JP2007219704A - Image position measuring method, image position measuring device, and image position measuring program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a positioning method and device of sub-pixel precision capable of performing a high speed processing in a real time without complicating calculation even when similarity distribution is anisotropic and realizing a highly precise position measurement. <P>SOLUTION: The image similarity of input image data as the object of retrieval and the template data of a preliminarily input reference image is calculated so that a similarity map can be prepared. Then, one-dimensional peak position estimation is operated to x and y axes with the data point of the maximum similarity detected from the similarity map as a center, and the peak position of sub-pixel precision is temporarily estimated, and the similar primary peak position estimation is operated to adjacent x and y axes with the estimated peak position of the sub-pixels interposed, and the correction of the peak estimation position is operated by linear interpolation by using the estimation results so that an image position can be determined. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a method and apparatus for performing highly accurate position measurement of parts and marks using images in the FA field and the like.

As one of the sub-pixel-accurate position measurement methods using images, an object or mark can be obtained by estimating the image coordinates using pixel matching and estimating the peak coordinates by interpolating the values. A method of measuring the position of the sub-pixel with sub-pixel accuracy is used. As a specific example,
(1) Method for estimating peak by independently performing function fitting etc. on one-dimensional similarity data of x and y coordinates (2) Isotropic curved surface for two-dimensional similarity data of x and y axes Method of estimating peak by fitting (3) There is a method of estimating a peak on the assumption of an anisotropic two-dimensional model on two-dimensional data of x and y axes.
The method of (1) is widely used because it is simple and requires a small amount of calculation. However, when the distribution shape of the similarity is not isotropic, that is, when the distribution is not uniformly distributed concentrically from the estimated peak position. An error occurs. In a real image, it is not completely isotropically distributed.
In the method (2), the amount of calculation increases because two-dimensional data is processed. If the similarity distribution is anisotropic, an error will occur.
The method (3) can measure with high accuracy even when the similarity distribution is anisotropic, but the amount of calculation is large because the model becomes complicated.

In Patent Document 1, first, the grid point position having the maximum grid point value is determined, and then the position of the adjacent grid point having the next maximum grid point value is determined along the x-axis and the y-axis. A quadrangle containing the pixel maximum value is determined, and the one-dimensional partial maximum value is positioned along each of the four sides, and the intersection of two straight lines connecting the one-dimensional partial maximum values located on opposite sides of the quadrangle is determined. It is described that the position of the subpixel maximum value is estimated. However, these differ in the method for estimating the position of the subpixel maximum value and the method for correcting the peak estimated position, and as in the present invention, one one-dimensional estimated value estimated from three points centered on the maximum point is used as it is. There was no one that could obtain high measurement accuracy with simple calculation and high-speed real time.
JP 2002-517045 A

  In a field such as FA, high-speed processing in real time is required, and therefore a method and apparatus for realizing high-accuracy position measurement even when the similarity distribution is anisotropic without complicating calculation are provided. Objective.

  In the present invention, peak estimation by function fitting or the like is performed on the one-dimensional similarity data independently of the x and y coordinates, and the estimation result is corrected two-dimensionally so that the peak position of the similarity can be easily and highly accurately determined. To estimate.

  According to the present invention, the peak position is estimated by two-dimensional processing for each coordinate axis by one-dimensional processing, and correction is performed using these, thereby improving the peak position detection accuracy. Since peak estimation by function fitting or the like is a one-dimensional process, the amount of calculation increases little, and almost the same measurement accuracy can be realized with a small amount of calculation compared to a method of fitting a two-dimensional curved surface. Further, even when the similarity is anisotropically distributed, the peak position can be correctly estimated. As described later, the reliability of the similarity having the longest distance among the eight neighboring points having the maximum similarity is low. Since points are not used, it is possible to reduce the influence of similarity errors.

  The image similarity between the reference image template data and the search target image data is calculated, and the one-dimensional peak position about the x and y axes centered on the detected maximum similarity data point from the created similarity map An estimation is performed to once estimate the position of subpixel accuracy, and correction is performed from the same one-dimensional peak position on the adjacent x and y axes with the estimated subpixel peak position interposed therebetween.

  FIG. 1 is a processing flow showing an embodiment of the present invention, and FIG. 2 is a block diagram showing an embodiment of the present invention. In the process flow of FIG. 1, template data is first input in step S1, and image input is performed in step S2. In step S3, image similarity is calculated between the input image and template data. In step S4, a similarity map is generated. Then, the maximum similarity is detected in step S5, and the position of the subpixel is estimated in step S6. Thereafter, the estimated position is corrected in step S7, and the measurement result is output in step S8.

  In the block diagram of FIG. 2, an image input unit 1 that inputs an input image, an image similarity calculation unit 2 that receives input of template data and calculates an image similarity with the image input unit 1, and the image The similarity calculation unit 2 includes a sub-pixel position estimation unit 3 that estimates the position of the sub-pixel and an estimated position correction unit 4 that corrects the obtained estimated position of the sub-pixel. Is output.

  FIGS. 3 and 4 illustrate a method of estimating a peak position by curve fitting of correlation values, and further estimating a position of a subpixel. Specifically, the subpixel is combined with FIGS. 1 and 2. A method of estimating the position of will be described. FIG. 3 is a diagram for explaining the maximum similarity peak position estimating method according to the present invention. As a result of calculating the image similarity with the search image data on the y = 242 axis using the template data input template, x Among the three points = 319, 320, and 321, x = 320 points are estimated to be the maximum similarity.

  FIG. 4 is a diagram for explaining peak position estimation by curve fitting of correlation values. Three points (x = 319, 320, and 321) plotted with the correlation values for the three points on the y = 242 axis plotted on the vertical axis are shown. The corresponding P, Q, R) is fitted with a parabolic quadratic curve, and the x coordinate at which the apex of the parabola is located is the peak estimated position on the y-axis 242 (point xa in FIG. 3). Show.

  This will be described with reference to FIGS. 5 and 6 as a way of obtaining a one-dimensional peak estimation position on each coordinate axis. FIG. 5 is a diagram showing one embodiment of the x-coordinate correction method in subpixel position estimation, and FIG. 6 is a diagram showing one embodiment of the y-coordinate correction method in subpixel position estimation. The point where the degree of similarity is maximized in FIG. 3 is the point of x = 320 and y = 242 in FIGS. 5 and 6, and this is M, and the point on x = 319,321 on the y = 242 axis Are points A and B, respectively, and points on x = 321, 319, and 320 on the y = 241 axis are C, D, and E points, respectively. In FIG. 6, the points x = 320 and 319 on the y = 243 axis are F and G points, respectively.

  In FIG. 5, the point xa on the x coordinate is obtained from the points M, A, and B on the y = 242 axis similarly to the description in FIG. 3. Similarly, the image similarity calculation on the y = 241 axis is performed. As a result, among the three points x, 319, 320, and 321 of points D, E, and C, the point E with x = 320 is estimated to be the maximum similarity, and similarly to the above, three points of D, E, and C are obtained. On the other hand, as a result of performing the fitting with a quadratic curve and estimating the peak position, the xb point on the x coordinate is determined.

  Next, with reference to FIG. 6, a method for correcting the position of the sub-pixel based on the estimated one-dimensional peak position for the y coordinate will be described. In FIG. 6, three points of y = 241, 242, and 243 on the axes of x = 320 and x = 319, that is, E, M, F; D, A, and G in total, are shown in FIG. The equivalent is not shown, but the same processing as the x-axis is performed, and the estimated pixel position close to the maximum similarity in the sub-pixel on the x-axis between E and M of y = 241 and 242 on x = 320 Is determined. Further, the yb point at the pixel estimated position close to the maximum similarity in the subpixel between D and A on the x axis between y = 241 and 242 between x = 319 and x = 319 is determined by the same processing, and thus x The coordinates xa and xb points and the y coordinate ya point are determined. In FIGS. 5 and 6, the point H at the intersection of x = 321 and y = 243 is not used. That is, among the eight points A to H in the vicinity of the maximum similarity point M, the H point having the lowest distance and the low reliability of the similarity degree is not used, so that the effect of the similarity error is reduced. Can do.

  Next, a method for correcting the position in the sub-pixel will be described. As shown in FIG. 5, on the line connecting the xa point and the xb point on the two x coordinates estimated at two points of y = 242 passing through the maximum similarity point and y = 241 adjacent thereto, the above is obtained. This shows a method of further improving the accuracy of the x coordinate of the peak by internally dividing with the coordinate of the ya point on y = 320 of the peak to obtain the xc point. Similarly, in FIG. 6, the peak obtained above is shown on a line connecting two points ya and yb on two y coordinates estimated at two points of x = 320 passing through the maximum similarity point and y = 319 adjacent thereto. This shows a method of further increasing the accuracy of the y-coordinate of the peak by internally dividing with the coordinates of the xa point on y = 242 in order to obtain the yc point. Furthermore, both of these may be executed. By repeating alternately, the accuracy can be further improved.

  In the second embodiment, as described in detail in the first embodiment, the method for obtaining the xa point, the xb point, the ya point, and the yb point is the same, but the correction of the position in the sub-pixel is performed in two-dimensional coordinates. The correction is performed in a lump. As shown in FIG. 7, it is an intersection of a straight line connecting the xa point on y = 242 and the xb point on y = 241 and a straight line connecting the ya point on x = 320 and the yb point on x = 319. The positions of the xc and yc coordinates are positions within the corrected subpixel.

Processing flow showing one embodiment of the present invention The block diagram which shows one Example of this invention The figure explaining the peak position estimation method of the maximum similarity of this invention The figure explaining peak position estimation by curve fitting of correlation value An embodiment of the x-coordinate correction method in subpixel position estimation Example of correcting y-coordinate in subpixel position estimation Example of batch correction of two-dimensional coordinates in subpixel position estimation

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Image input part 2 Image similarity calculation part 3 Sub pixel position estimation part 4 Estimated position correction part

Claims (5)

An image position measurement method for measuring the position of an object based on the degree of matching of an image,
Inputting template data of a reference image prepared in advance;
Inputting image data to be searched;
Calculating an image similarity between the input image data to be searched and template data of the reference image, and creating a similarity map;
Performing one-dimensional peak position estimation on the x and y axes from three points centered on the maximum similarity data detected from the similarity map, and once estimating the position of subpixel accuracy;
Performing the same one-dimensional peak position estimation on adjacent x and y axes across the estimated subpixel peak position, and correcting the peak estimation position using those estimation results;
An image position measuring method for determining an image position.   2. The image position measuring method according to claim 1, wherein the estimated position is corrected independently on the x and y axes, and one or both of the peak coordinates are corrected.   2. The image position measuring method according to claim 1, wherein the correction of the estimated position is performed collectively by using both x and y coordinates and linear interpolation. An image position measuring device that measures the position of an object based on the degree of matching of images,
Template data input means for inputting template data of a reference image prepared in advance;
Image data input means for inputting image data to be searched;
A similarity map creating means for calculating an image similarity between the input image data to be searched and the template data of the reference image, and creating a similarity map;
One-dimensional peak position estimation is performed with respect to the x and y axes from three points centered on the maximum similarity data detected on the similarity map, the position of subpixel accuracy is once estimated, and the estimated subpixels are estimated. Peak position estimating means for performing the same one-dimensional peak position estimation on adjacent x and y axes with respect to each other,
Correction means for correcting the peak estimation position using the estimation result;
An image position measuring apparatus characterized by determining an image position. On the computer,
Inputting template data of a reference image prepared in advance;
Inputting image data to be searched;
Calculating an image similarity between the input image data to be searched and template data of the reference image, and creating a similarity map;
Performing one-dimensional peak position estimation on the x and y axes from three points centered on the maximum similarity data detected from the similarity map, and once estimating the position of subpixel accuracy;
The same one-dimensional peak position estimation is performed on adjacent x and y axes with the estimated subpixel peak position in between, and the peak estimation position is corrected using the estimation results to determine the image position. And steps to
A program for running

Images