JPS6311718B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Application of the Invention] The present invention relates to a character pattern cutting method for cutting out two-dimensional image character data written on large drawings, forms, etc.

[Background of the invention]

For example, when an optical character reading device (hereinafter referred to as OCR) scans a document with characters written on it, a sensor with photoelectric conversion elements arranged one-dimensionally is used to detect the position of the sensor and the document. Two-dimensional scanning is performed by moving them relatively.

By the way, when the form becomes large and one sensor cannot cover the entire width of the form,
As shown in FIG. Scanning line 6 and scanning line 7 by sensor 2
Conventionally, it is known that one scanning line is constructed by aligning the connection points with the scanning line.

However, in the conventional method described above, even if the connecting points of the scanning lines of the sensors 1 and 2 are precisely aligned, the sensors 1 and 2 or the lenses 3 and 4
Due to variations in quantization, quantization errors, etc.
In the image data obtained during the process, a shift occurs at the connection point. Therefore, when such data is recognized, there is a problem that the recognition result is adversely affected.

Furthermore, when recognizing the image data obtained by the sensors 1 and 2, it is necessary to temporarily store the entire image data in a memory, which poses a problem in that a very large memory capacity is required.

Further, devices for reading large-sized drawings such as engineering drawings also have the same problem as described above, and no effective solution to the problem has been found at present.

[Object of the present invention]

The purpose of the present invention is to enable extremely accurate recognition;
Another object of the present invention is to provide a character recognition method that can reduce the capacity of a memory for storing image data.

[Summary of the invention]

In order to achieve such an object, the present invention divides the screen to be read into a plurality of parts, and configures the divided screens so that some of the adjacent divided screens overlap,
The feature is that characters are recognized in units of such divided screens, and the recognition results are compared to read the entire screen data.

[Embodiments of the invention]

Hereinafter, embodiments of the present invention will be explained in detail using screens.

FIG. 2 shows the principle of the character pattern cutting out method according to the present invention, and is an example of reading a large-sized drawing.

As can be seen from the figure, the large screen 10 has a plurality of small areas 11, 12, 1, which are units that can be processed at once.
3, . . . An overlapping area 14 is provided between adjacent small areas, for example, 11 and 13, and an overlapping area 15 is similarly provided between 11 and 12. When data on such a large screen 10 is read, it is performed in units of small areas. For example, photoelectric conversion is performed for each small region in the row direction indicated by the arrow in the order of small regions 11 and 12, and then photoelectric conversion is sequentially performed for the row including small region 13. The image data of each small area obtained in step 1 is sequentially recognized, and reading of the entire screen data is completed based on the recognition result of each small area.

FIG. 3 shows a block diagram of an embodiment of a large-sized drawing reading device that implements the character pattern cutting method according to the present invention, in which numeral 21 denotes a split screen input device;
22 is a split screen storage memory, 23 is a character/line recognition section, 24 is a recognition table storage memory, 25 is a control section, and 26 is a connection processing section.

Screens such as blueprints generally include characters 27 and lines 28 as shown in FIG. Longitudinal length is SX and
After photoelectrically converting the split screen of the area 11 which is SY by the split screen input device 21 and performing preprocessing such as binarization processing, the split screen is stored in the image memory 22 for storing the split screen.
Store in.

Therefore, this memory 22 stores the data as shown in FIG. 4b. However, the fourth
(Pi) in Figure b indicates the coordinates. Next, the image data stored in the memory 22 is read and sent to the character/line recognition section 23, and as a result of character cutout recognition and line recognition in the split screen, split screen recognition as shown in FIG. 4c is performed. A table 30 is obtained and stored in the recognition table storage memory 24. In this recognition result, a line is represented by the coordinates of feature points such as end points and bending points. For example, line a has coordinates P ₁ ,
P ₂ , P ₃ , and P ₄ are expressed as a line a (P ₁ , P ₂ , P ₃ , P ₄ ). Characters are also represented by character types and their coordinates. For example, the letter A is
It is expressed as (P8). Furthermore, although not shown in the figure, the curve is represented by a method such as polygonal line approximation. When recognition of the split screen of the small area 11 is completed in this way, the next split screen of the small area 12 is captured and recognized in the same manner as described above, and the entire area of the large screen 10 or the split screen of a desired area is performs recognition processing.

In this way, the recognition results for each split screen are stored in the memory 24 as the split screen recognition table 30. When these processes are completed, the connection processing unit 26 uses these split screen recognition tables 30.
The contents are read out, the contents are selected and the recognition results are connected between adjacent split screens, and as a result, a recognition table 29 as shown in FIG. 4d is created and stored in the recognition table storage memory 24. Store.

As mentioned above, when inputting adjacent split screens, an appropriate overlapping area is provided, so for example, only a part of the split screen of the small area 11 is included,
Character C that is unrecognizable with that cutout pattern
[Figure 4a] is included in the divided screen of the adjacent small area 13 in its complete form, so it is completely recognized in that area, the one that is deemed unrecognizable is discarded, and the one that is completely recognized is discarded. selected. As a result, accurate recognition results can be obtained in the recognition table 29. In addition, characters that are included in complete form in each of the split screens of two adjacent small areas, such as the character B, are included in each recognition result, but when connecting both recognition results, one is discarded and only the other is recognized. It is registered in table 29.

Furthermore, as described above, one split screen is stored in the image memory 22, and after recognition processing, the next split screen is stored in the image memory 22, so the image memory 22 stores one split screen. It is sufficient to have the capacity, and the memory capacity can be significantly reduced compared to the case where the entire screen is stored.

In addition, in the configuration shown in FIG. 3, the split screen input device 2
1. The operations of the character/line recognition section 23 and the connection processing section 26 are controlled by control data and timing signals from the control section 25.

The operation of each part of the apparatus shown in FIG. 3 will be explained in detail below.

FIG. 5 shows an example of a specific configuration of the split screen input device 21 shown in FIG.

In the figure, 31 is a large screen mounting table, 32 is an image input head such as a television camera, 33 is a movable rail, 34 and 35 are fixed rails, 36 is a motor for moving the image input head, 37 is a motor for moving the movable rail, 38 is a motor drive circuit, 39 is an image input head movement control circuit, and 40 is an image input control circuit. Further, 10 is a large screen, and 22 is an image memory.

In such a configuration, the image input head 3 is controlled from the control section 25 to the image input head movement control circuit 39.
Accordingly, the movable rail moving motor 37 and the image motor 36 are driven through the motor drive circuit 38 to move the image input head 32 along the movable rail 33 in the Y direction. , the movable rail 33 is moved in the X direction along the fixed rails 34 and 35, and the image input head 32 is moved to the target position.

Next, when the control unit 25 issues an input activation signal to the image input control circuit 40, the video signal of the divided screen of the small area captured by the television camera or the like of the image input head 32 is preprocessed, and the result is stored in the image memory. 22
is memorized.

Then, movement target values are input from the control unit 25 one after another, and each time the divided screen of the corresponding small area is stored in the image memory 22.

Note that when using an array sensor linearly arranged in the X direction in the image input head 32, after moving the image input head 32 to the target position, move the image input head 32 to a small area in the Y direction. input the video signal while moving it by the amount corresponding to
After this is completed, the image input head 32 is moved back by the amount of the overlapping area, and the video signal of the divided screen of the next adjacent area is input.

As the split screen input device 21, a device with a fixed image input head and a movable drawing can be used, or a device with a plurality of image input heads that cover the entire screen or a necessary part of the screen can be used. The outputs from these heads may be selectively taken out.

Conventionally, various methods have been used for recognition in the segmented character/line recognition section 23, and no particular explanation is required. A microprocessor may be used as the device.

To connect characters in the connection processing unit 26 in FIG. 3, the coordinates of characters in split screens in adjacent areas are compared, and if they match, one character is deleted, and if they do not match, both are registered in the recognition table. Do by doing.

Such processing can be easily performed by using a processor such as a microcomputer. Further, a dedicated device that performs the above-described functions may be configured as the connection processing section 26.

Note that the control unit 25 may be configured with a processor such as a microcomputer.

Further, as the memories 22 and 24, random access memories such as microcomputers can be used.

Furthermore, the apparatus shown in FIG. 3 may be configured with one microcomputer or the like.

FIG. 6 shows the principle of an example of OCR that implements the pattern recognition method according to the invention.

As shown in FIG. 6a, when reading characters on a large form 5, a plurality of sensors 1, 2
However, in the present invention, two adjacent sensors 1 and 2 perform overlapping scanning by a width sufficient to cover one character. As a result, the areas A1 and B1 scanned by sensors 1 and 2 become as shown in FIGS. 6b and 6c, for example. In this way, characters near the overlapping area of the scanning areas of the sensors 1 and 2 will necessarily be included in complete form in either area, as shown in FIGS. 6b and 6c. For example, in FIG. 6b, the character "5" is included in complete form in the scanning areas A1 and B1 of both sensors 1 and 2.
Further, in FIG. 6c, the character "4" is included in the scanning area A1 of the sensor 1, and the character "5" is included in the scanning area B1 of the sensor 2 in complete form.

〔Effect of the invention〕

As described above, according to the present invention, not only can a large screen be recognized extremely accurately, but also the capacity of the screen memory can be significantly reduced.

[Brief explanation of the drawing]

Fig. 1 is an explanatory diagram of conventional OCR, Fig. 2 is an explanatory diagram of the principle of the pattern recognition method according to the present invention, and Fig. 3 is an explanatory diagram of the principle of the pattern recognition method according to the present invention.
The figure is a block diagram of an embodiment of a large drawing reading device that implements the pattern recognition method according to the present invention, FIG. 4 is an explanatory diagram of the operation of FIG. 3, and FIG. 5 is an example of the split screen input device of FIG. 3. FIG. 6 is a diagram explaining the principle of OCR that realizes the pattern recognition method according to the present invention. 10 is a large screen, 11 to 13 are small areas, 14 and 15 are overlapping areas, 21 is a split screen input device, 22 is an image memory, and 26 is a connection processing section.

Claims (1)

[Claims] 1. Divide a screen with characters into a plurality of subdivisions so as to have parts that overlap with adjacent subdivisions, photoelectrically convert the image on the partial divisions, recognize the character pattern from the photoelectric conversion signal, and As the recognition result for the character on the boundary line of one of the pair of partial compartments, the recognition result of the character at the position corresponding to the position of the character on the boundary line in the other compartment of the pair of adjacent partial compartments is adopted. A character recognition method characterized by: