JPS60140482A - Picture collation system - Google Patents

Abstract

PURPOSE:To execute picture overlapping with high accuracy by correcting (eliminating or restoring) the deficiency part of an input picture and by forming the parameter for pattern matching. CONSTITUTION:Picture data and the density pattern of the designated registration picture are read from a file 60. The designated picture data are stored in a picture memory device 30 and the density pattern is stored in a main memory device 20 respectively. The collation picture on an input original is read from a picture input device 50, and it is stored on the picture memory device 30. Coordinates of the rotational center of the collation picture are calculated, and then, the density pattern of the collation picture is calculated. The parameter is calculated concerning the collation picture, the required area is obtained based on the parameter, and the rotational angle of the collation picture which is necessary for overlapping of the registration picture and the collation picture is obtained. Finally, the collation picture is shifted in parallel so that rotational centers of the registration picture and the collation picture will coincide, and both pictures are overlapped by rotating by only the required angle.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Application of the Invention] The present invention mainly relates to an image matching method suitable for pattern matching processing of images having rotationally symmetrical contours, such as seal images.

[Background of the invention]

There is a high demand for automating image verification such as seal verification. Speaking of seal verification as a typical example, this image verification involves superimposing a registered image of a seal that has been filed in advance and a second seal image (verification image) read in from an input device and evaluating the degree of match between the two. The main focus is so-called pattern matching processing.

FIG. 1 is a conceptual diagram of an image matching system to which the present invention is applied. From the registered images stored in file 5,
Extracts the corresponding image data by specifying the image name, and
A matching image 1 is read from the double-sided input device 2, and pattern matching processing is performed on both image data. At this time, the processing device 6 evaluates the degree of coincidence between the two image data by translating the images one rotation and overlapping the images one time.

In particular, when the target image is a two-dimensional image with a rotationally symmetrical contour, it is necessary to calculate the position of the center of rotation, and the accuracy of this calculation determines the success or failure of the pattern matching process.

Conventional method (Kaneko, Kamae: A method for positioning seal imprint patterns using the correlation of peripheral density with respect to the centroid.

IE81-16), IE81-16), the second
As shown in the figure, the coordinates xm+n, x
A method is used in which the coordinates (XO, YO) of the center of rotation are calculated using the following formula based on mix, Ymln, Ym&X.

However, as is clear from seal stamps, etc., images input to a computer generally undergo fluctuations such as blurring or blurring, so X +nln, XrraX in equation (1)
.

The values of ymtn and Ymax are influenced by the conditions of the input image and are likely to have errors. Therefore, the simple method described above has a problem in that the calculation accuracy of the rotation center position decreases.

In order to solve this problem, as shown in Fig. fM −e3++ , 13+t , &
+, A22 The applicant has already proposed a method for calculating the position of the center of rotation by also utilizing a straight line that diagonally intersects the Nohoka coordinate axis (Japanese Patent Application No. 58-55071).

For example, two sets of perpendicular lines that form 45 degrees with the coordinate axes? fMpu31 +
42 +4, +42 and the external contact point with the image (Plj shown in FIG. 6).

I-1 to 4. The coordinates (Xo, Yo) of the center of rotation are determined based on the coordinates of j=1 to 2).

The basic concept will be explained below. Now, let the coordinates of point Plj be ("j+y'j), and the midpoint Ql of the two points pH, p12
Let the coordinates of (ul, vl) be (+-1~4°j=1~
2). That is, u + - (x N + x + 2)/2 Equation (2) v
I−(y H+y I 2 ) /2 Formula (3) Formula (2
Based on the coordinates of the four points (q+) shown in ) to (3), the rotation center (Xo, Yo) of the image is determined as follows.

Xo=f (ul, u2.u3.u4) Formula (4) Yo
-g(v11v2.v3.v4) Formula (5) where, r
+g, respectively, represent a function that determines the representative value of (, 1) lv amount). Various functions can be used as f and g. For example, arithmetic mean f=ΣU amount/41g−Σv
It can be defined as l/4. Still (ut
), (vl) by removing the maximum value and minimum value and defining it as the average of the remaining two data.

In any case, since the coordinates (Xo, Yo) of the rotation center are determined as the representative values of (ul) and (vl), the calculation error of the rotation center due to fluctuations such as blurring or blurring of the image can be compared with the conventional method. can be reduced.

On the other hand, in order to perform pattern matching 7f of image data, in addition to calculating the position of the rotation center as described above,
It is necessary to calculate the rotation angle of the image. Regarding this rotation angle calculation, the applicant and others have already invented it (patent application 1982-
176262) There is a different method, which is explained as follows. That is, as shown in Fig. 4, the rotation center (xo, Yo) is the starting point, and the angle θ (0≦
The number of black pixels (pixels with a density of 1") existing on the half-line L(θ) forming θ≦2π) is represented by the symbol f(θ) (see Figure 4). f(0) is O It is defined for each θ of ≦θ≦2π, and this is called a 1a degree pattern. The density pattern of each of the verification image and registered image is r+(θ).

When f, (, the rotation angle ψ required to superimpose both images is calculated using the following formula.

ψ=M ax C(ψ) ... (6) ψ Here, 2π C(ψ)-. (f, +1 μm)×(f2(θ”−ψ
)-μ2) aθ...(7) μr= o r+(dθ/2π...(8) μ2-
o″f2(θ)dθ/2ff·, (9).

Equations (6) to (9) mean that the rotation angle ψ is determined so that the cross-correlation between the density patterns of the registered image and the verification image is maximized. Note that instead of equations (6) to (9), the rotation angle ψ can also be determined using the following equation, which is simpler to calculate.

ψ= M ln D tF) −(](1ψ Here, 2π D(ψ) −. If, (#) − f2 (θ ten ψ) lcl
-C11).

In the rotation angle calculation method for pattern matching outlined above, errors tend to occur in the calculation of the rotation angle when a part of the input image is blurred or missing, as can be seen in a seal stamp or the like. The reason for this is that "density turn", which is a basic parameter for calculating the rotation angle, does not take into account correction due to partial omission of the input image.

[Purpose of the invention]

SUMMARY OF THE INVENTION An object of the present invention is to provide an image positioning method that can accurately superimpose images even when a part of an input image is blurred or missing.

[Summary of the invention]

The present invention achieves the above object by correcting (deleting or reducing) missing parts of an input image and generating a parameter ζ parameter for pattern matching.

The basic principle will be explained below.

FIG. 5 shows an example of an input image (verification image) in which a portion (broken line) is missing, and FIG. 6 shows the density pattern f(θ) regarding the rotation center 6. In FIG. Here, f(θ) is O≦
It is defined for each θ of θ≦2π.

Further, in FIGS. 5 and 6, the region α≦θ≦β represents the missing portion. The broken line in FIG. 6 represents the density pattern assuming that there is no missing part in the input image.

α and β, which indicate the boundaries of the region of the missing part, are calculated as follows. Now, consider a half-line 5(0) that forms an angle θ with the X-axis starting from the rotation center 6.

Among the black pixels existing above S (, pay attention to the black pixel with the largest distance from the center of rotation, and let the distance from the center of rotation be d(). Figure 7 shows d( for each θ). θ)
- Give an example. In the figure, the dotted line represents the value of d(tl) assuming that there is no missing part in the input image. As is clear from the case of seal stamps, missing input images usually include outline images. Therefore, d(θ) takes a small value at the angle α≦θ≦β corresponding to the missing portion of the input image. In other words, if the point symmetry of the contour image is used, if there is no missing part in the input image, -d(θ1π) of d( holds true, but conversely, the relationship #d(θ+π) of d( If there exists an angle θ such that θ does not hold, then the input image is missing at that θ.
The angles α and β indicating the region of the missing part can be easily set.

The area Q excluding this missing part is defined as follows.

Ω-(θ10≦θ≦2πL-(θ1α≦θ≦β) ・・
・The rotation angle ψ required to superimpose the matching image and the registered image is calculated using Q with 0. Then, the density patterns of the matching image and the registered image are expressed as f, (θ), f2(θ
), the rotation angle ψ required to superimpose both images is calculated using the following formula.

ψ-Max-path ψ)...a1 ψ Here ...Q4 It is.

In other words, Equations 01 to α0 determine the rotation angle ψ so that the range corresponding to the missing part of the image is excluded from the density patterns of the matching image and the registered image, and the cross-correlation between the two is maximized. It feels childish.

In addition, the rotation angle ψ can also be determined using the following equation, which is simpler to calculate, instead of equations 01 to 100.

ψ=MinD(ψ)...α ψ Here, It is.

Note that in Part 2, for convenience of explanation, we have dealt with the case where the missing part of the input image exists in only one area, but the same applies to the case where the missing part of the input image exists in multiple areas.

[Embodiments of the invention]

Hereinafter, the present invention will be explained in detail with reference to Examples.

FIG. 8 is a block diagram of an image matching system according to the present invention. In the figure, 10 is a processing device such as a microprocessor, 20 is a main memory, 30 is an image memory, 40 is a keyboard, 50 is an image input device, 60 is a file, and 70 is a display. Image take of registered image (binary image)
is input in advance using the image input device 50 and stored in the file 60 together with the extracted density pattern. To store management information of registered images,
The registered image management table 80 shown in FIG.
hold it above. This registered image management table 80 stores image names 81 for all registered images, storage address 82 in the image data file 60, storage address 8 in the density pattern file 60, (x, y) coordinates 84 of the rotation center, etc. This is a table where

The processing flow of the processing device 10 is shown in FIG.

The processing contents of each step in FIG. 10 are as follows.

(a) Process 110: The command to be executed and the name of the registered image to be processed are read from the keyboard 40.

(b) Process 12D: The input command is decoded, and if it is an end command, the process is ended and the operation of the device is stopped. If it is not an end command, the process advances to step 130.

(c) Process 160: Find the image data storage address, density pattern, and (x, y) coordinates of the center of rotation for the designated registered image from the registered image management table 80.

(d) Process 140: Read the image data and density pattern of the designated registered image from the file 60. The read image data is stored in the image memory 30, and the density pattern is stored in the main memory 20, respectively.

(e) Process 150: Read the verification image on the input document from the image input device 5o,
It is stored on the image memory 60. However, Noboru 81 [fii
Store in a separate area from the image storage area.

(f) Process 160: The coordinates (Xo, Yo) of the rotation center of the matching image are calculated using equations (2) ~
Calculate according to (5).

(g) Process 170: Calculate the gunshot pattern f(θ) of the matching image. f(θ)
The calculation method will be explained using diagram m11. In FIG. 11, the matching image is expressed as a set of lattice points (marked with circles) at regular intervals a. Point 210 indicates the center of rotation of the matching image. A straight line 2 passing through the center of rotation 210 and forming an angle θ with the X axis
20 is shown. Starting from point 210, draw straight line 220
The matching image is sampled at regular intervals a on the top, and the total error of aI at these sampling points (marked 0 in Fig. 11) is 'a degree pattern f'. Generally, the points closely match the grid points (○ marks) of the matching image, so it is necessary to associate the sampling points with the grid points.Here, the m-th ( It is assumed that the sampling points m=1.2...,M) are associated with the lattice points in the i-th row and j column defined by the following equation.

i = (((m-1)xΔX×.l/a) ・(
IIj=(((m-1)XΔy+yol/a) −H Taso, ΔX : a
X sin θ 98. (a), and the symbol [] means truncation.

Now, the density of the grid point in the first row and column j of the matching image is expressed by the symbol gl
When expressed as glj, glj takes a value of 0 or 1 since this embodiment deals with binary images. At this time, the density pattern f is calculated using the following equation.

Here, M represents the number of sampling points on the straight line 220.

(h) Process 180: The parameter d shown in FIG. 7 is calculated for the matching image, and based on this, the region 12 of equation (6) is determined.

(i) Process 190: Determine the rotation angle ψ of the matching image necessary to superimpose the registered image and the matching image. The calculation formula for ψ is Equation 03~
CO, t, formula (1η~4 is used.

(j) Process 200: Translate the matching image in parallel so that the rotation centers of the registered image and the matching image coincide, rotate by an angle ψ, and superimpose both images. As a result of the superposition, a portion of both images that becomes clear and a mismatched portion are distinguished and displayed on the display 70. Although one embodiment has been described above, the idea of the present invention can be realized by either software or nodeware.

In the above embodiments, for convenience of explanation, binary images have been described, but the principles of the present invention can also be applied to multivalued images or color images.

Furthermore, as shown in process 110 in FIG. 10, the method of specifying the name of the sharp registered image to be processed on the keyboard 40 has been described in the above embodiment. Unlike this, file 6
It is also possible to use a method in which the registered image, its density pattern, and the (x, y) coordinates of the center of rotation are sequentially read out from 0, and the superimposition with the verification image is repeated. In other words, file 6 the registered image with the highest matching stable with the matching image.
You can search from 0.

In the embodiment described above, a method of generating parameters for pattern matching by excluding missing portions of an input image was shown, but it is easy to think of various modified methods other than this. For example, a missing contour image can be restored by utilizing the point symmetry of the contour image, but in this case, the rotation angle may be calculated using a density pattern that includes the restored contour image.

The present invention also provides automatic format checking to determine whether a predetermined image exists at a predetermined position on a document, and format identification to identify a format based on the image butter 7 existing at a predetermined position on a document. It can also be applied to

〔Effect of the invention〕

As explained above, the present invention eliminates or restores missing parts of input images to generate parameters for pattern matching, so even if a part of an input image is blurred or missing, image overlay can be performed. It can be done with high precision.

[Brief explanation of the drawing]

FIG. 1 is a conceptual diagram of an image matching system to which the present invention is applied, FIGS. 2 and 6 are explanatory diagrams showing a method for calculating the coordinates of the center of rotation of an image in the present invention, and FIG. 4 is an illustration of [o! Concentration on a radial half line passing through the center of rotation); explanatory diagram showing a turn,
Figures 15 to 7 are explanatory diagrams showing an example of a verification image in which a part of the sealing material is missing according to the present invention, its density pattern destination, and parameter d (θ), and Figure 8 is an explanatory diagram showing an example of a collation image in which a part of the sealing material is missing according to the present invention, and its density pattern destination and parameter d (θ). A block diagram of an image matching system as an embodiment, FIG. 9 is a configuration diagram of a registered image management table according to the present invention, FIG. 10 is a processing flow diagram according to the present invention, and FIG. 11 is a density pattern calculation method according to the present invention. FIG. 10... Processing unit, 20... Main memory, 30...
Image memory, 40...Keyboard, 50...Picture input device i'z, 60...File, 70...Display. Figure 1 Figure 2 Figure 4 Layer B ?fl Figure 7 Figure 8 Figure 9 Figure 1θ Figure

Claims (1)

[Claims] 1. For two-dimensional image data having a rotationally symmetrical contour,
When a missing part of the image is detected based on the density distribution of pixels along a radial half-line passing through the rotation center position, multiple images are superimposed based on the data obtained by removing the missing part from the density distribution. An image matching method comprising the step of determining a rotation angle of an image required for image matching. 2. For two-dimensional image data with a rotationally symmetrical contour,
2. The image matching method according to claim 1, further comprising the step of deleting missing contour images.