time inspection It is characterized by.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a flowchart illustrating a method for detecting linear spots of an optically related plate element for a flat panel according to an embodiment of the present invention.

In Figure 1, the linear spot detection method of the light-related plate element for a flat panel of the present invention is basically a median filtering step (step S2), a zero padding step (step S3), a directional filter bank processing step (step S4), projection It comprises a step (step S5), and a linear spot discrimination step (step S7), preferably, further comprises a low frequency component removing step in step S6.

Linear smudges in LCD panels are caused by a number of causes. When the transistor fails due to static electricity or voltage imbalance occurs, it occurs horizontally or vertically, and when it is caused by rubbing of the rubbing cloth during the rubbing process, it appears in the form of several lines in the moving direction of the panel. The linear blobs are elongated from the beginning to the end in the horizontal or vertical direction of the image and are not constant in position, width, and brightness. An example thereof is illustrated in FIG. 2A, and an example of the histogram leveling image is illustrated in FIG. 2B. In addition, a scan stain occurs in a scanning process, and is an uneven color appearing in the rainbow image in the whole image, and an example image is shown in FIG. 2C, and the histogram-leveled image is shown in FIG. 2D.

The directional filter bank (DFB) employed in the present invention was first proposed by RHBamberger and Smith (MJTSmith), and the method used in the present invention uses frequency scrambling. The removed method was used. As shown in FIG. 3A, the direction-specific band components of the input image are separated into eight direction-specific bands as shown in FIG. 3B using the directional filter bank. 3C and 3D show examples of a test image for applying the directional filter bank and an example band image for each direction. The result of applying the DFB to the original image of FIG. 3C is shown in FIG. 3D by stretching to 0 to 255 levels. It can be seen that only a straight line in a specific direction stands out for each band. As shown in FIG. 3D, the horizontal defect is 5, 6 band, and the vertical defect is characteristic in bands 1 and 2, so that important information regarding the linear defect in a specific direction can be obtained by using each band. It can be seen that.

In one embodiment of the present invention shown in Figure 1, in the pre-processing step to reduce the influence of noise in the original image input in step S1 and to create an image of a size that can be applied to the DFB median in steps S2 and S3 The filtering step (median filtering step) and the zero-padding step, that is, the zero insertion step, are performed. In other words, the difference in brightness value between the defective part and the peripheral part of the acquired image is only about 3 in the 8-bit gray image, and the boundary is not clear, and there is noise such as stepping by the noisy pixel of the camera. Can be. Pixels with a large difference in brightness from the surroundings use a median filtering that preserves the sharpness of the boundary while reducing the effects of noise because they act as a strong directional component to the DFB application result. In order to apply the DFB, since the image must be a square that is a multiple of 8, zero insertion is performed on the original image.

After preprocessing the input image in this way, in step S4, the band component for each direction of the input image is separated into 8 bands for each direction by using the directional filter bank (directional filter bank (DFB) processing step). After passing through the DFB, the coefficient values of each band have a real value between -1 and 1, and values for detecting the linear components in the horizontal direction in the 5th and 6th bands appear. The values in these two bands have opposite signs. FIG. 4A illustrates an image having a linear speckle created using a signal generator, and FIG. 4B illustrates a stretched result of applying a DFB to an input image. In the image of FIG. 4B, a horizontal band exists at 1/2 of bands 5 and 6 and bands 5 and 6 have different codes.

Subsequently, in step S5, the projection step is performed. A slightly different projection process is used to detect linear and scanned blobs. In both cases, the projection is horizontal. Linear blobs appear straight in the image, but scan blobs do not have a straight line component on the left and right sides of the image, but have a horizontal line component only in about 30% of the area. Therefore, in the case of linear staining, the entire area of the DFB result is projected, and the scanning staining projects only 30% of the center. Since the codes of the bands 5 and 6 representing the linear components are opposite to each other, the difference between the bands 5 and 6 is projected to make the linear components more prominent. Longitudinal linear smears are processed after rotating the image 90 degrees. Equations (1) and (2) shown in Fig. 6 represent projections of linear speckles and scan stains, respectively. In equation (1), ω is the horizontal length of the whole image. In the DFB result, the width of bands 5 and 6 is reduced to 1/4 of the original image, and the length is reduced to 1/2. 5A shows a screen diagram of an image of a projection result using Equation (1).

Then, it is preferable to remove low frequency components by high pass filtering (HPF) from the projection result 1-dimensional data (step S6). As shown in FIG. 5B, the resulting image after removing the low frequency component caused by the nonlinearity of the luminance distribution through the high pass filtering (HPF).

Then, in the linear spot determination (Defect analysis), the position and width of the linear spot are determined from the position and the interval of two peaks of different sign and large absolute value from the data of the projection result (step S7). The results relating to such linear blobs are output or stored in step S8.

On the other hand, the method of the present invention can be carried out by systems that are configured as the inspection apparatus of the light-related plate element for the sensitive flat panel developed by the applicant and filed in application number 2002-11358 dated March 4, 2002. .

According to the configuration and operation of the linear spot detection method of the flat panel light-related plate element according to the embodiment of the present invention described above, the background having nonlinearity through the direction subband feature using the directional filter bank and the line projection technique There is an effect, such as being able to automatically detect the linear stain present in the efficiently.

1 is a flow chart showing a method for detecting a linear stain of the light-related plate element for a flat panel according to an embodiment of the present invention,

2A and 2B are screen views illustrating an example of an image including a linear spot and a histogram leveling image for the linear spot detection method of FIG. 1, and FIGS. 2C and 2D include scan spots similar to linear spots. A screen diagram showing an example of an image and its histogram leveling image,

 3A and 3B are diagrams illustrating a frequency division map of an 8-band directional filter bank, and FIGS. 3C and 3D are screen diagrams of test images and band images for each direction for applying the directional filter bank.

4a and 4b is an image of the result of processing the original image and the original image of the linear spot with a directional filter bank,

5A and 5B are screen diagrams of an image of a projection result and a high pass filtering result;

6 is an equation showing projection of linear speckles and scan stains.

<Description of the symbols for the main parts of the drawings>

S1: video input step S2: median filtering step

S3: zero padding step S4: directional filter bank processing step

S5: Projection Step S6: Low Frequency Component Removal

S6: Linear spot discrimination step S7: Result output storage step

Claims (3)

In the defect detection method of the flat panel light-related plate element for acquiring an image from the flat panel light-related plate element and detecting the defective linear stain through the image processing, A median filtering step of filtering the median value so as to preserve the sharpness of the boundary while reducing the influence of noise caused by pixels having a large difference in brightness from the surroundings in the input image; In order to apply the directional filter bank (DFB), the zero padding step of zero-inserting the original image because the image should be a square of multiples of eight; A directional filter bank (DFB) processing step of separating the directional band components of the input image into eight directional bands by using the directional filter bank; A projection step of projecting the entire area of the DFB result so that the linear component becomes more pronounced; And And a linear spot discrimination step of determining the position and width of the linear spot at positions and intervals of two peaks of different sign and absolute value from the data of the projection result. Linear smear detection method. The method of claim 1, further comprising removing low frequency components by high pass filtering (HPF) in the projection result 1-dimensional data between the projection step and the linear spot discrimination step. Linear smudge detection method of light related plate element for panel. According to claim 1, In order to detect the horizontal linear unevenness in the projection step, since the sign of the band 5, 6 indicating the linear components are opposite to each other, the difference of the band 5, 6 is projected, the vertical linear The smudge is a linear smudge detection method for a flat panel light-related plate element, characterized in that the image is processed after rotating the image 90 degrees.