JPH069060B2 - Image registration device - Google Patents

Description

DETAILED DESCRIPTION [Overview] In an image registration device, a rotation angle difference candidate between two images to be registered is obtained by comparing angle histograms with a short processing time, and processing is performed based on this candidate. Then, the amount of alignment between the two images is extracted from the amount of shift of the center of gravity of each of the images. As a result, the processing time can be greatly reduced and the alignment amount can be obtained with good accuracy.

[Industrial application field]

The present invention relates to an image registration device for performing registration between a plurality of images, and more particularly, to a process for extracting a translational and rotational angle difference between both images from the amount of deviation of the center of gravity of each image for alignment. The present invention relates to an image registration device improved to perform at high speed.

When the reference image registered in advance and the input image are collated as in the case of performing the seal collation, both images are aligned in order to perform the collation without error. Then, in order to perform the image registration at high speed and accurately, various methods for automatically performing the image registration processing have been proposed.

As the method, the rotation angle difference between the two images to be aligned is obtained from the angle histogram, the rotation angles of the two images are adjusted, and then the difference between the projections of the two images on the x and y coordinate axes is calculated. There is a method of obtaining the parallel movement amount of the two images and a method of obtaining the parallel movement amount and the rotation angle difference between the two images from the shift amount of the center of gravity for each image to be aligned.

In this method, if the rotation angle cannot be accurately obtained from the angle histogram, some rotation angle difference candidates must be obtained and the parallel movement amount must be obtained for each of them, but There is a problem in that it takes a lot of time to obtain the optimum amount of parallel movement, because it takes a lot of time to perform the process of obtaining. The method of (1) has a problem that it takes a lot of time for processing because the center of gravity for each angle is rotated for all angles as described in more detail below.

The present invention relates to the method (1) and is improved so as to speed up the processing.

[Conventional technology]

FIG. 3 shows an image registration method based on the above method proposed by the same applicant (Japanese Patent Laid-Open No. 60-11804).
0) is shown in a block diagram, and FIG.
FIG.

In FIG. 4, (a) is a reference image and (b) is a collation image, both of which are linear images having a certain width.

The collation image data (FIG. 4 (b)) stored in the collation image memory 210 is differentiated by the spatial differentiating circuit 220, and the density gradient and its direction in each pixel are obtained.

The center-of-gravity extraction circuit 230 classifies the respective density gradient directions according to the direction as shown in FIG. In FIG. 4 (d), the centroids of the six constituent pixel groups classified by direction are indicated by circles. The coordinate data of each centroid obtained in this manner is used as the collation image centroid storage register 240.
Stored in.

Similarly, the reference image is also classified into the direction-dependent constituent pixel groups of the density gradient as shown in FIG. 4 (c), and the center-of-gravity coordinate data thereof is stored in the reference-image center-of-gravity storage register 250. In FIG. 4 (c), the barycenters of the six constituent pixel groups classified by direction are indicated by crosses.

The center-of-gravity rotation circuit 260 uses the matching image center-of-gravity storage register 24.
The collation image read by 0 is rotated as shown in FIG. 4 (e). The rotation step is performed, for example, once.

The center-of-gravity position comparison circuit 270 compares the respective barycentric coordinates of the rotated collation image with the barycentric coordinates of the reference image of the reference image barycentric storage eddy 250. When the comparison of one angle is completed, the rotation angle of the centroid rotation circuit 260 is increased by one step (1 degree), and the angular centroid coordinates of both images are compared at that rotation angle.

By this comparison process, as shown in FIG. 4 (f), x and y
A rotation angle at which the values of the barycentric positions of the pixel groups for each direction of the reference image and the collation image best match at positions separated by a certain distance indicated by dx and dy in the coordinate direction is obtained. The center-of-gravity position comparison circuit 270 outputs the rotation angle θopt and the parallel movement amounts dxopt and dyopt at this time as the optimum rotation angle and the parallel movement amount.

The coordinate conversion circuit 280 performs coordinate conversion of the collation image read out from the collation image memory 210 based on θopt, dxot, and dyopt input from the center-of-gravity position comparison circuit 270, and the data of this conversion collation image is output to the output image memory 2
Store in 90.

Through the above processing, it is possible to obtain a collation image that is aligned so as to best match the reference image.

[Problems to be solved by the invention]

The method for obtaining the translation amount and the rotation angle difference between both images based on the deviation amount of the angle-dependent center of gravity of both images to be aligned as described in the previous section is a step in which the angle-dependent center of gravity of the collation image is set between 0 and 359 ° in small steps. The position of the center of gravity for each angle of the reference image and the collation image is compared for each angular rotation angle. Therefore, it takes time to rotate the center of gravity for each angle.
There has been a problem that it takes a lot of time until the entire processing up to 9 ° is completed.

The present invention is an image registration method that is improved so that the processing for obtaining the parallel movement amount and the rotation angle difference between both images from the shift amount of the center of gravity for each of the images for which the registration is performed can be performed extremely faster than before. It is intended to provide a device.

[Means for solving problems]

Means taken by the present invention to solve the above-mentioned problems in the conventional method will be described with reference to FIG.

FIG. 1 is a block diagram showing the basic configuration of the present invention.

In FIG. 1, reference numeral 110 is a reference image angle histogram extracting means for extracting an angle histogram of a reference image which is a reference image when performing alignment. Generally, in a grayscale image, a change in density value in the vicinity of each pixel can be detected as a density gradient. This feature is obtained as the magnitude direction in which the density value changes for each pixel. An angle histogram is obtained by counting the number of pixels for each direction.

Reference numeral 120 is a reference image angle-based centroid extraction unit that extracts the angle-based centroids of the reference image. The angle-based centroid is the centroid of each pixel group classified by direction.

Reference numeral 130 is a collation image angle histogram extraction means for extracting an angle histogram of the collation image which is an image for collation for position alignment.

Reference numeral 140 is a collation image-by-angle centroid extraction unit that extracts the centroid by angle of the collation image.

A rotation angle difference candidate selection unit 150 compares angle histograms of the reference image and the collation image and selects candidates for the rotation angle difference between both images.

Reference numeral 160 denotes a registration amount extraction means, which accurately extracts a translation amount and a rotation angle difference between the collation image and the reference image, which are processed based on the selected rotation angle difference candidate, from the shift amount of the center of gravity between the images. To do.

[Work]

The rotation angle difference candidate selection means 150 compares the reference image angle histogram input from the reference image angle histogram extraction means 110 with the matching image angle histogram input from the matching image angle histogram extraction means 130 to determine the rotation angle difference. Select a candidate.

This makes it possible to obtain a rotation angle difference candidate that gives a rough rotation angle difference between the reference image and the matching image in a short processing time.

The alignment amount extraction means 160 accurately obtains the translation amount and the rotation angle difference between both images based on the selected rotation angle difference candidate from the shift amount of the center of gravity for each angle between the collation image and the reference image.

As described above, according to the present invention, the rotation angle difference between the reference image and the collation image and the parallel movement amount can be accurately obtained, but the processing takes a lot of time. (Processing by the alignment amount extraction means 160) and extraction processing of alignment parameters (rotational angle difference candidate selection means 150) by an angle histogram that does not require a rough rotation angle difference between both images but the processing does not take time.
Processing), the processing time can be significantly shortened compared to both the conventional method and the method, and the translation amount and rotation angle difference between the reference image and the collation image can be obtained with good accuracy. Can be asked. Therefore, accurate alignment of both images can be performed.

〔Example〕

An embodiment of the present invention will be described with reference to FIG. FIG. 2 is a block diagram showing the configuration of an embodiment of the present invention.

(A) Configuration of the embodiment In FIG. 2, reference image angle histogram extraction means 1
10, the reference image angle-based centroid extraction unit 120, the collation image angle histogram extraction unit 130, the collation image degree-based centroid extraction unit 140, the rotation angle difference candidate selection unit 150, and the alignment amount extraction unit 160 will be described with reference to FIG. As I did.

In the reference image angle histogram extraction means 110, 1
Reference numeral 11 is a reference image angle histogram extraction unit that extracts the angle histogram of the reference image. A reference image angle histogram register 112 stores the extracted reference image angle histogram.

In the reference image angle-based centroid extraction means 120, reference numeral 121 is a reference image angle centroid extraction unit that extracts the angle-based centroids of the reference image. Reference numeral 122 is a centroid register for each reference image angle, in which the extracted centroids for each reference image angle are stored.

In the collation image angle histogram extraction means 130, 1
A reference image angle histogram extraction unit 31 extracts the angle histogram of the reference image. A reference image angle histogram register 132 stores the extracted reference image angle histogram.

In the collation image angle-based centroid extraction unit 140, 141 is a collation image angle-centroid extraction unit that extracts the angle-specific centroids of the collation image. 142 is a centroid register for each collation image angle,
The extracted centroids by collation image angle are stored.

In the rotation angle difference candidate selection means 150, a rotation angle difference candidate selection unit 151 compares the angle histograms of the reference image and the collation image, and selects candidates for the rotation angle difference between both images. A rotation angle difference candidate register 152 stores the selected rotation angle difference. A plurality of rotation angle difference candidates are generally selected.

In the alignment amount extraction means 160, 161 is a center-of-gravity rotation unit, which is extracted by the matching image angle-based center-of-gravity extraction means 140 based on the rotation angle difference of each rotation angle difference candidate selected by the rotation angle difference candidate selection means 150. The center of gravity for each angle of the collated image is rotated. Reference numeral 162 denotes a rotation angle-based center of gravity register that stores the angle-based center of gravity of the rotated collation image. Reference numeral 163 denotes a parallel movement amount calculation unit, which shifts the angle-based centroid of the reference image read from the reference-image-angle-based gravity center register 122 and the rotation-angle-based centroid of the matching image read from the matching image-angle-based gravity center register 162. From this, a parallel displacement amount for each angle and a dispersion value indicating the degree of its dispersion are obtained. This processing is performed for each rotation angle center of gravity of each rotation angle difference candidate, and the parallel movement amount and the rotation angle difference of the matching image that has become the rotation angle difference candidate with the smallest variance value among them are set to the optimum translation amount and Output as a rotation angle difference.

Reference numeral 170 is a reference image memory in which reference image data is stored, and 180 is a reference image memory in which reference image data is stored.

(B) Operation of the embodiment Among the plurality of images to be aligned, the data of the reference image to be the reference image is stored in the reference image memory 170, and the data of the collation image to be the image for collation is the collation image memory. Stored in 180.

The reference image angle histogram extraction unit 111 extracts the angle histogram of the reference image from the data of the reference image read from the reference image memory 170 and stores it in the reference image angle histogram 112.

The reference image angle-based centroid extraction unit 121 includes the reference image memory 1
The centroid by angle of the reference image is extracted from the data of the reference image read from 70, and the centroid register by angle of the reference image 1
It stores in 22.

Similarly, the collation image angle histogram extraction unit 131 and the collation image angle-based centroid extraction unit 141
The angle histogram and the centroid by angle of the collation image are extracted from the data of the collation image read out from 0 and stored in the collation image angle histogram register 132 and the centroid register by angle of the collation image 142, respectively.

The rotation angle difference candidate selection unit 151 compares the angle histograms of the reference image and the matching image, selects a candidate for the rotation angle difference between the two images, and selects the rotation angle difference candidate register 15
Store in 2.

The angle histogram of the angle of one image expresses the change of the density value in the image regardless of the position information. When the image is rotated, the changing direction of the density value also changes according to the rotation angle. That is, when a plurality of images are a set of images obtained by rotating the same image, the angle histogram obtained therefrom has a relationship in which the images are translated on the angle axis.

Conversely, if the angle histogram of one image (reference image) is used as a reference and the angle histograms of other images (matching images) are translated in parallel on the angle axis, the distance between both angle histograms can be calculated. The amount of angular movement that minimizes the distance gives the rotational angle difference between both images.

The rotation angle difference candidate selection unit 151 selects the rotation angle difference candidate between the reference image and the collation image based on the amount of angular movement that minimizes the distance between the angle histograms based on the above-described principle. A plurality of rotation angle difference candidates are selected in order from the one having the smallest distance between angle histograms. By selecting a plurality in this way, it is possible to suppress a decrease in calculation accuracy due to noise.

The center-of-gravity rotation unit 161 determines, based on the rotation angle difference of each rotation angle difference candidate selected by the rotation angle difference candidate selection means 150,
The centroids by angle of the collation image extracted by the centroid extraction unit by collation image angle 140 are rotated, and the results are stored in the centroid register by rotation angle 162.

The parallel movement amount calculation unit 163 calculates the deviation between the centroids by angle of the reference image read from the centroid register 122 by reference image angle and the centroids by rotation angle of the collation image read by the centroid register by collation image 162. The amount of parallel movement for each angle and its variance are calculated. This processing is performed for each rotation angle center of gravity of each rotation angle difference candidate, and the parallel movement amount and the rotation angle difference of the matching image that has become the rotation angle difference candidate with the smallest variance value among them are set to the optimum translation amount and Output as a rotation angle difference.

The shape obtained by compressing the shape of one image into only position information on the image for each angle is the center of gravity for each angle. Therefore, when a plurality of images are collation of images obtained by translating the same image, two images among them are collated. The amount of parallel movement of the center of gravity of each of the (reference image and the collation image) for each direction is constant regardless of the direction, and at this time, the variance value is minimum. On the contrary, when the parallel movement amount differs in each direction and the variance value is large, the rotation angles of both images are not equal. Therefore, the one with the smallest variance value of the parallel movement amount becomes the optimum parallel movement amount, and the rotation angle difference at that time becomes the optimum rotation angle difference.

If the collation image stored in the collation image memory 180 is subjected to coordinate conversion based on the obtained optimum translation amount and rotation angle difference, it is possible to obtain the collation image most aligned with the reference image.

Although one embodiment of the present invention has been described above, each configuration of the present invention is not limited to each configuration of this embodiment. For example, the rotation angle difference candidate includes one case. The optimum translation amount and rotation angle difference are not limited to those using the dispersion value.

〔The invention's effect〕

As described above, according to the present invention, the following various effects can be obtained.

(B) It is possible to obtain the parallel movement amount and the rotation angle difference between the reference image and the collation image with a processing time extremely shorter than that of the conventional image registration method and with good accuracy.

(B) The two images can be accurately aligned, and both images can be accurately matched.

[Brief description of drawings]

 1 ... Explanatory diagram of basic configuration of the present invention, FIG. 2 ... Explanatory diagram of configuration of one embodiment of the present invention, FIG. 3 ... Explanatory diagram of configuration of prior invention, FIG. 4 ... Inventive invention In FIGS. 1 and 2, 110 ... Reference image angle histogram extraction means, 120 ... Reference image angle-specific centroid extraction means, 130 ... Collation image angle histogram extraction means, 140 ... Collation image angle-specific centroid extraction means Means, 150 ... Rotation angle difference candidate selection means, 160 ... Positioning amount extraction means,

Claims (1)

[Claims] 1. An image registration apparatus for performing registration between a plurality of images, comprising: (a) reference image angle histogram extraction means for extracting an angle histogram of a reference image that serves as a reference image when performing registration. 110), and (b) Reference image angle-specific centroid extraction means for extracting the angle-specific centroid that is the centroid of each pixel group classified by the reference image direction.
(120), (c) collation image angle histogram extraction means (130) for extracting the angle histogram of the collation image that becomes the collation image when performing the alignment, and (d) classified by the direction of the collation image. Collation image Centroid extraction means by angle for extracting centroids by angle which is the centroid of each pixel group
(140), (e) Rotation angle difference candidate selection means (150) for comparing the angle histograms of the reference image and the matching image and selecting candidates for the rotation angle difference between both images, and (f) selection Positioning amount extraction means (160) for extracting a translation amount and a rotation angle difference between both images from the displacement amount of the centroids by angle of the collation image and the reference image processed based on the candidate of the rotation angle difference, An image alignment device characterized by the above.