JPS6380373A - Pattern recognition device - Google Patents

Abstract

PURPOSE:To recognize a specific pattern in a short time by designating specific picture elements of the pattern and subjecting only specified picture elements to logical operation and recognizing the pattern with a small number of operations. CONSTITUTION:Information of a reference picture element A and featuring picture elements B, C, and E is inputted to define picture elements A, B, C, and E as specific picture elements of a pattern. Coordinates (X, Y) of specific picture elements to the reference picture element and their values P are set to the value of the picture element A. When the heading offset value of picture data is defined as SOFF, a heading offset value ROFF of objective picture data is obtained in accordance with an equation I where XH indicates the size in the horizontal direction. Next, feature points B and C are subjected to the operation in accordance with a formula II with respect to picture data, and results and the feature point E are subjected to the operation in accordance with a formula III, and results are transferred to an output area. These operations are executed in all areas. Thus, the pattern is satisfactorily recognized with a small number of operations.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a pattern recognition device that recognizes a specific pattern from image data.

[Conventional technology]

A conventional pattern recognition device of this type is shown in FIG. Furthermore, FIG. 8 shows the results of the conventional pattern recognition device shown in FIG.
5. It is a figure which shows the pattern recognition method called template matching shown in pp107-111. 7th
In the figure, 1 is a template memory that stores template data, 2 is a storage device that stores image data of the drawing, and 3 is an arithmetic device that measures the degree of coincidence between the data stored in the template memory 1 and the storage device 2. .

Next, the operation will be explained. First, when recognizing the pattern shown in FIG.
Template data shown in FIG. 3(b) is prepared and compared with the image data of the drawing stored in the storage device 2' by the arithmetic device 3. The arithmetic unit 3 checks whether pixels within the size of the template data in the image data (portion within the broken line in FIG. 8C) match the pixel values of the template data, and determines whether the patterns match. This operation moves the matching range in the image data one pixel at a time (arrow in Figure 8(c)).
, is performed on the entire image data, and the results are stored in the storage device 2.

[Problem that the invention seeks to solve]

Since the conventional pattern recognition device is configured as described above, the calculation of the number of pixels of template data is performed as many times as the number of pixels of image data.

If the template data is 82 x 82 pixels and the image data is 100 x 100 pixels, the number of calculations is approximately 1.
,000,000,000 times. Even if the calculation device 3 can perform calculations 1.000,000 times per second, it will take 1000 seconds. As described above, there was a problem that processing took a long time. This invention was made to solve the above problem, and it is possible to obtain a pattern recognition device that recognizes a specific pattern from image data in a short time. The purpose is to

[Means for solving problems]

The pattern recognition device according to the present invention specifies a certain pixel having a characteristic value in a pattern to be recognized, obtains pixel data having that characteristic, stores it in a storage device, and executes the process using a central processing unit. A specific pattern is recognized by performing arithmetic processing. Furthermore, pixels having a plurality of characteristics are determined by high-speed logical operations.

[For production]

In order to specify a specific pixel in the pattern recognition device of the present invention, a pixel that is a characteristic of a pattern is specified, and a logical operation is performed only on the specified pixel, thereby recognizing a pattern with a small number of operations.

For the operations on the identified pixels, logical operations are performed on the pixels of one word on the image data at once using an internal register of the central processing unit.

〔Example〕

An embodiment of the present invention will be described below with reference to the drawings. 1st
In the figure, 4 is a storage device that stores the calculation result of the binary image data of the drawing and the data of the pattern to be recognized, and 5 is the storage device that stores the recognition data in which a corridor is formed by 101 data of the pattern size to be recognized. temporary storage device, 6
is a central processing unit (abbreviated as CPU). Note that the central processing unit 6 has a function of performing logical operations on areas specified in bit units and other areas at high speed, and executes the above operations at approximately several Mbytes/second.

Next, the operation will be explained. First, in the flowchart of FIG. 2, initial settings are performed in step 7. Binary image data such as a drawing, a reference pixel of a pattern to be recognized, pixel values characteristic of the pattern, and coordinates for the reference pixel are input into the storage device 4. Further, an area for calculation result data is secured in the storage device 4 according to the size of the image data and initialized to "1". The following description will be made assuming that the pattern shown in FIG. 3(a) is recognized. In this case, in step 7, information on the reference pixel A and the characteristic pixels B, C, and E is input as shown in FIG. 3(b).

Then, let A, B, C, and E be specific pixels of the pattern.

In step 8, the coordinates (X,
Y) and its value P are set to the values of the reference pixel.

Next, in step 9, a target area for the logical operation is set based on the coordinates (x, y). That is, the one-dimensional distance (difference in offset value) between the target image data and the image data when the image data is arranged as a one-dimensional memory is determined. The start offset value of the image data is S. If it is FF, the start offset value R8FF of the target image data is ROFF = 5OFF -XHX Y + X
- It is determined by (1). Here, XH indicates the horizontal size 44 of FIG.

In step 10, the value of P is determined and pro, that is, P
If = 1, go to step 11, if P = o, step 1
Move on to 2.

In step 11, the following logical operation is performed on the image data.

R-(RoFF)・(5OFF) ・=・(
2) Here, ' (SOFF) is the content of the image data (starting offset value of the image data), (RO□) is the content of the target image data, R is the register that stores the calculation result,
・represents logical product.

This calculation will be explained with reference to FIG.

FIG. 4(a) is a diagram showing image data, FIG. 4(b) is a diagram explaining a logical operation between the image data and target image data, and FIG. 4(c) is a diagram showing the calculation result.

The logical operation in Q11) in FIG. 4 will be explained using the case of feature point B in FIG. 3 et al. Here, for the sake of clarity, the explanation will be given on a two-dimensional memory, and then the operation on a -dimensional memory will be explained.

First, the coordinates of feature point B with respect to reference pixel A are (XB, Y
B) and the value is 1”, so S(x+XB, y+YB)
By overlapping and S(X, y), it becomes as shown in Fig. 4, etc.).

41 is image data of S(x,y), 42 is S(x+
XB-3'+YB) Indicates image data. For the pixel (xo, yo) that is the reference pixel of the pattern on the image data, S (xo +XB -'fo +YB) is 111. Therefore, the specific pixel (x, y) on the pattern is 1
If it is 1, there is a possibility that the reference pixel of the pattern exists at the pixel where S (x+X, )'+Y) is 111.

Superimposing image data 41 and 42 is equivalent to performing a logical product of image data 43 and 41 that are shifted by coordinates (XB, YB) with respect to image data 41. Considering this as a logical operation on -dimensional memory, image data 45...starting offset value 8OFF to 3
The 2-bit data and the target image data 46 . . . 32-bit data starting from the leading offset value ROFF are ANDed.

In step 12, the following logical operation is performed.

R←(ROFF)·R (3) R is a register that stores the calculation result, and (Rob?) indicates the inverted contents of the target image data.

This calculation will be explained with reference to FIG. Figure 5(a) is a diagram explaining E in Figure 3(a), Figure 5(c)
is a diagram showing calculation results. The coordinates of pixel E with respect to the reference plane amateur are (XE, Yl) and the value is 101, so S
When (x+XB, y+Yi) and S<x, y) are superimposed, it becomes as shown in Fig. 5~). 51 is 5 (X-7), 5
2 indicates S (x+XB, y+Ym).

A pixel (xo,
yo) then S (XO+XE,)'o+Y]if )
becomes 101.

Therefore, if the specific pixel (x, y) on the pattern is 101, there is a possibility that the reference pixel of the pattern exists at the pixel (x-y) on the image data where S (x+X, y+Y) is 01.

Similarly to step 11, using the image data 51 as a reference (
Good image data 53 and 4 with coordinate deviation of XE #YE)
Performs a logical AND of 1.

The calculations in steps 11 and 12 are repeated several times on the target image data. In Fig. 3(b), feature point B,
The calculation in step 11 is performed for C, and the calculation in step 12 is performed for those results and the feature point E, and the results are transferred to the output area.

The above calculation is executed for the entire area.

FIG. 6 shows pattern recognition by this device.

That is, from the image data of FIG. 6(a), FIG. 3(a)
Recognize patterns. The result of performing the logical operation on feature point B is shown in FIG. 6(b), the result of performing the operation on feature point C is shown in FIG. 6(c), and the result of performing the operation on feature point E is shown in FIG. The calculation result shown in FIG. 3(d) is obtained, and the pixel where the reference pixel of the pattern in FIG. 3(a) exists is one! The remaining pattern is recognized. However, in this patent, since the calculations for the feature points B, C, and E are performed using registers, the situations shown in FIGS. It will be done.

To specify a specific pixel on a pattern, select a pixel that is characteristic of the pattern in relation to the image data.

The calculations of equations (2) and (3), steps 11 and 12, are performed at high speed by the central processing unit 6, which is capable of data calculations on variable length fields.

In the case of image data of 1000 x 100 Q pixels, even if the computing power of the central processing unit f6 is 4 Mbytes/sec, the computing time for one specific pixel on the pattern is approximately 0.0
3 seconds, if the pixels specified on the pattern are 10 pixels, then operation A
The time is approximately 0.3 seconds. Even if all 32×32 pixels are designated as specific pixels, the calculation time is about 30 seconds. Processing time can be shortened by specifying specific pixels of a pattern and performing high-speed logical operations.

In the above embodiment, the calculation output results from the central processing unit are stored in the storage device, but if they are stored in the storage device for display on the display device, the calculation results can be immediately displayed on the display device. be done.

In addition, although only logical product was performed as a logical operation on image data, in order to provide margin for pattern recognition, feature points m
This is executed by finding n cases that hold true out of
≦m) Yes. Further, regarding this processing, since the number of logical operations between image data increases, the number of operations of internal registers increases, and the speed of the storage layer can be increased. (for example,
The internal register operation speed is approximately 32 Mbytes/sec. ) [Effects of the Invention] As described above, according to the present invention, by specifying specific pixels of a pattern and performing logical operations only on the specified pixels, pattern recognition can be performed with a small number of operations, resulting in high-speed pattern recognition. By performing logical operations, it is possible to recognize specific patterns in a short time.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the configuration of a pattern recognition device according to an embodiment of the invention, FIG. 2 is a flowchart showing the operation of the pattern recognition device according to an embodiment of the invention, and FIGS. ), Figures 4(a)-(c). FIGS. 5(a) to (e) and 6th oral history) to (d) are explanatory diagrams showing the operation of a pattern recognition device according to an embodiment of the present invention, and FIG. 7 shows the configuration of a conventional pattern recognition device. The block diagram and FIGS. 8(a) to 8(c) are explanatory diagrams showing the operation of a conventional pattern recognition device. In the figure, 4 is a storage device, 5 is a temporary storage device, and 6 is a central processing unit. Patent Applicant: Mitsubishi Electric Corporation, ------- Figure 1 6: Central MI Actual Processing System Figure 2 Figure 3 Figure 4 ■ Figure 5 Figure 6 t

Claims (1)

[Claims] A storage device that stores binary image data of a drawing and data of specific pixels of a pattern to be recognized, and a logical operation between data in an area specified in bits of the storage device and data in other areas. , a temporary storage device that stores the position information of specific pixels of the pattern to be recognized, and an area specified in bit units of the storage device and other areas and types that are controlled while using registers inside the central processing unit (described later) to create an image. A pattern recognition device comprising a central processing unit that collectively performs logical operations on pixels for one word on data.