JPH10160632A - Image processing method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To detect a low contrast defect, e.g. rubbing streak, surely by subjecting an input digital data to wavelet conversion and detecting an area defect while quantizing over a wide range of a screen. SOLUTION: A digital image data is taken into an image memory (S1 ) and subjected to wavelet conversion (S2 ). Wavelet conversion is a signal processing method for converting a time region signal into a frequency region signal and analyzing the signal. An image energy is then calculate (S3 ) at a part where high-pass information in the direction of x-axis of an image data subjected to wavelet conversion is combined with high-pass information in the direction of y-axis. A criterion threshold is set for the quantity of image energy and a decision is made (S4 ) whether an image exceeds the threshold to represent a rejectable image containing a rubbing streak or the image does not exceed the threshold to represent a high quality acceptable image. The decision results are displayed (S5 ).

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image processing apparatus used for an image quality inspection apparatus such as an LCD (liquid crystal) tester or a CCD tester for inspecting an object using an image and an image recognition apparatus for recognizing the object using an image. About the method.

[0002]

2. Description of the Related Art Conventionally, there are many image processing methods for inspecting, recognizing, judging, and diagnosing an object using an image. In these methods, most digital images handle images digitally, and image processing is performed using a computer.

In a digital image handled by an image quality inspection method used for an LCD tester, a CCD tester, or the like, each pixel (each point for forming one screen) in the image has a plurality of gradations such as 256 gradations or 1024 gradations. Are 640 pixels in the X-axis direction and 480 pixels in the Y-axis direction, for example.
It is an aggregate in which pixels are arranged. When this image is, for example, uniform gray image information, if it is an ideal image, all the pixels of the digital image have the same pixel value, that is, the same density value. However, in actuality, since the noise of the signal level and the slight variation among the pixels are present, the pixel values of all the pixels cannot be all the same. If the variation of the value is about ± 20 gradations, a uniform and clear screen can be seen by humans.

[0004] However, the CC which constitutes the image processing apparatus is used.
If the D or LCD has a defect and the pixel value has a variation of about ± 20 or more, for example, while the surrounding pixels have a pixel value of about 400, only the pixel in question has a pixel value of 5
If you have a value like 00, this will be noticed by humans as soon as you see it. Such a pixel is called a defect, and a pixel in which a pixel having a difference of at least a certain degree from the pixel value of a surrounding pixel exists alone at only one point is called a point defect, and the point defects are arranged on a straight line. The thing is called a line defect.

Further, even if a single point does not significantly differ from the pixel value with the surroundings and a pixel is not recognized as a point defect, some pixels having a slightly higher or lower pixel value are partially aggregated. If the pixel value variation in a portion is larger than in other portions, a human recognizes "unevenness" in this portion. Such a defect as an aggregate of a plurality of pixels is called an area defect.

In an LCD panel, it is necessary to give an orientation to liquid crystal molecules. For this purpose, an orientation film is formed on the surface of a glass substrate in contact with a liquid crystal layer. And L
During the manufacturing process of the CD panel, there is a step of rubbing the alignment film in a certain direction with a cloth, which is called a rubbing treatment. The liquid crystal molecules injected between the two glass substrates have their long axes aligned in the direction in which the alignment film is rubbed by the rubbing treatment. In an LCD panel, if the arrangement of liquid crystal molecules becomes non-uniform, the amount of transmitted light becomes non-uniform.
Rabink treatment is a very important step. However, the rubbing of the alignment film may become uneven in a streak shape for some reason. As a result, when a uniform white image is entirely displayed on the manufactured LCD panel, a streak-like luminance unevenness occurs in the displayed image. This uneven brightness is referred to as a “rubbing line”.

An image quality inspection apparatus detects these defects to determine whether the image is good or bad. The image quality inspection apparatus outputs the position, area and other information of these defects as a result by image processing. 2. Description of the Related Art Conventionally, threshold processing is employed to detect a point defect or a line defect in an image to be inspected for image quality. Since a point defect or a line defect is basically detected as a single point defect, the pixel value of the defective portion is sufficiently larger or smaller than the variation of the pixel value of the surrounding portion. The value is set, and if the defect is a bright defect, the pixel above the threshold is determined as a defect. If the defect is dark, the pixel below the threshold is determined as a defect, and the number of pixels or the pixel value of each defect is output.

On the other hand, detecting an area defect requires complicated processing. This is because the pixel value of each pixel in a portion that does not have an area defect is absorbed in the variation in the pixel value of a portion that does not have an area defect, and only pixel information indicating whether the pixel is a defect or noise is determined. It is because it cannot judge from the. At present, for example, the following processing is performed to detect an area defect.

First, an original image A which is an original image to be inspected
Is subjected to median filter processing to generate an image B from which point-wise noise components have been removed. Next, a difference image C between the filtered image B and the original image A is created. As a result, an image containing only noise components in units of points is obtained. The difference image C is subjected to a binarization process to create a binary image D. In this binarization, the threshold
A pixel that can become a defect is defined as an active pixel having a pixel value of 1, and a pixel having a pixel value equal to or less than a threshold is defined as a pixel value of 0. Pixel value 1 of binary image D
Active pixels may have components of noise, point defects, line defects, and area defects. The difference between the area defect portion and the non-area defect portion in the binary image D is that the pixel having the pixel value 1 is aggregated in the area defect portion, and the other portion is sparse. Therefore, the binary image D is subjected to image processing for removing the isolated point, that is, the pixel in which the pixel value of the target pixel is 1 but the pixel values of eight adjacent pixels are all 0.

A labeling process is performed on an image from which isolated points have been removed, and connected pixels are grouped. That is, a pixel having an image value of 1 is searched for in a binary image, a label is assigned to the pixel, and then, if there is a pixel connected to each pixel having an image value of 1, there is an identical label, that is, grouping is performed. In the area defect portion, since the pixels having the pixel value 1 are aggregated, the area of each label, that is, the number of pixels in the same group becomes large when labeling is performed.

[0011] The area of each label is calculated, and only labels having an area larger than a certain area are left. By this processing, a noise component that is not an area defect is removed, so that an area defect can be detected.

[0012]

Although the above-described area defect detection technique has already been established, rubbing stripes have extremely low contrast caused by a slight difference in the force of rubbing the alignment film. Since the luminance unevenness is about 3%, it is not easy to detect even if the above-described image quality inspection apparatus is used. This is because the contrast of the rubbing streak is almost equal to that of the noise, so that if the threshold value is lowered, the noise is also detected, and only the rubbing streak cannot be selectively detected.

When rubbing lines occur,
Streaks often occur over a wide area of an LCD image. Due to the nature of the processing, the above-described image quality inspection apparatus is limited to those having a small area for detectable area defects against such a widespread unevenness. In the case of a defect having a large area, for example, a defect having unevenness over 1/3 of the entire screen, information on such a large area defect at the time of generating the difference image C may be obtained by conventional processing. Lost. At present, there is no method for detecting such a defective defect over a wide area of the screen.

The present invention provides an image processing method for quantifying and detecting an area defect over a wide area of a screen.

[0015]

According to the present invention, input digital image data is subjected to wavelet transform processing on a screen constituting the image, and high-pass information in the x-axis direction in the wavelet-transformed image data is obtained. Then, an image energy amount in a portion where the image energy is combined with the high-pass information in the y-axis direction is obtained, and the inspection recognition of the object is determined based on the image energy amount.

According to another embodiment of the present invention, input digital image data is subjected to wavelet transform on a screen constituting the image, and x-axis high-pass information and y-axis high-pass information in the converted image data are used. The pixel value of the pixel is subjected to a threshold value processing for the combination portion of the binary image data into binary image data of an active pixel “1” and an inactive pixel “0”, and the number of active pixels in the binary image data is Count.

The binarized image or the image from which isolated points have been removed may be displayed as an image to make a judgment. Based on whether or not the counted number of pixels is equal to or greater than a predetermined value, it is determined whether the inspection target, which is the data source of the input digital image data, is good or bad. The wavelet transform is a signal processing method for converting a time-domain signal into a frequency-domain signal and analyzing the signal. For example, the book Charles K, CHUI ““ An Intro
ACTION DUCTION TO WAVELETS ”, Acad
Chapter in emic Press, 1992
1 (An Overview) and chapter 3
(Wavelet Transforms and T
The general description is given in the article entitled "Ime-Frequency Analysis". In image processing, the position of the time domain is used instead of time, that is, in the case of processing in the x-axis direction, the x coordinate of each pixel is used, and the value of the pixel at that position, that is, the luminance value is plotted instead of the signal amplitude. The wavelet transform is performed on this coordinate system in the same way as the time-frequency analysis, and the same analysis is performed thereafter. Wavelet transform for images is described in M. Barland,
“Wavelets in Image Commun
ICATION ", ELSEVIER, 1994.

The wavelet transform may be performed a plurality of times, that is, the wavelet transform is performed once or once on the combined portion of the low-pass information in the x-axis direction and the low-pass information in the y-axis in the image data subjected to the single wavelet transform. After performing a plurality of times, the binarization processing or the image energy may be obtained for the combination of the high-pass information in both the x and y axes.

[0019]

FIG. 1 shows an embodiment of the present invention. First, digital image data is taken into the image memory (S
₁ ). Wavelet (Wavele)
t) Perform a conversion process (S ₂ ). Although the wavelet transform has been described in the above-mentioned document, the wavelet transform processing will be briefly described here.

The wavelet transform for image data is a two-dimensional wavelet transform, which is realized by combining a one-dimensional wavelet transform for the x-axis direction and a one-dimensional wavelet transform for the y-axis direction of the image. First, the one-dimensional wavelet transform processing will be described. Although there are many basis functions used for performing the wavelet transform, a description will be given here using a Haar wavelet having the simplest structure. For other wavelet basis functions (filter coefficients of wavelet transform), the information to be output is almost the same except for the shape of the function.

At the time of wavelet transform, a basis function is composed of two types, a scaling function and a wavelet function. The scaling function is a function for outputting smoothing information of data, that is, low-pass information, and the wavelet function is a function for outputting detailed information of data, that is, high-pass information. In the case of the Harr wavelet, the scaling function is g ₀ = g ₁ = １ ／ and the wavelet function is h ₀ =
、, h ₁ = −２.

A sequence (data) S consisting of 16 numerical values
_{0, S 1, S 2,} ..., sequence after hard wavelet transform for S _{15 (data) t 0, t 1, ...} , t 15 , respectively obtained by the following calculation. _{t 0 = g 0 · s 0} + g 1 · s 1, t 8 = h 0 · s 0 + h 1 · s 1 t 1 = g 0 · s 2 + g 1 · s 3, t 9 = h 0 · s 2 + h _{1 · s 3 t 2 = g} 0 · s 4 + g 1 · s 5, t 10 = h 0 · s 4 + h 1 · s 5 · · · · t 7 = g 0 · s 14 + g 1 · s 15, t ₁₅ = h ₀ · s ₁₄ + h ₁ · s ₁₅ FIG. 2 shows a specific method of performing a wavelet transform on 16 number sequences (data). That is, FIG. 2A shows a scaling function, and FIG. 2B shows a wavelet function. When the wavelet transform is performed on the sequence s ₀ , s ₁ ,..., S ₁₅ shown in FIG. 2C, that is, two adjacent numerical values of the sequence are sequentially taken out as a set, divided into eight sets, and the left 1/2 to numbers S _L (= g ₀₎ multiplied by, even multiplied by 1/2 (= g ₁₎ with respect to the right of the numeric S _R, the sequence t ₀ by adding both multiplication results, t _1, ..., to give the t _7, also for each of the eight pairs, numeric S of the left
_L is multiplied by （(= h ₀ ), and the right numerical value S _R is −1 /
2 (= h ₁ ), and the results of both multiplications are added to obtain a sequence t ₈ , t ₉ ,..., T ₁₅ . FIG. 2D shows specific numerical values of the sequence t as a result of the operation for specific numerical values of the sequence s shown in FIG. 2C. The first half t _{0 of the} sequence t,
t _1, ..., t ₇ as a calculation result a scaling function with respect to the original sequence s, it is composed of a low-pass information of half the total length of the original sequence s, late _{t 8, t 9, ...,} t 15 Is composed of high-pass information of half of the total length of the original sequence s as a result of calculating the wavelet function on the original sequence s. FIG. 2E shows a graph of the sequence s, and FIG. 2F shows a graph of the sequence t. In each case, the horizontal axis shows the order of numerical values in the sequence, and the vertical axis shows the numerical values.

Although the resolution is reduced by half with respect to the original image by this wavelet transform, both low-pass information and high-pass information of the original signal can be obtained simultaneously. A sequence (data) t _{0 to} t ₁₅ obtained by one-time wavelet transform while the original signal has a resolution of 16
Indicates low-pass information and high-pass information with a resolution of 8. Noise in the original signal, for example, s ₁₅ = 5 in FIGS. 2C and 2E
If there is noise data with a value much larger than other numerical values such as 0, the high-pass information portion of the sequence t generally has data with a small absolute value, but in this case, t ₁₅ = −25 due to noise. Large information is output. Thus, the wavelet transform can be advantageously used to detect the noise component of the original data.

This wavelet transform can be applied to two-dimensional image data such as an LCD image. FIG. 3 shows an example in which a wavelet transform is specifically applied to image data. In the example of FIG. 3, the original image 21 (FIG.
A) is 640 × 480 digital data, and this image is first subjected to wavelet transform in the x-axis direction in the same manner as one-dimensional data. That is, 4 of a sequence (one-dimensional data) composed of 640 pieces of data arranged in the x-axis direction.
Wavelet transform is performed on each of the 80 data.
As a result, as shown in FIG. 3B, the image is divided into two parts on the left and right, the low-pass information image 22 on the left and the high-pass information image
Is obtained. The 640 × 480 image 2 shown in FIG. 3B
4 is now subjected to the same wavelet transform in the y-axis direction. In other words, the wavelet transform is performed on each of 640 pieces of a sequence (one-dimensional data) composed of 480 pieces of data arranged in the y-axis direction. This results in 6 shown in FIG. 3C.
A 40 × 480 image is obtained. FIG. 3C is 640 × 48.
The screen 0 is divided into upper and lower parts, and the y-axis direction low-pass information image 25 is on the upper side and the y-axis high-pass information image 26 is on the lower side.
Is obtained. Therefore, the image obtained in FIG.
3D, a screen of the same size as 640 × 480 is divided into four parts as shown in FIG. 3D, a low-pass information image 27 in both x and y directions in the upper left half, and a lower half in the upper right part. Image 28 combining high-pass information in the x-axis direction and low-pass information in the y-axis direction, image 2 combining low-pass information in the x-axis direction and high-pass information in the y-axis direction in the lower left quarter
9, the high-pass information images 30 in both the x-axis and y-axis directions have been obtained in the lower right-hand quarter. Here, the conversion process is first performed in the x-axis direction, and then the conversion process is performed in the y-axis direction to obtain the image of FIG. 3D. However, even if the order of x and y conversion is changed, the final two-dimensional wavelet transform is performed. After that, the same image of FIG. 3D is obtained.

An original image handled by an LCD tester or a CCD tester is usually a uniform image such as gray on the entire surface.
When such an image is subjected to wavelet transform, in the case of a clean image having no defects or the like, an image containing high-pass components after wavelet transform, that is, both x-axis and y-axis low-pass information images (part 27 in FIG. 3D) Almost no data appears in the converted images other than, and only the noise components generated when the image is captured are included, and most of the pixels have values close to 0. However, when an area defect having a certain large area is present in the original image, in the high-pass information image after the wavelet transform, pixels having a high pixel value are concentrated on a portion where the defect exists. By performing binarization processing, isolated point removal, and labeling processing on the image of the high-pass information, it is possible to detect an area defect as in the conventional method.

Here, the method of detecting an image defect according to the present invention is significantly different from the conventional method of detecting an image defect in that each image after the wavelet transform has a quarter of the area of the original image. That is, the binarization processing or the removal of the isolated points on the image after the wavelet transform is similar to the processing of dividing the original image into 2 × 2 pixels and processing this. Accordingly, there is a high possibility that an area defect component which is conventionally removed together with noise when removing an isolated point remains.

Further, an image subjected to low-pass processing in both the x and y directions after the wavelet transform (part 27 in FIG. 3D)
Is an image obtained by reducing the original image to one-fourth, so that this image can be further subjected to wavelet transform. If the wavelet transform is further applied to the x- and y-direction low-pass information portions of the image once subjected to the conversion process, the image output by this is further reduced to a quarter.
To an image having a size 1/16 of the original image. By performing image processing similar to the above-described image processing on the x- and y-direction bi-directional high-pass information portions of the converted image, an area defect can be similarly detected. The image processing in this case is the same as the processing in which the original image is divided into 4 × 4 pixels and processed, so that it is possible to cope with a large area defect over a wider range than the processing on one wavelet transform image. X of the image which became 1/16,
By performing a wavelet transform on the low-pass information portions in both y directions, it is the same as processing the original image having an area of 1/64 of the original image into 8 × 8 pixels. Such recursive and repeated application of the same wavelet transform is referred to as multi-resolution analysis. The first wavelet transform is referred to as a level 1 wavelet transform, and applying the same wavelet transform to the x- and y-direction low-pass information portions of the level 1 wavelet transform portion is referred to as a level 2 wavelet transform.

Brightness unevenness such as a rubbing streak, which is low in contrast but spreads over a wide area of the screen, is difficult to detect because it is covered with noise by conventional pixel-based processing. This can be easily and reliably detected.
The direction in which rubbing streaks occur depends on the direction in which the alignment film is rubbed, but this direction is determined to be a fixed direction, and is approximately 45 degrees with respect to the x-axis direction. Therefore, a rubbing streak also occurs in the direction of approximately 45 degrees. Therefore, in order to detect a rubbing streak in the wavelet transform result, a high-pass information portion (FIG. 3) is used in both the x-axis direction and the y-axis direction of the transform result, which is a portion indicating a line component in an oblique direction.
Attention should be paid only to the portion 30) in D.

In the x-, y-direction high-pass information portion 30 of the wavelet transform result, one pixel corresponds to a plurality of pixels of the original image. That is, the level 1 wavelet transform portion corresponds to 2 × 2 = 4 pixels, and the level 2 transform portion corresponds to 4 × 4 = 16 pixels. Therefore, since the noise component in the pixel unit that is not a defect is averaged, the influence of the noise component in the pixel unit on the converted pixel value is reduced, and the pixel value becomes close to zero. On the other hand, when there is a rubbing streak, a portion having a slightly higher pixel value and a portion having a slightly lower pixel value are alternately repeated at a certain interval, so that the pixel value of each pixel as a result of the wavelet transform becomes higher. Or lower. As a result, the difference between the pixel values of adjacent pixels increases, so that the absolute value of the pixel value of each pixel in the x- and y-direction high-pass information portions becomes larger than the pixel value of the portion without rubbing lines.

When the rubbing streak spreads over a wide area of the image, the number of pixels having a large absolute value of the pixel value increases in the x- and y-direction high-pass information portions of the wavelet transform result. Therefore, the presence or absence of a rubbing streak can be determined by quantifying the image energy of the entire portion, that is, by quantifying the absolute value of the pixel value of each pixel.

In order to quantify the absolute value of a pixel value over the entire screen, the absolute value of the pixel value is binarized by a threshold value, and the number of pixels having a pixel value equal to or larger than the threshold value is counted and quantified. However, in this embodiment, as shown in FIG.
Wavelet transform of the captured image data (S ₂ )
Its x, calculates the image energy of y directions pass information unit to use it in quality inspection (S _3). The image energy E is defined as follows when the number of pixels included in the screen is N and the pixel value of each pixel is x (i).

E = Σ _{i = 1} ^N × (i) ² / N If the obtained value of the energy E of the image is large, it can be determined that a rubbing stripe exists. A threshold of a predetermined criterion is provided for the magnitude of the image energy E. An image exceeding the threshold is a defective image having a rubbing streak, and an image below the threshold is a non-defective image. determination was carried out (S _4), and displays and outputs the result (S _5).

Alternatively, the obtained image energy E may be output and displayed so that the degree of oblique line components contained in the original image can be quantitatively known (S ₅ ). In the above embodiment, the wavelet transform, the calculation of the image energy amount, and the pass / fail judgment can be performed by a computer. Here, the operation and effect of the present invention will be specifically described with reference to FIG.

In FIG. 4, the image is simplified and shown as a one-dimensional sequence of pixel values a (i). That is, FIG. 4A is an image composed of 40 pixels, and the pixel value of the background is 0, but two defective pixels having a pixel value of 4 are arranged side by side, and an image in which the defective pixels occur periodically every four pixels is shown in FIG. Show. This periodic defect is assumed to be a rubbing line. Actually, the numerical sequence a (i) is vertically arranged as shown in FIG. 5, and the defective pixels having the pixel value 4 are shifted by one pixel for each row. Is done. When the wavelet transform is applied to this sequence a (i), as shown in FIG.
(I) is generated. The sequences b (0) to b (19) are the low-pass information of the original image, and the sequences b (20) to b (39)
Is the high-pass information of the original image. Since no characteristic appears in the high-pass information resulting from this wavelet transform, no defect can be detected at this stage.
In addition, although the defect, that is, the pixel value 4 is periodically generated in the low-pass information, these are all present as isolated points and cannot be distinguished from noise. Therefore, next, the low-pass information sequence b (0) ~
Further, a wavelet transform is applied to b (19). A sequence c (i) shown in FIG. 4C, which is a result of the level 2 wavelet transform, is obtained. Sequences c (0) to c (9) are level 2 low-pass components of the original image, and c (10) to c (10)
c (19) indicates a level 2 high-pass component of the original image.
The sequences c (20) to c (39) are the sequences b (20) to b (3
As in the case of 9), a high-pass component obtained by performing a wavelet transform of level 1 of the original image is shown. Here, the sequence c (10) to c of high-pass information of the level 2 wavelet transform
Focusing on (19) and finding the energy of this component, E = Σ _{i = 10} ¹⁹ c (i) ² /10=2.4. This indicates that a defect component has occurred.

Here, as shown in FIG. 4D, noise that is not a defect is mixed into the sequence a (i) of the original image. That is, the sequence d (i) is a (8), a (25), a of the sequence a (i).
(32) and a (38) in which noise having a pixel value of 8 is mixed in four places. On the other hand, when the level 2 wavelet transform is performed, a sequence e (i) is obtained as shown in FIG. 4E. When determining the energy of the sequence e sequence of the high-pass components of the wavelet transform of level 2 (i) e (10) ~e (19), E = Σ i = 10 19 e (i) 2 /10=2.4 next , The high-pass components of the level 2 wavelet transform of the noise-free sequence a (i), the sequences c (10) to c (1)
It has the same value as the energy of 9). Conventionally, it has been difficult to distinguish between a defective component having a pixel value of 4 and a noise having a pixel value of 8 which are continuous with two pixels. However, even if the pixel value of each defective component is small, it is spread over a wide range. By adopting a statistical method called wavelet transform in order to make use of the feature, the influence of noise is reduced, and the defect component can be corrected and quantified. Further, when the same processing is performed on an image having no periodic noise, the calculated energy E becomes close to zero. Therefore, by setting an appropriate threshold value, it is possible to determine whether the image quality is good or bad due to the influence of the rubbing streak component.

In the above description, the image energy is calculated after the wavelet transform. However, after the wavelet transform, the binarizing process may be performed in the same manner as in the related art. That is, as shown in FIG. 6, a digital image is fetched (S1), a secondary wavelet transform is performed on the digital image (S2), and then a high-pass information portion (portion 30 in FIG. 3) in both x and y axes directions is obtained. First, a binarization process is performed to convert the pixel into an active pixel and an inactive pixel, that is, a binary image having pixel values of 0 and 1 is created (S3).
The threshold value of the binarization may simply be a threshold value in which the pixel value is positive and a threshold value in which the pixel value is 0 or less. An isolated point removal process is performed on the binary image (S4). This processing may be performed by a conventional method. Since most of the noise components are isolated points, noise other than the area defect is almost removed by the processing by the removal of the isolated points. In other words, in this image, most of the pixels having the pixel value 1 are the area defect or the rubbing stripe defect.

Therefore, in this image, a pixel having a pixel value of 1,
That is, the number of active pixels is counted (S5), and whether the count is equal to or less than a predetermined criterion is determined as a non-defective product or a defective product (S6), and the determination result is output and displayed, for example (S7). ). It is also possible to output and display the counted number of active pixels without making a pass / fail determination to know the degree of image quality.

As a modification, the above-described process of removing isolated points,
That is, step S4 is omitted, and the binary image is counted in step S
5 may be supplied directly. That is, when the wavelet transform is performed once, the area of the high-pass information portion in both the x and y directions is 1
As described above, the influence of the isolated point is also reduced to 1/4, and if the wavelet transform is performed, for example, three times, the effect of the isolated point pixel can be almost ignored. That is, if the maximum value of the isolated point pixel is about 60 to 70 times the normal value, the isolated point removal processing step S4 may be omitted by performing the wavelet transform three times.

As will be understood from the above description, the advantage based on the wavelet transform described in the embodiment for calculating the image energy is similarly obtained when this binarization process is used.

[0040]

As described above, according to the present invention, a wavelet transform is applied to an image to be inspected for image quality, and defect detection processing is performed on high-pass information in both the x and y directions as a result of the conversion. Therefore, it is possible to easily and reliably determine the presence or absence of a defect, such as a rubbing stripe, which has a very low contrast and spreads over a wide area, and the degree of the defect.

[Brief description of the drawings]

FIG. 1 is a flowchart showing a processing procedure according to an embodiment of the present invention.

2A is a diagram showing an example of a scaling function, FIG. 2B is a diagram showing an example of a wavelet function, and C to F are data examples of a sequence s before conversion as a specific example of one-dimensional wavelet conversion processing, 3 is a diagram showing each data value of a sequence t, a change graph of data of a sequence s before conversion, and a change graph of data of a sequence t after conversion.

3A is a diagram showing an original image, and FIG. 3B is x for the original image of A;
C is an image subjected to wavelet transform in the direction, C is an image subjected to wavelet transform in the y direction with respect to the image of B,
D is a diagram for explaining a combination value of each pixel of B and C.

4A is a diagram illustrating an example of one-dimensional display of original image data having a rubbing streak defect, FIG. 4B is a diagram illustrating image data obtained by performing a wavelet transform on the original image data of A, and FIG. A diagram showing image data subjected to wavelet transform, D is a diagram showing an example of one-dimensional display of original image data having a rubbing streak defect including noise, and E is an image obtained by performing a wavelet transform twice on the original image of D It is a figure showing data.

FIG. 5 is a diagram illustrating a two-dimensional image in which the one-dimensional images of FIG. 4A are sequentially shifted by one pixel.

FIG. 6 is a flowchart showing a processing procedure according to another embodiment of the present invention.

Claims (12)

[Claims] An image processing method for inspecting, recognizing, and judging an object using an image, comprising: a wavelet transforming step of performing a wavelet transform on input digital image data on a screen constituting the image; An energy amount calculating step of calculating an image energy amount for a combination portion of the x-axis direction high-pass information and the y-axis direction high-pass information in the wavelet-transformed image data. 2. A pass / fail judgment step of judging pass / fail of the inspection object as the input digital image data source according to whether or not the calculated image energy amount is equal to or less than a predetermined value. . 3. The image processing method according to claim 1, wherein the energy amount calculation step is a step of dividing the sum of squares of each pixel value of the combination part of the high-pass information by the number of pixels. 4. An image processing method for inspecting, recognizing, and judging an object using an image, comprising: a wavelet transforming step of performing a wavelet transform of input digital image data on a screen constituting the image; A threshold value processing is performed on the pixel value of each pixel of a combination portion of the x-axis direction high-pass information and the y-axis direction high-pass information in the wavelet-transformed image data, so that a binary image of an active pixel and an inactive pixel is obtained. A binary processing step of obtaining data; and a pixel number counting step of counting the number of active pixels in the binary image data. 5. The image processing method according to claim 4, wherein the counted number of active pixels is equal to or less than a predetermined value.
A pass / fail judgment step of judging pass / fail of the inspection object as a data source of the input digital image data. 6. The image processing method according to claim 4, wherein an isolated pixel in the image is removed from the binary image data, and the binary image data from which the isolated pixel is removed is included in the counting step. The method includes an isolated point removing step of forming binary image data. 7. The image processing method according to claim 1, wherein the wavelet transforming step comprises: x-axis-direction low-pass information and y information in the wavelet-transformed image data of the input digital image data. In this step, the wavelet transform is performed at least once on the combined portion of the axial low-pass information. 8. Inspection and recognition of an object using the image,
A recording medium on which a program for executing a process of an image processing device for performing a determination or the like is recorded, wherein the program comprises: a wavelet transform step of performing a wavelet transform on input digital image data on a screen constituting the image; An energy amount calculating step of calculating an image energy amount for a combined portion of the x-axis direction high-pass information and the y-axis direction high-pass information in the wavelet-transformed image data. 9. The recording medium according to claim 8, wherein the program determines the quality of the inspection object as the input digital image data source according to whether the calculated image energy amount is equal to or less than a predetermined value. It is characterized by including a judging step. 10. A recording medium storing a program for executing a process of an image processing apparatus for inspecting, recognizing, and judging an object using the image, wherein the program stores input digital image data in the storage medium. A wavelet transform step of performing a wavelet transform on a screen constituting an image; and a pixel value of each pixel for a combination part of the x-axis direction high-pass information and the y-axis direction high-pass information in the wavelet-transformed image data. A threshold value process to obtain binary image data of an active pixel and an inactive pixel, and a pixel number counting step of counting the number of active pixels in the binary image data. Features. 11. The recording medium according to claim 10, wherein said program determines the quality of an inspection object as a data source of said input digital image data according to whether said counted number of active pixels is equal to or less than a predetermined value. And a pass / fail judgment step. 12. The recording medium according to claim 9, wherein the wavelet transforming step includes the step of converting low-pass information in the x-axis direction and low-pass information in the y-axis direction in the image data obtained by wavelet transforming the input digital image data. The wavelet transform is performed at least once on the combination part.