JPH0154751B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to an image processing device, and more particularly to an image-to-image rotation angle difference detection device for determining a rotation angle difference between two congruent or similar images.

For example, in a seal matching device, which is an applied field of image processing technology, a method such as pattern matching is used to match a matching seal impression, that is, a matching image, and a registered seal impression, that is, a reference image that is used as a reference for matching. A process of comparing the two and checking the degree of similarity between the two is often used.

However, if the angle or orientation of the image to be matched is different from that of the reference image, matching is difficult as is.

Therefore, an inter-image rotation angle difference detection device is used to detect the rotation angle difference between two images and remove the rotation angle difference between the two images.

At this time, it is desirable that the rotation angle difference between the two images be detected accurately and in a short time.

[Conventional technology]

FIG. 4 is a block diagram showing the configuration of a conventional example of an image rotation angle detection device, and 1 and 2 are block diagrams showing image data of a reference image and a matched image, respectively, in the x-axis direction and the y-axis direction on orthogonal xy coordinates. A first image memory and a second image memory store the density of each pixel arranged at equal pitches.

3a and 3b and 4a and 4b are a first angular histogram generating unit 3 and a second angular histogram generating unit 3 that generate angular histograms in the main change direction of density for each pixel in the image data of the reference image and the matched image, respectively. It is a component of the generation unit 4, and 3
a and 4a scan the image data stored in the first image memory 1 and the second image memory 2, respectively, using, for example, a 3×3 differential calculation window, and calculate the difference dx in the density of each pixel in the x direction. and the difference dy in the y direction, and using θ=arctan(dy/dx), the first angle calculation unit and the second The angle calculation units 3b and 4b count the number of pixels for each angle θ in the main change direction of density obtained by the first angle calculation unit 3a and the second angle calculation unit 4a, respectively, and calculate the number of pixels to 0 degrees. They are a first histogram memory and a second histogram memory that store angles from 359 degrees to 359 degrees as addresses.

5 are first histogram memories 3b, respectively.
and an angular difference detection unit that detects a rotation angle difference between two images based on the angle histogram stored in the second histogram memory 4b.

6, the image to be matched stored in the second image memory 2 is calculated based on the rotation angle difference θ detected by the angle difference detection unit 5, using a table, for example, x′=x cosθ+y sinθ y′= An image rotation unit rotates by performing a conversion of −y sin θ+y cos θ, and 7 is a third image memory that stores the image rotated by the image rotation unit 6.

Reference numerals 8 and 9 denote a first projection unit and a second projection unit that create a projection onto the x-axis and a projection onto the y-axis of the image data stored in the first image memory 1 and the third image memory 7, respectively. and 1
0 and 11 are a first projection memory and a second projection memory that store the x-axis projection and y-axis projection obtained by the first projection section 8 and the second projection section 9, respectively.

Further, 12 verifies the rotational angle difference detected by the angular difference detection unit 5 based on the distance between the projection stored in the first projection memory 10 and the projection stored in the second projection memory 11. This is the angle difference verification section.

Next, the fifth image is used as the reference image and the image to be matched.
Congruent images with a rotation angle difference of 45 degrees as illustrated in Figures a and b are stored in the first image memory 1, respectively.
To explain the case where the images are stored in the second image memory 2, first, the first angular histogram generating section 3 and the second angular histogram generating section 4
As a result, angle histograms as shown in FIGS. 5(c) and 5(d) are obtained, respectively.

The angular difference detection unit 5 converts the angular histograms in FIG. 5 c and d in the angular axis direction (lateral direction in the figure).
By comparing the images in Figure 5 a and b (with respect to the reference image, counterclockwise rotation is positive and clockwise rotation is negative), the rotation angle difference between the images in Figure 5 a and b is determined to be 45 degrees or -135 degrees. To detect.

FIGS. 6a and 6b are images of FIG. 5b rotated by -45 degrees and 135 degrees, respectively, and their respective angular histograms exactly match those of FIG. 5c.

That is, the rotation angle difference between two images cannot be determined using only the angle histogram.

Therefore, the image rotation unit 6 rotates the image to be matched by -
The angle difference verification unit 1 calculates the projection obtained by the second projection unit 9 from the image rotated by 45 degrees or 135 degrees and the projection obtained by the first projection unit 8 from the reference image.
This is verified using 2.

The projection shown in Figure 7a is obtained from both the image to be matched rotated by -45 degrees and the reference image, and the projection shown in Figure 7b is obtained from the image to be matched by rotated 135 degrees. As a result, it has been verified that the rotation angle difference between the reference image and the matched image is 45 degrees.

In the case of the images illustrated in FIGS. 5a and 5b, two rotational angle differences of 45 degrees and -135 degrees with a difference of 180 degrees were obtained by the angle difference detection unit 5. In cases such as seal verification, especially when dealing with kanji, four types of rotation angle differences, each with a difference of 90 degrees, are often detected.

[Problem that the invention seeks to solve]

In the inter-image rotation angle difference detection device having the above configuration, for example, four types of rotation angle differences may be detected as described above, and in such a case, it is necessary to perform the verification four times, which takes a long time. The problem was that it took time.

[Means for solving problems]

The inter-image rotation angle difference detection device according to the present invention includes an angular histogram generation unit that generates an angular histogram by counting, for each angle, the main direction of change in density at each pixel in image data expressed as the density of each pixel. , an angular difference detection unit that detects a rotation angle difference between two images based on the generated angle histogram; and an angular difference detection unit that detects at least one of the two image data so as to remove the detected rotation angle difference. an image rotation unit that rotates; a shape histogram generation unit that generates a shape histogram by counting the number of occurrences of each of a plurality of predetermined shape patterns in each partial region in the image data; The above-mentioned problem is solved by including an angular difference verification section that verifies the detected rotational angle difference using a shape histogram.

[Effect]

That is, in the present invention, the rotation angle difference detected by the angle difference detection unit is verified by comparing the shape histograms generated from each of the two images from which the rotation angle difference has been removed. For example, even if four types of rotation angle differences having a difference of 90 degrees are detected, the correct rotation angle difference can be determined just by comparing one of these types.

〔Example〕

The gist of the present invention will be specifically explained below using Examples.

FIG. 1 is a block diagram showing the configuration of an embodiment of an image-to-image rotation angle difference detecting device according to the present invention, and the same reference numerals as those in the conventional example in FIG. 4 refer to the same objects.

Others 13a and 13b and 14a and 14b
are 3×3 in the image data stored in the first image memory 1 and the third image memory 7, respectively.
For each partial area of a pixel, for example, as shown in Figure 2,
Calculate the number of times each of the 12 shape patterns appears,
These are components of a first shape histogram generation section 13 and a second shape histogram generation section 14 that count and generate a shape histogram for each pattern, and 13a and 14a are components of the first image memory 1 and the third image memory 1, respectively. A first pattern extraction unit that scans the image data stored in the image memory 7 using a 3×3 pattern detection window and extracts patterns of 12 types of shapes illustrated in FIG. and a second pattern extraction section.

13b and 14b are the first pattern extraction section 13a and the second pattern extraction section 14, respectively.
They are a third histogram memory and a fourth histogram memory that count the number of each of the 12 types of shape patterns extracted by a and store it for each address from 1 to 12.

15 is the shape histogram stored in the third projection memory 13b and the fourth histogram memory 1
This is an angular difference verification section that verifies the rotation angle difference detected by the angular difference detection section 5 by comparing it with the shape histogram stored in the angular difference detection section 4b.

The 12 types of patterns illustrated in Figure 2 are basically three types: , and , and for each, -90 degree rotation is given to and , and +180 degree rotation is given to each of them. , and , and also gives +90 degrees of rotation.

Hereinafter, the first image memory 1 stores an image as shown in FIG. To explain the case where
First, the angle difference detection unit 5 detects α degrees・(α−90)
Four types of rotation angle differences are detected: degree, ((α+180) degree, or (α+90) degree), and images such as those shown in FIG. 3 a, b, c, and d are obtained for each by the image rotation unit 6. .

In the first shape histogram generation unit 13, the third
A shape histogram as shown in FIG. e is generated, and a shape histogram as shown in FIG.

For example, if the rotation angle difference detected by the angle difference detection unit 5 is α degree, the first shape histogram generation unit 13 obtains a shape histogram as shown in FIG. A shape histogram of the same shape is obtained in the histogram product 14, and by comparing these, the angle difference verification unit 15 determines the rotation angle difference α detected by the angle difference detection unit 5 as correct. do.

Further, if the rotation angle difference detected by the angle difference detection unit 5 is, for example, (α-90) degrees, the angle difference verification unit 15 calculates the shape as shown in FIG. 3e obtained from the reference image. By comparing the histogram with a shape histogram as shown in FIG. It is verified that there is an error of 90 degrees, so the correct rotation angle difference is determined as α+90 degrees.

In the above embodiment, the shape histogram is obtained by counting the frequency of appearance of patterns of 3×4 types for each partial region of 3×3 pixels, but the size of the partial region and the pattern The numbers are not limited to those listed above.

〔Effect of the invention〕

As explained above, according to the present invention, it is not necessary to verify all four types of rotational angle differences obtained by the angle difference detection unit, each having a difference of 90 degrees, so that verification can be performed in a short time. Verification can be performed.

[Brief explanation of drawings]

FIG. 1 is a block diagram of an embodiment of the present invention, FIGS. 2 to 3 a to h are explanatory diagrams of the same embodiment,
FIG. 4 is a block diagram of the conventional example, and FIGS. 5 a to d, 6 a, b, and 7 a, b are explanatory diagrams of the conventional example. In the figure, 3a is a first angle calculation section, 4a is a second angle calculation section, 3b is a first histogram memory,
4b is a second histogram memory, 5 is an angular difference detection unit, 6 is an image rotation unit, 13a is a first pattern extraction unit, 14a is a second pattern extraction unit, 13a
14b is a first histogram memory, 14b is a second histogram memory, and 15 is an angular difference verification unit.

Claims (1)

[Claims] 1. An angular histogram generation unit that generates an angular histogram by counting the main direction of change in density at each pixel in image data expressed as the density of each pixel for each angle, and two an angular difference detection unit that detects a rotation angle difference between images; an image rotation unit that rotates at least one of the two image data so as to remove the detected rotation angle difference; a shape histogram generation unit that generates a shape histogram by counting the number of patterns each of a plurality of predetermined shapes appears in all partial regions for each pattern; An image-to-image rotation angle difference detection device comprising: an angle difference verification section that verifies an angle difference.