JPH01237778A - Character recognizing method - Google Patents

Abstract

PURPOSE:To loosen the limit of a writing character to the same level as a general slip by discriminating a tendency in the character of a writer from a recognized and outputted standard pattern and reading even a resemble character which can not be identified only with a character form. CONSTITUTION:A slip 2 is carried to a reading part, which is composed of a light emitting source 3 and a photo-detecting element 4, by a document carrying mechanism 1. This element 4 converts a pattern, which is successively inputted, to an electric signal and outputs it. As such an element 4, a one-dimensional semiconductor CCD sensor, etc., is used. Here, a binarizing circuit 5 converts an output signal from the element 4 to the binarizing signals of '1' and '0' by a prescribed threshold and outputs the signal to a memory 6. This memory 6 stores the pattern on the document as the binary picture of the '1' and '0' and a semiconductor memory, etc., is used for this memory. From the binary picture to be outputted from the memory 6, a segment/recognizing part 7 segments the character pattern in each character, and the result is compared with the standard pattern. Then, a recognized result is outputted.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a character recognition method, and particularly to a character recognition method suitable for identifying similar characters.

[Prior art] In conventional character recognition methods, when recognizing characters written on a form, etc., the characters to be recognized are registered in a recognition device as a standard pattern, and the unknown character pattern is compared with the standard pattern. death,! The standard pattern with good matching is output as the recognition result.

In recent years, character recognition devices have been realized that can recognize not only standard-shaped characters but also similar characters as shown in FIG. 6, for example, due to improvements in recognition performance. This type of character recognition device analyzes the structure and characteristics of characters in detail and identifies similar characters based on their shape.

As such conventional technology, for example, Toshiba Review 4
The technique described on pages 1014 and 1015 of Vol. 1, No. 12 (1986) is known.

[Problems to be Solved by the Invention] In general, similar characters have particularly strong character quirks (characteristics) depending on the person who wrote them, and the above-mentioned prior art does not take into consideration the characteristics of similar characters by the person who writes the characters. There was a problem in that there was a limit to the recognition of similar characters in different categories.

This invention was made in view of the problems of the prior art described above, and makes it possible to identify similar characters, which is limited by identification based on the structure and characteristics of the characters, by considering the characteristics of the character shape by the person writing the characters. The purpose is to provide a character recognition method.

[Means for Solving the Problems] The character recognition method of the present invention compares an unknown input pattern with a standard pattern to find a standard pattern with good consistency as a candidate character pattern, and selects a category indicated by the found candidate character pattern. It is applied to character recognition methods that determine and output recognition results from among the following, and in particular, for each standard pattern,
The number of times the standard pattern is output as a recognition result is counted as a first value, the number of times the category indicated by the standard pattern is output as a recognition result is counted as a second value, and the first value is is divided by the second value to obtain and memorize the third value, and for the category indicated by the standard pattern selected as a candidate character pattern, the above is applied to all standard patterns indicating the category excluding the candidate character pattern. The third value is referred to, and if the third value exceeds a predetermined value, the candidate character pattern is deleted from the candidates.

[Operation] In this invention, as shown in Figure 2 (a) = (b) and (c), the font of the characters entered differs greatly depending on the person who wrote it, and conversely, the font of the characters written by the same person is in the same category. This was done by noting that the characters are almost the same. It is something.

In other words, within the same category, the rate at which a certain standard pattern is output as a recognition result (the third value above) exceeds a predetermined value, and a standard pattern other than this standard pattern is used as a candidate character. When a new standard pattern is selected, the newly selected standard pattern is deleted from the list of candidate characters.

[Examples] Hereinafter, the present invention will be described in more detail with reference to Examples shown in the accompanying drawings and figures.

FIG. 3 is a block diagram of an optical character reading device showing one embodiment of the present invention. In FIG. 3, reference numeral 1 designates a document transport mechanism 1, 2 a document, 3 a light emitting source, 4 a light receiving element, 5 a binarization circuit, 6 a memory, and 7 a cutout/recognition section consisting of a microprocessor.

In FIG. 3, a form 2 is conveyed by a form conveyance mechanism 1 to a reading section comprising a light emitting source 3 and a light receiving element 4. As shown in FIG. The light receiving element 4 converts the sequentially inputted patterns into electrical signals and outputs the electrical signals. As the light receiving element 4, a -dimensional semiconductor CCD
Sensors etc. are used. The binarization circuit 5 converts the output signal from the light receiving element 4 into II l 11. by a predetermined threshold value. '0
'', and outputs it to the memory 6. The memory 6 stores the pattern on the form as a binary image of ``1'' and ''0'', and uses a semiconductor memory or the like. It will be done. The cutout/recognition unit 7 cuts out a character pattern character by character from the binary image output from the memory 6, compares it with a standard pattern, and outputs a recognition result.

FIG. 4 is a block diagram of a microprocessor that constitutes the extraction/recognition section 7. As shown in FIG. The extraction/recognition unit 7 includes an arithmetic unit 8, an internal work memory 9, an external interface adapter 10, and a common bus 11.

The arithmetic unit 81 cuts out and recognizes the binarized image in the memory 6 by executing an internal program, and outputs the result to the outside via the external interface adapter 11. The internal work memory 9 stores data and results necessary for extraction and recognition processing, and is also used as a dictionary memory for storing standard patterns.

FIG. 1 is a flowchart 1- showing the similar character recognition operation in the extraction/recognition unit 7, and FIGS. 5(a) to (f)
) is a diagram showing a specific example of the similar character recognition operation shown in FIG. 1. In step 100 of FIG. 1, the character pattern extracted as a - character is recognized. The recognition process compares a previously registered standard pattern with an unknown character pattern, and selects a consistent standard pattern. Output as recognition result. In the specific example shown in FIG. 5(a),
The character pattern P0 cannot be determined to be either the number "1" or the number 7 #l, and as a result of the recognition process, two candidate character patterns S 119 Stz are output as shown in FIG. 5(b). Here, the candidate character pattern S 11 is one of the plural candidate character patterns prepared for the category II 113, and the candidate character pattern StX is one of the plural candidate character patterns prepared for the category It 7 ++. one of. In the conventional technology, in such a case, the character cannot be determined based on the shape alone, so a code indicating that the character is unrecognizable is output as a recognition result.

However, in this embodiment, in step 200, it is determined whether there are two or more candidate character patterns.

That is, it is determined whether a similar candidate pattern exists in the candidate character pattern. If it is determined that there is only one candidate character pattern and no similar candidate pattern exists, this candidate character pattern is immediately output as a recognition result in step 600.

In the embodiment shown in FIG. 5(b), since there are two candidate character patterns, it is determined that a similar candidate pattern exists, and the process proceeds to step 300.

In step 300, for one of the plurality of candidate character patterns, it is determined whether or not the characters of the candidate character pattern have already been recognized in another candidate character pattern (standard pattern). refers to the recognized output value N for the A4 pattern, and determines whether the recognized output value N exceeds a certain reference value, for example 0.5.

In the specific example shown in FIG. 5(c), it is determined whether or not the character "'1" has already been recognized by a candidate character pattern (standard pattern) other than the candidate character pattern S1. Extraction/recognition unit 7 shown in Figure 4
The recognition output value N stored in the internal work memory 9 in
This is done with reference to. Here, the recognition output value N is the value obtained by dividing the number of times a character has been recognized using the standard pattern by the total number of times that the category has been recognized and output. In the specific example shown in FIG. 5(C), the recognition output value N for the standard pattern S i 1 is 1/3. Here, the denominator (3) of the recognition output value N indicates the total number of times that the category ``1'' has been recognized and output so far, and the numerator (1) is the standard pattern Syu, which has been output as the category at 1 ++ so far. Shows the total number of times it was recognized and output. In this case, category 1
11 ++ has already been read once by the standard pattern S, and the category It 1 ++ has been read three times in total. Therefore, in this case.

Candidate character pattern S-ko is candidate character pattern S-ko.

The recognized output value N is 1/3 and does not exceed 0.5. If it is determined in step 300 that the recognition output value N does not exceed 0.5, the process proceeds to step 500, where it is determined whether or not all candidate character patterns have been checked. If it is determined that the recognition output value N of the candidate character pattern exceeds 0.5, the process proceeds to step 400, where processing is performed to delete the candidate character pattern recognized in step 100 from the candidates.

In this embodiment, as described above, if the recognition output value N is 0.5 or more, the candidate character pattern is deleted from the candidate characters. This is because there is a high possibility that the candidate character pattern has been misrecognized. That is, the same characters written by the same person are similar and are often recognized using the same standard pattern.

After the process in step 400 is completed, step 5
The process advances to 00, and it is determined whether or not all candidate character patterns have been checked.

If it is determined in step 500 that the processing has not been completed for all candidate character patterns, the process returns to step 300 and repeats the above processing.
In the case of the specific example shown in Figure (c), step 300.4
As shown in the figure, in the case of the candidate character pattern S ffl, the recognition t of the standard pattern st* is performed.
Since the force value N is 8/8't'A LJo, 5, as shown in FIG. 5(d), the candidate character pattern St
t is removed from the candidates in step 400.

If it is determined in step 500 that all candidate character patterns have been checked, the process proceeds to step 600, where characters in categories corresponding to the candidate character patterns that were not deleted in step 400 are output as recognition results. In the case of the above specific example, II 1 #l is output as the recognition result, as shown in FIG. 5(e).

Next, in step 700, the cutout/
The recognition output value N of the internal work memory 9 in the recognition unit 7 is reversed, and the process ends.

In the above embodiment, the recognition output value N may be set to be cleared to 0 every time a new form is processed, or
Further, the conditions for clearing may be set in advance using format information or the like.

Further, similar characters may be determined after reading all the characters on the form.

Further, the standard pattern of the candidate character deleted in step 400 or the standard pattern other than the standard pattern whose recognition output value N exceeds a predetermined value is invalidated in the recognition process of step 100 and removed from the candidate characters to be read. In this case, the standard pattern that was invalidated when the person filling it out or the filling conditions changed may be valid.Furthermore, the standard pattern to be invalidated is selected from the candidate character patterns, but the recognition The resulting a semi-pattern may be a standard pattern in the same category of another academic form.

As described above, according to the embodiments of the present invention, even characters that cannot be distinguished based on the shape alone can be read by knowing the tendency of the characters appearing by the person writing them, and the characters can be recognized with human ability.

[Effects of the Invention] According to the present invention, the tendency of the characters written by the user can be determined from the recognized and outputted standard pattern, and even similar characters that cannot be identified by a single character shape alone can be read.

Restrictions on characters to be entered can be relaxed to the same level as general slips.

In addition, even handwritten numerals that have different character shapes depending on the US and West German standards can be read without changing the dictionary structure (especially similar characters), so both standards are used together, as in the US. This has the effect of facilitating the spread of character recognition devices in areas where

Furthermore, a dictionary composed of a large number of standard patterns is used to read nine different types of printed characters.

Even in multi-font reading where a plurality of printed fonts are read, even if there are similar character shapes between the fonts, if there are no similar characters within the same font, it is possible to separate them. In this case, if invalid standard patterns are excluded from the reading target, the reading time can also be increased.

[Brief explanation of the drawing]

Fig. 1 is a flow chart showing an example of implementing the character recognition method of the present invention, Fig. 2 (a) l Cb) * (c) and Fig. 6 are diagrams showing examples of filling in similar characters, and Fig. 3 is a flowchart showing an example of implementing the character recognition method of the present invention. A diagram showing an example of a character recognition device to which the invention is applied, FIG. 4 is a block diagram showing a specific example of the extraction/recognition unit 7 shown in FIG. 3, and FIGS. 5(a), (b), (c) . (d), (e), and (f) are diagrams showing specific examples of the processing of the flowchart shown in FIG. 1. 5... Binarization circuit, 6... Memory, 7... Extraction/recognition unit, 8... Arithmetic device, 9... Internal work memory. Agent Patent Attorney Tadashi Akimoto Figure 1 Figure 2 Figure 3 Figure 4 Figure 6 Figure “1” “7” Figure 5

Claims (1)

[Claims] 1. Compare an unknown input pattern with a standard pattern to find a standard pattern with good consistency as a candidate character pattern, and determine and output a recognition result from among the categories indicated by the found candidate character pattern. In a character recognition method, for each standard pattern, the number of times the standard pattern is output as a recognition result is counted as a first value, and the number of times the category indicated by the standard pattern is output as a recognition result is further counted. to obtain the second value, and divide the first value by the second value to obtain the third value.
The value of is calculated and memorized, and for the category indicated by the standard pattern selected as a candidate character pattern, the third value is referred to for each standard pattern indicating the category excluding the candidate character pattern. A character recognition method characterized in that, if the third value exceeds a predetermined value, the candidate character pattern is deleted from the candidates. 2. The character recognition method according to claim 1, wherein the deleted candidate character pattern is excluded from recognition targets in subsequent character recognition. 3. For a certain standard pattern, when the third value exceeds a predetermined value, all standard patterns other than the standard pattern in the category indicated by the standard pattern are excluded from recognition targets for the next character recognition. The character recognition method according to claim 1, characterized in that: