JPS62281090A - Recognition method for pattern information - Google Patents

Abstract

PURPOSE:To correctly recognize a pattern in a short time by matching the central value for respective divided areas to a previously held standard pattern, further taking a pattern matching only on the specific area of a part of a reading pattern to recognize the pattern when the matching is not taken. CONSTITUTION:The standard pattern stored in the standard feature pattern 51a of a ROM 51 is matched to an extracted feature pattern stored in the reading pattern 24. Consequently, when there are plural coinciding standard patterns, a pattern matching in the specific small area is carried out. In the respective areas, a black density is present or not and whether there is more than at least one black density or not in the area 71 may be discriminated. In this manner, a desired pattern can be specified by a simple pattern matching processing.

Description

Detailed Description of the Invention 3. Detailed Description of the Invention [Field of Industrial Application] The present invention relates to a pattern information recognition method for automatically reading and recognizing unknown characters such as alphabets and numbers.

[Prior Art] FIG. 13 shows character recognition processing of a conventional character recognition device of this type.

The characters on the manuscript paper P are read character by character and photoelectrically converted as shown in step S81, and further converted into a binary data pattern of 1°' and "ooo" in step S82. In the following step S83 , Pre-processing to make the processing described later effective, that is, noise processing such as black dots on paper P on which characters are recorded based on the binarized pattern in step S82, and smoothing of character/figure boundary surfaces. A series of processes including the following are performed.

Next, in step S84, feature extraction processing is performed to extract several features (intersections, branch points, number of loops, stroke length information, etc.) necessary for character recognition. Depending on this extraction result, some characters with common features are selected from a large number of characters.

Further, from step S85 onward, dictionary collation processing and recognition processing are performed as identification processing for selecting a unique character from among the characters.

First, it is checked whether it is possible to guide (search) the dictionary according to the features extracted in step S85. If the feature extracted here is not worthy of guiding a dictionary, the process advances to step S88, and unrecognizable is output. On the other hand, if dictionary guidance is possible based on the extracted feature points, step 38
In step 8, the built-in dictionary is searched and sequentially compared with the previously extracted features. If a match is found as a result of the comparison in step S87, the matching result is outputted as a recognition output in step 388. If a match cannot be found, it is output as unrecognizable.

[Problems to be solved by the invention] However, according to this conventional recognition method, simple characters (
Since the above-mentioned complicated processing is performed even for cases where only characters (which are easily recognized), alphabets, and numbers are included, there is a drawback that the processing time is too long.

Further, the configuration for realizing the above processing is also complicated, and there are various problems in terms of increased cost and reliability.

[Means for Solving the Problems] The present invention has been made in view of the problems of the prior art described above, and includes the following configuration as a means for solving the above problems.

That is, a reading means for optically reading pattern information on a reading surface, an area dividing means for dividing the pattern information read by the means into each predetermined area, and a representative pattern for each area divided by the area dividing means. a specifying means for specifying the value; a first matching means for comparing the specific pattern of the specifying means with a standard specific pattern held in advance; and, if the value is not specified by the means, a reading pattern by the reading means; The second matching means performs pattern matching of the specific pattern of the identifying means in at least one specific area of a part of the information.

[Operation] In the above configuration, the pattern information read by the reading means is divided into predetermined regions by the dividing means, a representative value for each divided region is determined by the dividing means, and this representative value is stored in advance. Performs matching with standard pattern,
If matching is achieved, it is recognized that a pattern corresponding to the standard pattern has been read, and if matching is not possible, pattern matching is further performed only on a specific area of the read pattern to recognize the pattern.

[Example] Hereinafter, an example according to the present invention will be described in detail with reference to the drawings.

FIG. 1 is a block diagram of one embodiment of the present invention.
2 is an optical disk that stores the image information read by the reader 1; 3 is a host computer that controls the entire embodiment; 4 is a keyboard; 5 is a A CRT displays image information, operation information, etc., and 6 is a printer that prints out the image information. Further, reference numeral 7 denotes an auto feeder that automatically feeds originals one by one to the original reading surface of the reader 1.

In the host computer 3, 50 is a CPU that controls the entire apparatus according to the control program shown in FIG. 4 stored in the ROM 51. The CPU 50 is preferably configured with a microcomputer MC68000 manufactured by Motorola, for example. 51 is a ROM, ROM5
1 includes a standard feature pattern storage area 51a. 5
2 is a CRT interface that controls the interface with the CRT 5; 53 is a keyboard interface that controls the interface with the keyboard 4; and 54 is an optical disk interface that controls the interface with the optical disk. Further, 55 is a reader interface that controls the interface with the reader 1, 56 is a printer interface that controls the interface with the printer 6, and 57 is an R that stores the processing progress, the reading mark recognition processing progress, etc.
AM, 60 is a bus that connects each component.

A detailed storage area of the RAM 57 is shown in FIG.

In FIG. 2, 21 is a pattern buffer, 22 is a line histogram buffer, 23 is an area histogram buffer, and 24 is a read pattern.

FIG. 3 shows an external view of this embodiment having the above configuration.

In the figure, the same components as in FIG. 1 are given the same numbers. 8
9 is an interface cable between the reader 1 and the host computer 3, and 9 is an interface cable between the printer 6 and the host computer 3.

The operation of this embodiment having the above configuration will be explained below with reference to the flowchart of FIG.

In FIG. 4, as in FIG. 13, the process of reading and recognizing a pattern on a document set on the reader 1 will be explained as an example.

First, in step S1, the CPU 50 activates the reader 1 via the reader interface 55, scans the surface of the document to be read, and reads image data on the surface of the document.

For example, the manuscript surface is not shown in FIG.
Apply as shown. The OMR sheet is used for registering keywords in an electronic file system, manually setting a telephone number for a facsimile, and setting modes such as the number of copies and reduction ratio of an image forming apparatus such as a copying machine.

If the format shown in FIG. 11 is fixed and a fixed pattern of characters is written on it, subsequent processing becomes easier. In addition, 'A' to 'Z' are shown in Figure 11.
” and “t” to “o” (7) A regular pattern is shown.

The basic pattern of the format shown in Figure 11 is three lines in the X direction,
It is part of the shape that combines the character ``Ta'', which is made up of three strokes in the X direction, and the ``X'' strokes in two diagonal directions.

Below, O"' is written at the mark position on the manuscript surface,
The recognition process when t, +-= are read will be explained as an example.

The signal photoelectrically converted and read by the reader 1 in step S1 is sent to the reader interface 55 as shown in step s2, and converted into a binary pattern of "1" and "No" by the analog-to-digital converter built into the reader interface 55. It is converted into a digital signal, and the converted digital signal is input to the CPU 50.

The CPU 50 sequentially stores the thus input digital data in the pattern buffer 21 of the RAM 57. In this embodiment, the character pattern is divided into areas of 48 bits x 48 bits per character and stored in the pattern buffer.
1.

In step s3, the CPU 50 divides this digital data into a plurality of predetermined areas. The reading pattern is, for example, the fifth
It is divided into each predetermined area shown in 1 to 1o in Figure (A). Then, in step s4, a stroke input due to continuous black density within a predetermined range for each line of each divided area is detected, and a histogram of the input stroke distribution is determined. Note that when there is black density, the histogram is 1°, and when there is no black density, the histogram is 0. Then, the histogram of the stroke distribution obtained in step S5 is stored in the line histogram buffer 22 of the RAM 57. .

This histogram detection is not limited to this example.
Instead of the above process, it may be determined that the stroke is made manually when the black density continues to be equal to or greater than a predetermined amount.

For example, the histogram of the stroke distribution for each line in the horizontal direction in region 1 divided as shown in FIG. 5(A) is as shown in FIG. , 2, 2, 2.2''.

Subsequently, in step S6, the minimum value of the histogram for each line within the predetermined region obtained earlier is obtained, and this is stored in the region histogram buffer 23 of the RAM 57 as one of the representative values of the predetermined region. In region 1 mentioned above, it is 2°. Similarly, in step S7, the maximum number of strokes generated in the histogram distribution for each line within the predetermined region previously obtained is obtained, and this is also stored in the region histogram buffer 23 of the RAM 57 as one of the representative values of the predetermined region. Area 1 mentioned above
In this case, the maximum number of strokes generated in the histogram distribution is also 0.2''.

Then, in step S8, a matching pattern for the reading pattern is determined (extracting features) from the representative values stored in the area histogram buffer 23, and is stored in the reading pattern 24 in the RAM 57.

This matching pattern (feature extraction) can be set freely.For example, the representative value (feature) of each region for each region is set to the maximum number of strokes that occur in the histogram distribution, and the maximum number of strokes generated in the histogram distribution is used as the representative value for each region, and for other regions, The minimum value of the histogram can be used as the representative value of the area.

Here, an example will be explained in which the representative value of each area is the minimum value of the histogram. When the minimum value of the histogram is used as the representative value of each area, the matching pattern stored in the read pattern 24 of the RAM 57 is as shown in FIG. 6(A).

In the following step SIO, standard characteristic pattern 5 of ROM51
The standard pattern stored in 1a is matched with the extracted feature pattern stored in the read pattern 24. As a result, if no pattern matching is obtained in step Sit, the process proceeds to step S12, where an unrecognizable message is output and the process ends. Then, the next pattern reading process is executed again.

If a match is found, the process proceeds to step S13, where it is checked whether there is only one pattern that matches.

If there is only one pattern, the process advances to step S14, where it is recognized as a matched pattern input, and the recognition result is output. On the other hand, if there are a plurality of matched standard patterns, the process advances from step S13 to step S15.

For example, in the case of the human cover turn shown in FIG. 11, the following human forces are considered to be the same pattern.

■``U'' and V'' ■ ``5'' and ``Z''
“2” and “S”■ “L” and “T” and “1 °° and
"7" Therefore, in such a case, pattern matching is performed in a specific small area specific to each pattern.

The matching areas for each pattern are shown in FIGS. 7 to 9. Figures 7 (A) and (B) show the case of ■ above, Figures 8 (A) to (D) show the case of ■, and Figures 9 (A) to (D) show the case of ■. There is.

In each of these areas, there is a black density or not, and it is sufficient to determine whether or not there is at least one or two black densities in the area.

A desired pattern can be specified by such a simple pattern matching process.

If the candidate pattern is “U”, standard feature pattern 5
The standard feature pattern stored in 1a is shown in Figure 6 (B).
This pattern corresponds to the reading pattern shown in FIG. 6(A). However, as shown in Fig. 10, the candidate character pattern in Fig. 6 (B) is not only °゛U'°;
It is the same as the "v" pattern. So step 5
A pattern matching process for a specific small area, which is the feature extraction process No. 15, is executed. In this case, the seventh
The black density of the area shown in FIG. 71 is examined.

If the human power is "U", the read data is the pattern shown in 72, so a large amount of black density data exists in this area. This allows the CPU 50 to specify the reading pattern as "U". Note that the detection of the black density of this small area may not be determined based on one or more, but it may be determined that black density is present when black density data of a predetermined value or more is detected. This makes it possible to further eliminate the influence of dust and the like.

As explained above, according to this embodiment, (1) It is possible to easily and quickly narrow down patterns with common features that cannot be recognized by comparing stroke distributions alone. That is, by simply performing pattern matching on only a specific small area, one of the many candidates can be specified.

■Furthermore, since the amount of data to be collated is small, processing time is significantly reduced.

In this way, the drawbacks of the character recognition device in recognizing simple characters can be overcome, the processing time required for simple character recognition can be shortened, and character recognition can be performed at high speed and with high accuracy.

Furthermore, it is possible to recognize numbers in a fixed pattern (printed type) or letters of the alphabet at high speed and with high precision.

[Effect of the invention] As explained above, according to the present invention, in a very short time,
Moreover, accurate pattern recognition processing can be performed.

[Brief explanation of the drawing]

Fig. 1 is a block diagram of an embodiment according to the present invention, Fig. 2 is a detailed configuration diagram of the RAM shown in Fig. 1, Fig. 3 is an external view of this embodiment, and Fig. 4 is a pattern of this embodiment. Recognition control flowchart, FIGS. 5(A) and (B) are diagrams showing an example of storing read patterns in the pattern buffer of this embodiment, and FIGS. 6(A) and (B).
) are diagrams showing matching patterns according to this embodiment, FIGS. 7(A), (B), FIGS. 8(A) to (D), and
Figures (A) to (D) are diagrams showing examples of setting small regions for feature extraction in this embodiment, Figure 10 is a diagram explaining the u'' recognition process in this embodiment, and Figure 11 is a diagram illustrating the Figure 12 is a diagram showing a standard human cover turn used in this example, Figure 12 is a diagram showing an OMR sheet used in this example, and Figure 13 is a flowchart showing conventional pattern recognition processing. , 2... optical disc, 3...
・Host computer, 4...Keyboard, 5...
CRT, 6...Printer, 7...Auto feeder,
5 (1-CPU, 51''・ROM, 51a--Standard feature pattern storage area, 52...CRT interface, 53...Keyboard interface, 54...
Optical disk interface, 55...Reader interface, 56...Printer interface, 57...
-RAM, 60...is a bus.

Claims (5)

[Claims] (1) A reading process that optically reads the pattern information on the reading surface, an area dividing process that divides the read pattern information into predetermined areas following this process, and each area divided in the area dividing process. a first matching step of comparing the specific pattern of the specific step with a standard specific pattern held in advance, and if the value is not specified as 1 in this step, A second step of performing pattern matching of the specific pattern of the specific step of at least one specific area of a part of the read pattern information of
A pattern information recognition method characterized in that pattern information is recognized by a matching step. (2) The pattern information recognition method according to claim 1, wherein the identifying step detects a histogram of stroke distribution and identifies the minimum value of the detected histogram of stroke distribution as the representative value of the divided area. . (3) Pattern information recognition according to claim 1, characterized in that the identifying step detects a histogram of stroke distribution and identifies the maximum number of occurrences of the detected histogram of stroke distribution as a representative value of the divided area. Method. (4) Pattern information recognition according to claim 2 or 3, characterized in that the identifying step detects a continuity of black density within a certain range and identifies the detected continuity of black density as a stroke input. Method. (5) In the second matching step, pattern matching of a specific pattern in a specific area is determined based on whether the black density of the specific pattern in the specific area is greater than or equal to a predetermined amount. The pattern information recognition method according to any one of item 4.