JP2007299248A - Method for correcting image position - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a method for correcting an image position capable of promptly and accurately calculating an amount of positional shift between the images while aligning a shot image of an inspection target and its reference image. <P>SOLUTION: All of the brightness data of each pixel G in a shot image S and a reference image M are added respectively along a vertical row direction and horizontal row direction for each pixel line X<SB>1</SB>, X<SB>2</SB>, ..., X<SB>n</SB>, Y<SB>1</SB>, Y<SB>2</SB>, ..., Y<SB>m</SB>for obtaining sum numeric sequences and a difference between the sums of a position corresponding to both images is calculated for each pixel line of vertical rows and horizontal rows. Furthermore, differences between the sums are obtained by shifting the pixel line one by one in each image and then a correction value is decided from a position having the least difference. In this process, a predetermined number of pixels on both the side end parts of a vertical row and a horizontal row of the images is excluded from objects to be compared, and so a maximum correction limit is obtained by subtracting the excluded amount. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a technique for correcting a positional deviation between images when, for example, a captured image obtained by photographing a mounting state of a substrate on which electronic components are mounted and a reference image thereof are inspected by pattern matching.

Conventionally, as a technique for inspecting a mounting state of an electronic component mounted on a board, a reference image obtained by photographing a board that is normally mounted is prepared in advance, and a photograph obtained by photographing the reference image and the board of the inspection object is prepared. A technique is known in which pass / fail is determined by comparing with an image. In recent years, in order to inspect the mounting state, it has become necessary to acquire an inspection object image at a high magnification as the mounted parts become finer, especially when shooting at a high magnification. For example, if the camera is moved to the inspection site on the XY stage, it is necessary to repeat the movement in units of microns to obtain accuracy. However, there is a limit in trying to increase the accuracy mechanically, and the acquisition of a photographed image without error has reached the limit. Nevertheless, in order to increase the inspection accuracy of such a miniaturized mounting component, it is necessary to superimpose and compare the reference image and the inspection target image without shifting one pixel.
Therefore, in the conventional technique, in such a comparison of two images, in order to align both images, a cell in a characteristic area having a reference image is selected as a template, and this selected cell is used as a template. A technique is known in which template matching is performed on another image, and the positional deviation amount of the image is obtained based on the matching result. (See Patent Document 1.)
JP-A-9-35063

  However, the conventional techniques as described above have inaccuracies in correction due to variations in the pattern of cells selected as a template and limitations on the location of selected cells, and it takes time to calculate the amount of misalignment. There was a problem.

  In view of the above, an object of the present invention is to correct the problems such as the limitation of the location of a selected cell and the inaccuracy of correction, and to calculate the positional deviation amount of an image quickly and accurately.

  In order to achieve the above object, the present invention provides a correction method for correcting a positional deviation between two images at the time of pattern matching in which a captured image of an inspection object is compared with a reference image. All the pixel lines are added in the vertical direction and the horizontal direction to obtain an addition numerical value sequence for each pixel line in each direction, and a difference between the addition numerical values of the corresponding pixel lines in both images is obtained. The difference is obtained while shifting the addition value of the addition value sequence one by one, and the correction value is determined by finding the position with the smallest difference.

In other words, if the conventional image position correction method is a method in which a characteristic cell in a narrow area is selected and aligned, in the case of the present invention, the entire image is overlooked and corrected with the brightness distribution of the entire image. This is a method for obtaining a value. Then, for example, if the pixel is moved by XX pixels in the horizontal direction and XX pixels in the vertical direction, the point that the difference in the brightness distribution of both the images can be minimized can be found, and this point is used as the correction value. . When the correction value is obtained, both images can be made to coincide by moving one of the images in the horizontal direction by XX pixels and the vertical direction by XX pixels. By doing so, the correction is performed more accurately and the calculation process is simple and simple, so that the correction value can be determined quickly.
The vertical and horizontal pixel lines referred to here are, for example, when the entire image is composed of horizontal 1000 pixels and vertical 500 pixels, with respect to the vertical direction 1000 pixel lines and the horizontal direction. 500 pixel lines are formed. In the addition numerical value sequence, an additional numerical value sequence consisting of 1000 numerical values is formed in the horizontal direction, and an additional numerical value sequence consisting of 500 numerical values is formed in the vertical direction.
Further, the “position where the difference between the addition values is the smallest” is a position where x / y by the least square method described later is minimized.

  Further, in the present invention, when the difference is obtained while comparing the added numerical value columns one by one in the vertical and horizontal directions, a predetermined number of pixel lines at both end portions of the vertical and horizontal rows are compared as the maximum correction limit. In addition, the maximum correction amount was set as the maximum amount of displacement.

That is, if both images are relatively moved by correction, the upper and lower and left and right end portions of the image corresponding to the correction value become meaningless regions where both images do not overlap. In addition, the area where both images do not overlap is an area where the brightness of the pixel cannot be obtained from the image on one side, and a special work or the like is required to compare from the brightness difference between the two images. If this area is excluded from comparison, no special means are required. In view of this, efficient comparison can be achieved by excluding predetermined areas at both the upper, lower, left and right end portions from the comparison target as the maximum correction limit.
The specific value of the maximum correction limit is determined based on experience values and the like.

  Further, in the present invention, when comparing the difference between the addition numerical value sequences of the photographed image and the reference image, when the total number of the addition numerical value sequences is y and the difference between the respective addition numerical values is x, the least square method is used. When x / y is larger than the specified amount, an error is determined.

When the reference image and the photographed image are overlaid and compared by pattern matching, it is not completely impossible to compare with completely different comparison objects. In such a case, since it is necessary to exclude as an error, such an error is determined by the least square method.
Here, the least square method is a method that approximates a function that approximates this most with a straight line from a value that includes an error such as measurement data, and when y is the number of samples and x is an error, y = If the slope x / y of the straight line represented by ax + b exceeds a certain amount, an error in the total number of samples is regarded as too large, and this is regarded as an error, and such a least square method is adopted. The error can be reasonably judged.

In the present invention, in addition to the numerical sequence obtained by adding the brightness of the vertical and horizontal rows of the photographed image and the reference image, a numerical sequence obtained by adding the lightness of the diagonal row is also obtained and compared.
In addition to the vertical and horizontal rows, by adding a numerical sequence that adds the lightness of the diagonal row, the image can be corrected not only in the X and Y axes but also in the rotation direction. can do.

When pattern matching is performed by superimposing the reference image and the captured image, the entire image of both images is looked over and the correction value is obtained from the overall brightness distribution, so that the overlay accuracy can be improved and the correction amount can be reduced. Can be determined quickly. Then, when calculating and comparing the difference between the numerical addition values in the vertical and horizontal directions one by one, it is possible to efficiently compare by excluding a predetermined number of pixel lines at both ends. It can be reasonably judged by adopting the least squares method for judging.
In addition, the accuracy of image alignment can be further improved by taking into account a numerical sequence obtained by adding the brightness of pixels in an oblique column.

Embodiments of the present invention will be described with reference to the accompanying drawings.
Here, FIG. 1 is an example of a reference or photographed image, and FIGS. 2 and 3 are explanatory diagrams for explaining the position correction method of the present invention.

  The image position correction method according to the present invention is an image processing inspection in which a captured image of an inspection object such as a mounting substrate is inspected by comparing it with its reference image. Instead of extracting a part as a selected cell and aligning both images based on this, it is possible to quickly determine the amount of misalignment of both images by looking over the entire image and instantly superimpose both images Has been.

  The basic principle of position correction according to the present invention is that a brightness spectrum is created based on the brightness of each pixel G, which is the minimum unit of image information as shown in FIGS. 1 and 2, and the brightness spectrum of a photographed image of the inspection object is obtained. Are compared with the brightness spectrum of the reference image and compared while changing the relative positions of both images, and the relative positional relationship with the best matching degree is obtained to determine the correction amount.

The specific method is as shown in FIG. 2, the pixel G now image, _Y 1 in the transverse direction (X _{direction) X 1, X 2 ... X} n n columns of the vertical direction (Y-direction), Y ₂ ... Y _m composed of m columns, and when X ₁ , X ₂ ... X _n and Y ₁ , Y ₂ ... Y _m are called pixel lines, the brightness of each pixel G is determined for each pixel line. All values are added in the column direction and the row direction to obtain an added numerical sequence in each direction. The addition numerical value sequence means, for example, that the addition value of the brightness of all the pixels G existing on the X ₁ pixel line is a simple single digit of 5, and all the values existing on the X ₂ pixel line The added value of the brightness of the pixel G of 6 is set to 6, the added value of the brightness of all the pixels G existing on the X ₃ pixel line is set to 7, and the same applies to all the pixels existing on the X _n pixel line. When the added value of the brightness of G is 4, the added numerical sequence in the X direction is represented by n numerical sequences such as 567.
Similarly, the numerical sequence in the Y direction is represented as m numerical sequences.
Incidentally, the number of pixels of a general image used in an image processing inspection apparatus or the like is about 1000 to 1400 pixels in the X direction and about 800 to 1100 pixels in the Y direction.

  Then, two types of numerical values such as a vertical / horizontal row of a reference image to be a reference and a vertical / horizontal row of a photographed image of the inspection object are obtained.

In the comparison of both images, the numerical values of the vertical and horizontal addition numerical values corresponding to each image are compared, and the difference is obtained while shifting the addition numerical values one by one in the vertical and horizontal directions. To be compared.
This will be described using specific numerical values as an example. Now, the number of pixel lines in the X direction of the reference image is 13, the added numerical value sequence is 5556768766555, and the added numerical value sequence in the X direction of the photographed image is 5676876655554. In this case, for example, when comparing the added numerical value sequence of the captured image with respect to the reference image while shifting one by one in the right direction, the difference (absolute value) between (5556768766555) and (* 567687665555) is | (* 011121101000) |, and the difference (absolute value) between (5556768766555) and (** 56768766555) is | (** 00000000000) | for two gaps, and (5556768766555) and (*** 5676876655) for the three gaps And the difference (absolute value) becomes | (*** 1112110100) |, and in the case of two gaps, the overall matching degree is the highest. Therefore, in this case, it can be seen that a correction amount of two pixels in the X direction is necessary.

  In the Y direction, correction values are obtained in the same manner, the correction amounts in the X and Y directions are immediately determined, and both images can be matched with high accuracy.

  By the way, when both the reference image M and the captured image S are made to coincide with each other by the correction procedure as described above, as shown in FIG. 3C, both end portions of the images M and S in the X direction and the Y direction are arranged. In addition, useless portions δx and δy that do not overlap are generated. The areas δx and δy are areas where brightness cannot be obtained from the image on one side and are unsuitable areas for comparison. In this embodiment, predetermined areas at both end portions of the image are preliminarily used. The pixel G of the portion is excluded from the comparison target as the maximum correction limit, and the maximum amount of shift when the images M and S are compared with each other is set as the maximum correction limit.

The maximum number of corrected pixels on both sides is determined by an empirical value based on the relationship between the amount of image shift caused by, for example, play of the image pickup apparatus, the range of the shooting area, and the like.
By the way, in this embodiment, the maximum number of pixels for correction is about 5 to 15 pixels near the end in both the X and Y directions (about 10 to 30 pixels on both sides) so that the inspection is not affected. It is set to a proper area.

  Further, it can be said that the spectrum of the reference image and the spectrum of the photographed image are not exactly the same, and in some cases, the comparison may be performed using completely different comparison objects. In this case, it is necessary to determine that the error is not subject to correction, and the minimum position of the difference between the addition values may be unknown. For this reason, in the present invention, these are discriminated using the least square method. Hereinafter, the determination method will be described.

  In the case of taking the addition numerical sequence in the X direction (5556768766555) of the reference image and the addition numerical sequence in the X direction (5676876655554) of the captured image as an example, the difference (absolute value) between the two is as described above. With both images as they are, | (0120112111001) | (total 11), with the captured image shifted to the right by one line of pixel, | (* 011121101000) | (total 8), shifted by two lines of pixel Thus, | (** 00000000000) | (total 0), with a shift of 3 lines of pixels, it becomes | (*** 1112110100) | (total 8), and it is clear that it can be matched by the shift of 2 lines of pixels. I understand.

  On the other hand, when the reference numerical image X-direction addition numerical sequence is (5556768766555) and the photographed image addition numerical sequence is (6866554934477), the difference (absolute value) between the two images remains as it is. 1310214232122) | (total 24), with the captured image shifted by one line to the right, | (* 130113333112) | (total 22), with a shift of two lines of pixel | (** 12103223211) | (Total 18), with the captured image shifted by one line in the left direction, | (311122142222 *) | (total 23), with the pixel shifted by two lines in the same direction, | (11013353212 **) | It is not clear which position is close to the reference image or the reference image is wrong.

Therefore, in the present invention, when the total number (number of samples) of the addition numerical sequence is y and the difference (error) is x, it is determined that an error occurs when x / y is larger than a specified value using the least square method. When x / y is within the specified value, the position where x / y is minimum is determined as the correction position. This least squares method is a method in which a plurality of data (x ₁ · y ₁ ) (x ₂ · y ₂ )... (X _n · y _n ) is obtained, and a straight line y = ax that is most likely to be fitted to them is obtained. + b is calculated, and when the horizontal axis is y and the vertical axis is x and the straight line y = ax + b is expressed, if x / y (straight angle of the straight line) is larger than the specified value, the number of samples The error is judged as an error because there is too much error in general y.
This is because, for example, when there is an extremely large error in the total number y of a certain added numerical sequence and there is almost no error in other places, x / y is small by adopting the least square method. On the other hand, if there is an error throughout the total number y of the added numerical sequence, even if the total error is smaller than the locally large error in the above example, x / y is a large value. This is a reasonable judgment method for judging the degree of matching of the images in general.

  Note that the addition of the brightness of each pixel G, the comparison of both images, the operation of the least square method, and the like are operations that can be quickly processed in the field of expertise for computers, and are calculated in a very short time and concluded. Is obtained.

  Further, in the above example, the brightness of each pixel of the image is added in the pixel lines in the vertical and horizontal directions to obtain an added numerical sequence in the X and Y directions, and the correction amount in the X and Y directions is obtained. However, a numerical sequence obtained by adding the brightness values of the diagonal direction and the diagonal columns parallel to the diagonal direction may be obtained and compared in both images. In this case, by adding an addition numerical value sequence in two directions in addition to the vertical row and the horizontal row, the rotational direction of the image can be corrected, and the correction can be made more accurately.

In addition, this invention is not limited to the above embodiments. What has substantially the same configuration as the matters described in the claims of the present invention and exhibits the same operational effects belongs to the technical scope of the present invention.
For example, the inspection object or the like is arbitrary. In addition, when comparing pixel lines while shifting, the addition numerical sequence of either image may be shifted.

  When aligning the reference image and the captured image, it is possible to calculate the correction amount quickly and accurately by determining the positional deviation based on the brightness distribution of each pixel of the image. In the inspection, the inspection time can be shortened and the inspection accuracy can be improved.

Example of reference image or captured image Explanatory drawing explaining the vertical and horizontal pixel line of the pixel of this invention Explanatory drawing for demonstrating the alignment situation of a reference image and a photographed image

Explanation of symbols

M: Reference image, S: Captured image, G: Pixel.

Claims (4)

A correction method for correcting a positional deviation between two images in pattern matching in which a captured image of an inspection object is compared with a reference image, and the brightness of each pixel of both images is set to each pixel in a vertical direction and a horizontal direction. All the lines are added together to obtain an addition value sequence for each pixel line in each direction, and the difference between the addition values of the corresponding pixel lines in both images is obtained, and the addition value of each addition value sequence is obtained in the vertical and horizontal directions. A method for correcting the position of an image, wherein the difference is obtained while shifting one by one and a correction value is determined by finding a position having the smallest difference. When a difference is obtained while comparing the added numerical value columns one by one in the vertical and horizontal directions, a predetermined number of pixel lines at both ends of the vertical and horizontal rows are excluded from comparison targets as the maximum correction limit, and The image correction method according to claim 1, wherein the maximum correction limit is a maximum amount of displacement. When comparing the difference between the addition values of the photographed image and the reference image, if the total number of addition value sequences is y and the difference between the addition values is x, x / y is a specified amount using the least square method. The image position correcting method according to claim 1, wherein if there is more, it is determined as an error. 4. The addition numerical value sequence obtained by adding the lightness values of the diagonal rows in addition to the addition numerical value sequence value obtained by adding the vertical and horizontal lightness values of the photographed image and the reference image is obtained and compared. The image position correction method according to claim 1.

Images