JPH09274659A - Image aligning method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To enable the exact alignment of an image by adding mutual correlation coefficients in the longitudinal and lateral directions of a search area and defining a position, where these added values respectively become maximum values, as a matching position. SOLUTION: Concerning the search area in an input image divided into the blocks of M×N picture elements, inside that search area, mutual correlation coefficients C (a, b) at respective positions are operated according to a specified expression. Then, the operated results of the mutual correlation coefficients are stored in arrangement R (a, b) (S1). Next, the arrangement R (a, b) is added in (x) direction, SX(1) to SX(M-X+1) are added in (y) direction, and SY(1) to SY(N-Y+1) are respectively provided (S2). Then, among the SX(1) to SX(M-X+1) and among the SY (1) to SY(N-Y+1), the positions to respectively take the maximum values, namely, a column number x0 and a row number y0 are extracted (S3). Finally, the column number x0 and the row number y0 are decided as the matching position (S4).

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image alignment method in an automated technique using image processing, for example, a character recognition technique, an automated product visual inspection technique, and the like.

[0002]

2. Description of the Related Art In an automated technique using image processing, when a predetermined reference image (template) and an input image have a relationship in which they overlap with each other by a parallel movement, the reference image is searched for in the input image. The operation of moving within the area to find the most suitable place, that is, template matching is often used. At the time of matching, it is important to accurately determine at which position in the input image the reference image matches. Therefore, in the automation technology using image processing, the cross-correlation coefficient method is often used in which the cross-correlation coefficient between the reference image and the input image is calculated while moving the reference image within the search area of the input image. Has been. FIG.
It is a figure explaining the template matching using the conventional cross correlation coefficient method. As shown in FIG. 3A, X ×
The cross-correlation coefficient C (a, b) is moved while moving the reference area of Y pixels in the search area in the input image of M × N pixels.
Is calculated. When the coordinates (a, b) at the upper left of the reference image move in the range from the coordinates (1, 1) to the coordinates (M−X + 1, N−Y + 1) and the cross-correlation coefficient is calculated, for example, A cross-correlation coefficient matrix as shown in 3 (b) is obtained. In FIG. 3B, since C (x0, y0) is 0.9, which is the maximum, it is determined that the same image as the reference image exists (matches) at the position of a = x0, b = y0. Regarding such a conventional technique, for example,
"Image Processing Handbook" edited by Morio Onoue, p303-
It is disclosed in p304.

[0003]

However, the conventional cross-correlation coefficient method as described above is sensitive to noise, and the maximum value of the cross-correlation coefficient becomes unclear. There is a lot of fear The noise referred to here is not only noise generated by an image capturing camera or an image processing device, but also a difference that always exists between the reference image and the input image, for example, when capturing the reference image and the input image. It is noise in a broad sense, including differences caused by changes in the lighting conditions of. Due to the above noise,
An example in which accurate matching cannot be obtained will be described with reference to FIG. Here, FIG. 4 is a chart for explaining erroneous determination in the conventional cross-correlation coefficient method caused by the noise. For example, in the matrix of the cross-correlation coefficient shown in FIG. 4, the cross-correlation coefficient is the position (x0-1, y0).
Takes the maximum value in the reference image and the input image information by the noise (x0-1,
This is because they overlap at the position of (y0), and the correct matching position is (x0, y0). In such a situation, in the conventional cross-correlation coefficient method, the matching position is determined only by the maximum value described above, and the surrounding correlation is not taken into consideration. Therefore, (x0-1, y0) is the matching position. Will be judged. SUMMARY OF THE INVENTION The present invention provides a method for improving the cross-correlation coefficient method to solve the problems in the conventional techniques as described above, and to provide a method for performing accurate image registration with almost no increase in calculation time. The purpose is.

[0004]

In order to achieve the above object, the present invention is one in which a predetermined reference image is two-dimensionally moved in a search area in an input image, and the reference image is set at each position of the search area. And the input image, a cross-correlation coefficient between the reference image and the input image is calculated by calculating a cross-correlation coefficient between the reference image and the input image. Is added in the vertical direction and the horizontal direction of the search area, and the position where the added value in the vertical direction and the added value in the horizontal direction are respectively maximum values is set as the matching position. Configured as a method. Furthermore, it is an image registration method in which the added value is the added value of the squared values of the cross-correlation coefficient.

[0005]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to the accompanying drawings to provide an understanding of the present invention. It should be noted that the following embodiments are examples embodying the present invention, and do not limit the technical scope of the present invention. FIG. 1 is a block diagram showing the processing procedure of the image registration method 0 according to the embodiment of the present invention, and FIG. 2 is a table showing the cross-correlation coefficient matrix of the image registration method 0. In addition, S1 and S in FIG.
2, S3 and S4 indicate processing procedures (steps), and x and y respectively represent the horizontal direction and the vertical direction of the image. FIG.
As shown in FIG. 4, the image registration method 0 according to the embodiment of the present invention moves a predetermined reference image two-dimensionally within a search area in the input image to thereby make a space between the reference image and the input image. This is the same as the conventional technique in that the cross-correlation coefficient is calculated and the maximum value in the above-mentioned cross-correlation coefficient matrix is used as the criterion. However, in the embodiment of the present invention, the cross-correlation coefficient is added in the x direction and the y direction (s
2) The position where the added value in the x direction and the added value in the y direction give the maximum value is extracted (s3), and is set as the matching position (s).
4) The point is different from the conventional technology. Hereinafter, details of the image alignment method 0 according to the present embodiment will be described. First, the reference image divided into blocks of X × Y pixels is two-dimensionally moved in the search area in the input image divided into blocks of M × N pixels. In this case, the coordinates (a, b) at the upper left of the reference image are from (1, 1) to (M−X +
1, N-Y + 1). Within the search region, the cross-correlation coefficient C (a,
Calculate b).

[0006]

[Equation 1]

In the above equation, I is the intensity value of the input image whose search area is (a, b) and whose coordinates are (x, y). T is the intensity value of the reference image whose coordinate is (x, y). I and T with a minus sign are the average values of I and T, the subscript of sigma is the variance value of I and T, and the calculation result of the cross-correlation coefficient obtained by this equation is the array R
It is stored in (a, b) (s1). Then the array R (a, b)
Are added in the x direction, and SX (1) to SX (M−X + 1)
Are added in the y direction, and SY (1) to SY (N−Y + 1)
Respectively (s2). And SX (1) to SX
Medium (M-X + 1) and SY (1) to SY (N-Y + 1)
In each of them, the position having the maximum value, that is, the column number x0 and the row number y0 is extracted (s3). Finally, the column number x0 and the row number y0 are determined as matching positions (s4). FIG. 2 shows the calculation results of the processing procedures s1 and s2 and the addition value in the x direction and the addition value in the y direction according to the processing procedure s3, and the row and column that take the maximum value in each (the row with the calculation result indicated by a thick frame and Column) shows the extraction results. still,
For comparison, the calculation result of the processing procedure s1 uses the same result as that of FIG. 4 used in the description of the conventional technique, and the position where the cross-correlation coefficient C (a, b) is maximum is (x0− 1,
The same applies to the fact that the cause of becoming y0) is due to noise.

In FIG. 2, it is not the position (x0-1, y0) that the image registration method 0 according to the present embodiment determines to be the matching position.
The position is (x0, y0). This is because the image registration method 0 adds the cross-correlation coefficient in each of the x and y directions and uses the position where both added values are maximum simultaneously as the matching position. , Position (x0-1, y0) [C (x0-1, y0) =
0.921] is rejected and the position (x0, y0) [C (x
0, y0) = 0.919] was judged to be the most comprehensively matched position. As described above, according to the present invention, it is possible to perform accurate image alignment that is less likely to be affected by noise. Further, the processing time increased in the image registration method 0 is substantially only the calculation time in the processing procedure s2. Since the calculation in the processing procedure s2 is extremely simple, it is negligible compared with the calculation time in the processing procedure s1, and the calculation time hardly increases.

[0009]

EXAMPLE In the image registration method 0 described above,
Although the processing procedures s1, s2, and s3 are separately performed, the calculation time can be shortened by mixing the processing procedures s1, s2, and s3. Such an image registration method is also an example of the image registration method in the present invention. Also,
Although the cross-correlation coefficient is added as it is in the processing procedure s2 in the image registration method 0, the square value of the cross-correlation coefficient may be added instead. Such an image registration method is also an example of the image registration method in the present invention. Further, in the image alignment method 0, the cross-correlation coefficient is calculated for all the coordinates in the search area, but when the search area is wide, the calculation is not performed for all the coordinates from the beginning. The image registration method 0 is limited to the predetermined coordinates of the search area, or for the skipped areas at appropriate intervals.
Evaluation may be performed and the coordinate intervals may be sequentially reduced to shorten the calculation time and align the images. Such an image registration method is also an example of the image registration method in the present invention.

[0010]

Since the image registration method according to the present invention is configured as described above, even if the image information has noise, the operation time is hardly increased as compared with the conventional technique. ， The images can be aligned accurately. Furthermore, if the added value is set to the added value of the squared value of the cross-correlation coefficient, the difference in correlation is emphasized,
It becomes possible to perform more accurate image alignment.

[Brief description of drawings]

FIG. 1 is a block diagram showing a processing procedure of an image registration method 0 according to an embodiment of the present invention.

FIG. 2 is a table showing a cross-correlation coefficient matrix according to the image registration method 0.

FIG. 3 is an explanatory diagram for explaining template matching in a conventional technique.

FIG. 4 is a table showing a cross-correlation coefficient matrix in the related art.

[Description of Symbols] x ... Horizontal direction in search area y ... Vertical direction in search area a ... x coordinate at upper left of reference image b ... y coordinate at upper left of reference image X ... Number of pixel divisions in x direction of reference image Y ... Pixel division number in the y direction of the reference image M ... Pixel division number in the x direction of the input image N ... Pixel division number in the y direction of the input image s1 ... Cross-correlation coefficient calculation s2 ... Mutual correlation in the x direction and the y direction Add the number s3 ... Extract the maximum value from the added value s4 ... Determine the matching position

Continued Front Page (72) Eiji Takahashi Eiji Takahashi 1-5-5 Takatsukadai, Nishi-ku, Kobe-shi, Hyogo Kobe Steel Works, Ltd. Kobe Research Institute

Claims (2)

[Claims] 1. A predetermined reference image is two-dimensionally moved in a search area in an input image, and a cross-correlation coefficient between the reference image and the input image is calculated at each position of the search area, In the image registration method in which the position where the cross-correlation coefficient is the maximum value is the matching position between the reference image and the input image, the cross-correlation coefficient is added in the vertical and horizontal directions of the search area, An image alignment method, wherein a position where the added value in the vertical direction and the added value in the horizontal direction have respective maximum values is set as the matching position. 2. The image registration method according to claim 1, wherein the added value is an added value of square values of the cross-correlation coefficient.