JPH07220081A - Segmenting method for graphic of image recognizing device - Google Patents

Abstract

PURPOSE:To segment a two-dimensional code graphic with such constitution that at least adjacent two or four sides form straight lines or a graphic similar to that at high speed and accurately. CONSTITUTION:A two dimensional image is inputted in step S1, and edge detection is performed in step S2, and straight line sampling is performed by Hough transformation and least square approximation in step S3, and the combination of the straight lines is selected in step S4. In step S5, the size of the graphic is detected, and the sampling of remaining two straight lines are performed in step S6. and image recognition processing is performed in step S7, then, the coordinates of the four sides and four corners of a detection target graphic are decided. Processing is repeated by returning to the combination selection of the straight lines in step S4 until correct straight line detection or graphic recognition are performed in steps S5, S6. and S7.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method of cutting out a figure of an image recognition device such as a two-dimensional code reading device.

[0002]

2. Description of the Related Art An image recognition apparatus of this type reads an image containing a rectangular two-dimensional code figure in which data is arranged in a matrix having a configuration in which at least two sides or four sides are straight lines or a figure similar thereto. A two-dimensional code figure or a figure similar thereto is cut out from the read image and is recognized.

In such an image recognition apparatus, as shown in FIG.
As shown in (a), the position of the two-dimensional figure 2 to be cut out is detected from the read two-dimensional image 1. This position detection is found in, for example, Japanese Patent Publication No. 62-3475. The pattern matching method is often used.

This pattern matching method is shown in FIG.
A plurality of standard patterns 3, 3, ... As shown in (b) are used, the two-dimensional image 1 is scanned by this standard pattern 3, and the degree of similarity is expressed as a numerical value.
This is a method of cutting out a portion of the two-dimensional image 1 close to the numerical value and detecting the position of the target two-dimensional figure 2.

As for position detection, for example, Japanese Patent Publication No. 1-
A contour line edge tracking method as seen in Japanese Patent No. 35385 is also often used.

This contour line edge tracing method is a method for catching the feature of the contour of a figure, and as shown in FIG.
It traces and captures the outline as shown in.

[0007]

The above-mentioned pattern matching method is known as a very effective method for cutting out a target two-dimensional figure 2 from a two-dimensional image 1 having a certain characteristic. Therefore, a rectangular two-dimensional code figure in which data is arranged in a matrix having a configuration in which at least two sides are straight lines and a rectangular two-dimensional code figure in which data is arranged in a matrix configuration in which four sides are straight lines are provided. It is also effective when cutting out a figure having a characteristic as described above, but when an image is read, the two-dimensional code figure always exists in parallel to the two-dimensional image as shown in FIG. In most cases, the two-dimensional code figure 2 is read with an inclination with respect to the two-dimensional image 1 as shown in FIG.

For this reason, the pattern matching method is actually used.
In order to cut out the dimensional code figure 2, as shown in FIG. 14 (b), standard patterns 7 representing straight lines with various inclination angles,
.. are prepared and the two-dimensional image 1 is sequentially scanned to detect the position of the target two-dimensional figure 2, which causes a problem that it takes a lot of time to cut out the figure.

Further, unlike the pattern matching method, the above-described contour line edge tracing method has no problem in detecting the position of an inclined straight line, and searches only a characteristic portion in a two-dimensional image. Therefore, it takes a relatively short time to cut out a figure, but there is a problem that it is difficult to specify the starting point for contour edge tracking. Further, even if the starting point is successfully taken, if a figure with a scratch 8 is read in as shown in FIG. 15, the part of the scratch 8 is traced as a contour as shown by a dotted arrow 9 in the figure, and originally, There was a problem of recognizing a contour different from that of.

Therefore, according to the present invention, by using the Hough transform method and the least-squares approximation method for detecting the position of a straight line, a rectangular shape in which data is arranged in a matrix shape in which at least two sides or four sides are straight Provided is a method of cutting out a figure of an image recognition device capable of cutting out a two-dimensional code figure or a figure similar thereto at high speed and accurately.

[0011]

The invention according to claim 1 includes a rectangular two-dimensional code figure in which data is arranged in a matrix having a configuration in which at least two sides of the circumference are straight lines or a figure similar thereto. In an image recognition device for reading an image and recognizing a two-dimensional code figure or a figure similar to the image read out from the read image, a Hough transform method and a least-squares approximation are applied to the image of the two-dimensional code figure or a figure similar thereto. The position of the two straight lines that intersect each other by the method, the step of detecting the length of the two straight lines detected in this step, and the position of the two straight lines detected in each step. Based on the length, a step for detecting the positions of the remaining two straight lines that intersect each other is provided, and a two-dimensional code figure or a figure similar to it is cut out at each step. Lies in the fact.

According to the second aspect of the present invention, a rectangular 2 in which data is arranged in a matrix having a structure in which four sides are straight
An image recognition apparatus that reads a two-dimensional code figure or an image including a figure similar to the two-dimensional code figure and cuts out a two-dimensional code figure or a figure similar to the two-dimensional code figure from the read image and recognizes the two-dimensional code figure or a figure similar thereto. On the other hand, a step of detecting the positions of four surrounding straight lines by the Hough transform method and the least square approximation method is provided, and a two-dimensional code figure or a figure similar thereto is cut out by the step.

The invention according to claim 3 is the invention according to claim 1 or 2, wherein the step of detecting the position of the straight line comprises:
A certain straight line is detected by extracting some straight line candidates having priorities and sequentially combining the extracted straight line candidates in the order of higher priority.

[0014]

Embodiments of the present invention will be described below with reference to the drawings.

First Embodiment FIG. 1 is a diagram showing a two-dimensional image area 11 when a rectangular two-dimensional code figure is read and a two-dimensional code figure 12 to be cut out in the image area 11. The two-dimensional code figure 12 is a rectangular two-dimensional code figure in which data is arranged in a matrix having a configuration in which the two sides 12a and 12b on the periphery are straight lines. That is, the range of data is actually arranged up to the range indicated by the alternate long and short dash line in the figure.

2 which is a straight line of the two-dimensional code figure 12
As shown in FIG. 2, the edges 12a and 12b are horizontally scanned at regular intervals to extract edge points indicated by ◯ and Δ in the figure. Further, as shown in FIG. 3, the two sides 12a and 12b are scanned in the vertical direction at regular intervals to extract edge points indicated by ◯ and Δ in the figure.

The scanning in the horizontal direction and the vertical direction is not performed for all the pixels in the entire screen, but is performed at regular intervals in order to shorten the processing time.

By this scanning, the attribute of the rising edge (edge point indicated by ◯ in the figure) or the falling edge (edge point indicated by Δ in the figure) is determined.

When the edge point is detected in this way, a straight line is subsequently extracted by the Hough transform method. The Hough transform method is known as an algorithm for extracting straight lines from an image. That is, in the Hough transform method, the distance R from the origin of a straight line having an angle θ passing through arbitrary sampling points is calculated, and the distance R corresponding to the angle θ is calculated at all angles of all sampling points, and all the angles θ This is a method in which the straight line represented by the combination that has the largest number among the combinations of and the distance R is used as the extraction straight line.

When this Hough transform is performed by a computer, a two-dimensional array having θ and R as parameters is secured, and voting for the array element is performed from each sampling point. At this time, if θ is finely taken, the number of calculations increases, and a problem of processing speed arises. Further, if R is taken strictly, it becomes necessary to take a very large two-dimensional array. Therefore, in the present embodiment, a step of roughly taking θ and R and performing strict line extraction is provided later.

FIG. 4 is a diagram showing the result of straight line extraction after the Hough transform in FIG. 2, and FIG. 5 is a diagram showing the result of straight line extraction after the Hough transform in FIG. .

Since the Hough-transformed straight line 13 in FIG. 4 and the Hough-transformed straight line 14 in FIG. 5 roughly take θ and R, they are straight lines having angles and distances slightly different from the actual straight line to be detected. .

Therefore, the least squares approximation is carried out using each point supporting this straight line as a sampling point, and the precise straight line 15 shown in FIG. 4 and the precise straight line 16 shown in FIG. 5 are obtained by calculation.

By the way, when rough Hough transformation is performed, a straight line in which rising edges and falling edges are mixed may be detected by Hough transformation. In such a case, if the least-squares approximation is performed using all edge points supporting this straight line as sampling points, a straight line 18 different from the straight line 17 to be actually detected is detected as shown in FIG. Occurs.

Therefore, in this embodiment, after the Hough transform is performed by the horizontal scan and the vertical scan, as shown in FIG. 7, a combination of straight lines where the detected straight lines by the Hough transform intersect at about 90 degrees, that is, straight lines. 19 and 20 are selected, and the existing direction of the figure is determined by the distribution of the edge points supporting the straight lines 19 and 20 as shown by solid arrows 21 and 22 in FIG.

That is, the direction in which a lot of edge points are present around the intersection of two straight lines detected by the Hough transform is determined to determine the existing direction of the figure.

When the direction of existence of the figure is determined in this way, it becomes clear whether the straight line to be detected is a rising edge or a falling edge, and by performing the least-squares approximation by using only the edge points having the attribute corresponding thereto as sampling points. , Can detect the correct straight line.

Thus, as shown in FIG. 8, the straight line 2 corresponding to the two sides 12a and 12b around the two-dimensional code figure 12 is formed.
3,24 will be strictly determined.

Since the figure to be detected is a rectangular figure, a square 25 is formed at the position on the opposite side of the two straight lines 23 and 24.
Is present, and the end lines 26 and 27 can be recognized as straight lines by Hough transform.

When the squares 25 on opposite sides are not connected, two straight lines 2 obtained by the least-squares approximation method
3 and 24 are parallel to the positions facing each other, and a virtual end line is drawn by the end points 28 and 29, and the end line and the distance,
A straight line that has a similar angle and is detected by the Hough transform is an end line 2
6,27.

In the detection of the end lines 26 and 27 at this time, as in the first detection of the two straight lines 23 and 24, strict straight line extraction is performed by performing least square approximation from the Hough transform.

FIG. 9 is a flow chart showing the above processing. That is, a two-dimensional image is input in the first step S1, edge detection is performed in the next step S2, straight lines are extracted by Hough transform in the next step S3, and combinations of straight lines are selected in the next step S4. Then, the size of the figure is detected in the next step S5, the remaining two straight lines are extracted in the next step S6, and finally the image recognition processing is performed in step S7. In the Hough transform of step S3, least square approximation is also performed in order to detect an accurate straight line.

In steps S5 and S6, the size of the figure is detected and the remaining two straight lines are extracted. If the detection or extraction fails, the process returns to the straight line combination selection in step S4, and the straight line combination is again performed. Make a choice.

Then, the size of the figure is detected again in step S5, and the remaining two straight lines are extracted in step S6.

Further, the image recognition processing in step S7 is 4
The whole process performed after detecting the position and size of the straight line of the book is shown. When the position of the figure detected in the process is found to be incorrect, the process returns to the straight line combination selection in step S4, and the straight line combination is selected again. Do.

Then, the size of the figure is detected again in step S5, the remaining two straight lines are extracted in step S6, and then the image recognition processing in step S7 is performed.

In the image recognition processing of step S7, the coordinates of the four sides and the four corners of the figure to be detected are determined.

As described above, first, the read two-dimensional image area 11 is subjected to the horizontal scanning and the vertical scanning at a constant interval so that the two sides 12a, 12 of the two-dimensional code figure 12 are scanned.
The rising edge point and the falling edge point for b are extracted, and the attribute of each edge point is determined.

Then, a rough Hough transform is performed to extract straight lines, and then a combination of straight lines intersecting the extracted straight lines at about 90 degrees is selected, and the existence of the figure is determined by the distribution of edge points supporting the straight lines. Determine the direction.

After the rising and falling edges of the straight line to be detected are clarified, the least square approximation is performed using only the edge points having the attributes on the detected straight line as sampling points to detect the correct straight line.

Next, by utilizing the fact that it is a rectangular figure from the two detected straight lines, the two straight lines on the opposite side are recognized by the Hough transform. At this time as well, the correct straight line is detected by performing the least-squares approximation from the Hough transform.

In this way, the four sides and the four corners of the figure to be detected are determined, and the two-dimensional code figure 12 is cut out from the two-dimensional image 11.

After the rough Hough transform is performed in this way, a combination of straight lines intersecting at about 90 degrees is selected, the direction of existence of the figure is determined, and then a correct straight line is detected by least square approximation. Since the two-dimensional code figure is taken, the processing time is short and the two-dimensional code figure can be cut out at high speed, and the two-dimensional code figure can be cut out accurately even if there is a scratch or the like.

Second Embodiment This is a two-dimensional image area 31 read as shown in FIG.
A rectangular 2 in which data is arranged in a matrix in which the four sides 32a, 32b, 32c, 32d of the periphery are straight lines.
The cutout of a figure when the dimension code figure 32 is present will be described.

This two-dimensional code figure 32 has two sides 32a,
32b is not only connected as straight lines 33 and 34, but is also opposite to the two sides 32a and 32b.
32d is also connected as straight lines 35 and 36. That is, the two-dimensional code figure 32 has straight lines 33 on all four sides.
It is a figure surrounded by ~ 36.

In the first embodiment, after extracting two straight lines, the end line of the opposite side is determined based on the size and end point of the figure. This is because since the opposite side is not a continuous line of black pixels, the end line obtained from it is less characteristic as a straight line than a continuous line, and it is difficult to correctly detect the end line from the entire image.

However, in this embodiment, two straight lines 3
3, 34 and the two end lines 35, 36 both have the characteristic of being strong straight lines.

Therefore, in the present embodiment, when cutting out a figure, as shown in the flow chart in FIG. 11, when edge detection is performed in step S2, all four sides of the figure to be detected are straightened by Hough transformation in step S13. Can be detected by extraction.

As in the first embodiment, the final correct straight line extraction is step S14 after the straight line is extracted by the Hough transform.
At step S7, a combination of straight line candidates having a plurality of priorities is selected by using the least squares approximation method, and step S14 is repeated until correct image recognition is performed at step S7, and finally step S7. Correct image recognition will be performed at.

In this embodiment, the steps of detecting the size of the figure and extracting the remaining two straight lines are not necessary as compared with the first embodiment, so the processing amount is reduced and the figure cutting out speed is further increased. be able to. Of course, also in this embodiment, as in the first embodiment, it is possible to accurately cut out a two-dimensional code figure even with respect to a scratch or the like.

In each of the above embodiments, the cutting out of a rectangular two-dimensional code figure in which data is arranged in a matrix having a structure in which two or four sides are straight is described, but the present invention is not necessarily limited to this. Of course, the present invention can also be applied to the cutting out of a graphic having a configuration in which two or four sides around it are similar to a two-dimensional code graphic.

[0052]

As described above, according to the present invention, by using the Hough transform method and the least-squares approximation method for detecting the position of a straight line, data can be formed in a matrix with a configuration in which at least two sides or four sides are straight lines. It is possible to provide a method of cutting out a figure of an image recognition device capable of cutting out a rectangular two-dimensional code figure or a figure similar thereto in a fast and accurate manner.

[Brief description of drawings]

FIG. 1 is a diagram showing a two-dimensional image area and a two-dimensional code figure according to a first embodiment of the present invention.

FIG. 2 is a diagram for explaining edge point extraction by horizontal scanning according to the embodiment.

FIG. 3 is a diagram for explaining edge point extraction by vertical scanning according to the embodiment.

FIG. 4 is a diagram for explaining straight line extraction by Hough transform and least square approximation based on edge points extracted by horizontal scanning in the same embodiment.

FIG. 5 is a diagram for explaining straight line extraction by Hough transform and least square approximation based on edge points extracted by vertical scanning in the same embodiment.

FIG. 6 is a diagram for explaining straight line extraction by erroneous Hough transform.

FIG. 7 is a diagram for explaining detection of a figure existing direction of the embodiment.

FIG. 8 is a diagram for explaining extraction of two remaining straight lines in the embodiment.

FIG. 9 is a flowchart showing a graphic cutout process of the embodiment.

FIG. 10 is a diagram showing a two-dimensional image area and a two-dimensional code figure according to the second embodiment of the present invention.

FIG. 11 is a flowchart showing a graphic cutout process of the embodiment.

FIG. 12 is a diagram for explaining graphic cutout in the first conventional example.

FIG. 13 is a diagram for explaining graphic cutout in the second conventional example.

FIG. 14 is a diagram for explaining the problem of the first conventional example.

FIG. 15 is a diagram for explaining the problem of the second conventional example.

[Explanation of symbols]

 11 ... Two-dimensional image area 12 ... Two-dimensional code figure

Claims (3)

[Claims] 1. A rectangular two-dimensional code figure in which data is arranged in a matrix having a configuration in which at least two sides are straight lines, or an image including a figure similar thereto is read, and the two-dimensional code figure is read from the read image. Alternatively, in an image recognition device which recognizes a figure similar to it by slicing the two-dimensional code figure or an image of a figure similar to the two-dimensional code figure, the two straight lines intersecting each other by the Hough transform method and the least-squares approximation method. The step of detecting the position, the step of detecting the lengths of the two straight lines detected in this step, and the rest of the surroundings that intersect each other based on the positions and the lengths of the two straight lines detected in each of the steps An image recognition characterized by providing a step of detecting the positions of two straight lines, and cutting out the two-dimensional code figure or a figure similar thereto by each of the steps. A method for cutting out a graphic of a recognition device. 2. A rectangular two-dimensional code figure in which data is arranged in a matrix having a structure in which four sides are straight lines or an image including a figure similar thereto is read, and the two-dimensional code figure or the image is read from the read image. In an image recognition device which cuts out and recognizes a graphic similar thereto, the positions of four surrounding straight lines are detected by the Hough transform method and the least-squares approximation method for the image of the two-dimensional code graphic or a graphic similar thereto. A step is provided, and the step 2
A method for cutting out a figure of an image recognition apparatus, which is characterized by cutting out a dimensional code figure or a figure similar thereto. 3. The step of detecting the position of a straight line extracts some straight line candidates having priorities and sequentially combines the extracted straight line candidates in descending order of priority to obtain a correct straight line. 3. The method of cutting out a figure of the image recognition device according to claim 1, wherein the figure is detected.