JPH04280394A - Character recognizing device and recognized character fixing method - Google Patents

Abstract

PURPOSE:To present the character recognizing device and the recognized character fixing method, which can obtain correct character recognition with much higher probability, concerning the device and the method to fix recognized characters by processing (post-processing) character data from an OCR (optical character reader). CONSTITUTION:A character reader 2 is provided to read characters described in the respective plural items of a slip 1, and a recognition processing device 3 is provided to fix the recognized characters from the character data read by the character reader 2. The recognition processing device 3 specifies character candidates to be recognized as the characters described in the respective plural items for each item and converts the character candidate to be recognized as the character described in the first item to the character candidate to be recognized as the character described in the second item related to the first item. Further, the specified character candidate to be recognized as the character described in the second item is compared with the converted character candidate, and the coincident character candidate is recognized and fixed as the character described in the second item.

Description

[Detailed description of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition device and method for determining recognized characters by processing (post-processing) character data from an OCR (optical character reading) reading device. With the spread of OCR reading devices and the like, it is desired that read data can be processed without operator correction and handwritten characters can be recognized more accurately.

0002

2. Description of the Related Art FIG. 6 is a diagram illustrating a conventional character recognition device and its operation. 1 is an example of a transaction slip at a bank, etc., and has a plurality of item entry columns (items No. 1 to No. 3).

[0003] Item No. 1 is a field for entering the furigana of the name. Item No. 2 is a field in which the name is written in Kanji. Furthermore, item No. 3 is a column for writing an address. Normally, characters are written by hand in these item columns.

[0004] In banks and the like, data is read from such handwritten slips or forms using an OCR reader 2 and business processing is performed. Therefore, it is necessary that handwritten characters be accurately read and recognized. Otherwise, the operator will have to do a lot of correction work and business processing will become inefficient.

[0005] The characters written in each item column of the handwritten slip 1 are as follows:
It is read by the OCR reader 2 and guided to the recognition processing device 3. The recognition processing device 3 identifies recognition candidate characters for each item field.

[0006] Item No. 1 is ``Yamamoto'' in katakana.
The four characters are handwritten. In the recognition processing device 3, “Y” is detected based on the data read by the OCR reading device 2.
A candidate character of "ya" or "a" is specified for the character. For the character "ma", candidate characters "a" or "ma" are identified. Furthermore, for the character “mo”,
Candidate characters for "mo" and "sa" are specified, and only "to" for "to" is specified as a candidate character.

The recognition probability is 90% for candidate character field a, 80% for candidate character field b, and 70% for candidate character field c. The candidate characters corresponding to the four characters of "Yamamoto" in combination of each character are the eight characters from "Yamoto" to "Mamasato".
The correct combination (7) is obtained.

The recognition processing device 3 further searches a memory dictionary (not shown) for each of these eight combinations (
8). As a result of the search, combinations that do not exist in the dictionary are marked with an "×",
Existing combinations are indicated by "○".

That is, "Yamamoto" and "Yamasato" exist in the dictionary. "Yamamoto" has a recognition probability of 90% x 7
0% x 90% x 90%, and the recognition probability of "Yamasato" is 90% x 70% x 80% x 90%. Therefore, among the combinations existing in the dictionary, "Yamamoto" has the highest recognition probability.

On the other hand, in item No. 2, the Chinese character "Yamamoto" is handwritten. Processing is performed in the same way as the handwritten characters in item No. 1. Therefore, the characters for ``yama'' are ``small'' and ``yama''.
For the characters ``mountain'' and ``hon'', candidate characters ``hon'' and ``tree'' are identified. From these candidate characters, four combinations (7) of "Komoto" to "Yamaki" are obtained.

Furthermore, a dictionary is searched for these four combinations (8). As a result, the combination of "Ogi", "Yamamoto", and "Yamaki" exists in the dictionary. The recognition probabilities are the same for "Ogi" and "Yamamoto," and the recognition probability for "Yamaki" is the smallest.

In the conventional embodiment, the character with the highest recognition probability among the character candidates existing in the dictionary is determined as the recognized character (9). Therefore, for the characters listed in item No. 1, "Yamamoto" is determined as the recognized character. Item N
Regarding the characters listed in o2, the recognition probabilities for "Ogi" and "Yamamoto" are the same, so in this case, the combination with the highest appearance frequency is given priority, and "Ogi" is determined as the recognized character. Ru.

[0013] Therefore, in the recognition character confirmation method in such a conventional device, mutually related items No. 1 and No.
2, an error occurs in the recognition result of item No. 2,
The discrepancy between the recognition results for items No. 1 and No. 2 is also ignored.

[0014]

In the conventional apparatus, as explained in FIG. 6, the final recognized character is determined based only on the recognition probability. Therefore, when there are many candidate characters, it is difficult to expect correct character recognition with a higher probability using conventional methods. SUMMARY OF THE INVENTION An object of the present invention is to solve such problems and provide a character recognition device and a method for determining recognized characters, which can achieve correct character recognition with a higher probability.

[0015]

Means for Solving the Problems FIG. 1 shows a diagram of the principle of the present invention. In the figure, 1 is a slip or a form, and handwritten characters are written in a plurality of items (items 1 to 3). In item 1 and item 2, item data characters related to each other, such as the relationship between a name and its furigana, are handwritten.

Reference numeral 2 denotes an OCR reading device, which reads handwritten characters written in each item of the slip 1. 3 is a recognition processing device, which performs post-processing on character data read by the OCR reading device 2 to determine recognized characters.

In the present invention, the recognition processing device 3 has means 4 for specifying, for each item, character candidates to be recognized as characters described in each of a plurality of items from the character data read by the OCR reading device 2. Multiple character candidates are given for one handwritten character.

Furthermore, a plurality of character candidates to be recognized as characters listed in the first item (item 1) are converted into character candidates to be recognized as characters listed in the second item related to the first item. It has means 5. For example, if the characters listed in the second item are the kanji of the name and the characters listed in the first item are their furigana, the characters should be recognized as the characters listed in the first item. The plurality of furigana characters that are candidates are converted by the means 5 into Kanji character candidates to be recognized as characters listed in the second item.

Next, the character candidates of the Kanji characters converted by the means 5, the character candidates to be recognized as the characters specified in the second item in the means 4, and the character candidates converted by the converting means 5 are compared. compare. It has means 6 for recognizing and confirming character candidates that match as a result of the comparison as characters described in the second item.

[0020]

[Operation] The present invention converts a character candidate to be recognized as a character described in a first item into a character candidate to be recognized as a character described in a related second item, and Compare with character candidates that should be recognized as written characters. For example, the characters listed in the first item are katakana or hiragana, and the characters listed in the second item are kanji related to the characters listed in the first item.

Katakana or Hiragana in the first item is converted into related Kanji characters and compared with Kanji character candidates to be recognized as characters listed in the second item. The kanji character candidates that match through the comparison are determined as the characters listed in the second item. Compared to kanji, katakana and hiragana have a higher probability of being correctly read and recognized.

Therefore, since the present invention utilizes the relationship between the characters described in the first item and the characters described in the second item, it is possible to read and recognize the characters described in the second item. It is possible to increase the certainty of

[0023]

Embodiment FIG. 2 shows an embodiment of the present invention. The same or similar parts as in FIG. 1, which is a diagram of the principle of the present invention, and FIG. 6, which is a conventional example, are given the same numbers. 1 is an example of a slip, as in the conventional example shown in FIG.
The furigana is handwritten in item No. 1, and the address is handwritten in item No. 3.

Reference numeral 2 denotes an OCR reading device which reads the handwritten characters of each item on the slip 1 and sends it to the recognition processing device 3. The recognition processing device 3 performs post-processing to recognize and confirm the characters read by the OCR recognition device 2. In the recognition processing device 3, the recognition processing operation, which is a post-processing function according to the present invention, is performed based on software instructions stored in a ROM (read-only memory), not shown. The operational flow based on this software instruction is shown in FIGS. 3 and 4.

The post-processing operation of the recognition processing device 3 is executed in accordance with the definitions for each item of the slip 1. FIG. 5 shows an example of the definition of each item for slip 1 (10). The contents of the definition include the type of item (A), the line position of the item on the slip (B), the column position (C), and the related item number (D).

[0026] In item No. 1, the furigana of the name is written. Therefore, in related item No. (D), item No. 2 in which the Kanji name is written is defined as a related item. Definitions for each item of the slip 1 are stored in the magnetic disk 11. This definition body is the magnetic disk 1
1 and is read from the RAM (not shown) of the recognition processing device 3.
(writable memory).

Therefore, the recognition processing device 3 first checks whether the definition program has been loaded into the RAM (step S1 in FIG. 3), and if it has not been loaded, the recognition processing device 3
1 and loaded into the RAM (step S2 in Figure 3).
). From the data read by the OCR reading device 2, item information corresponding to the item to be processed is acquired according to this definition body (step S3 in FIG. 3).

That is, when item No. 1 is the item to be processed, the OCR reading device 2 corresponding to the written character "Yamamoto"
The candidate character combinations (FIG. 2-7) are obtained from the candidate characters of each character, and the dictionary is searched for each candidate character combination (FIG. 2-8). As a result, if the candidate character combination exists in the dictionary, "○" is displayed, and if it does not exist, "×"
” is indicated. This processing operation is similar to the operation of the conventional device shown in FIG.

The definition of item No. 1 defines that the related item is item No. 2. Therefore, the recognition processing device 3 acquires the item information of item No. 2, which is a related item, as the next process (step S4 in FIG. 3).
).

[0030] In item No. 2, the Kanji name "Yamamoto" is written. Input the read data of the OCR reader 2 corresponding to the written character "Yamamoto", find candidate character combinations (Figure 2-7) from the candidate characters of each character, and search the dictionary for each candidate character combination (Figure 2-8). )do. As in the case of the written characters in item No. 1, if the candidate character combination exists in the dictionary, it is indicated by "○", and if it does not exist, it is indicated by "x".

According to the present invention, the candidate character (katakana) of item No. 1, which is the item to be processed, is then changed to the related item No.
2 (kanji) (step S5 in FIG. 4). As a result, the candidate characters "Yamamoto" are "Yamamoto" and "Yamamoto".
The candidate character ``Yamasato'' is converted to ``Yamazato'' (Figure 2-5).

[0032] Next, the converted characters “Yamamoto” and “Yamamoto”
"Yamazato" is used as the comparison source data, and on the other hand, the candidate characters "Ogi", "Yamamoto", and "Yamaki" that existed through the dictionary search for item No. 2 are used as the comparison destination data, and are compared and compared in order (steps S6 and 7). , 8, 9).

As a result of the comparison, when the comparison source data and comparison target data match, the matching character is determined as the recognized character of item No. 2 (steps S10, 11). If there is a final mismatch, characters with high recognition probabilities are determined for each of item No. 1 and item No. 2 as recognized characters (step S1
0, 12, 7, 8, 6, 13). In the example of FIG.
Since a match can be made with the characters ``Yamamoto'', ``Yamamoto'' is determined as the recognized character for item No. 1, and ``Yamamoto'' is determined as the recognized character for item No. 2 (Figures 2-6).

In the above embodiment of the present invention, kanji are the objects of comparison, but the same result can be obtained even if the kanji candidate characters in item No. 2 are converted to katakana and compared with the katakana candidate characters in item No. 1. Get results.

[0035] Furthermore, the entry items such as "zip code" or "bank code" may be used as the first item, and the entry items such as name, address, etc. may be used as the second item for comparison. The recognition rate for "postal code" or "bank code" is 98
Since the rate is as high as ~99%, this case is particularly advantageous for increasing the probability of character recognition for the second item.

[0036]

[Effect] As explained in detail above, the present invention provides
Even if there are a plurality of recognized character candidates for character data read by an OCR reading device, an accurate character candidate can be selected by comparing with character candidates of related items. This provides a character recognition device that can perform more accurate character recognition.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing the principle of the present invention.

FIG. 2 is an embodiment of the present invention.

FIG. 3 is the first half of an operation flowchart in the embodiment of FIG. 2, and is common to the operation in the conventional example.

FIG. 4 is the second half of the operation flowchart in the embodiment of FIG. 2, and shows a portion of the operation flowchart that is a feature of the present invention.

FIG. 5 is an explanatory diagram of a definition body for each item of a slip in the embodiment of FIG. 2;

FIG. 6 is an example of a conventional recognition device.

[Explanation of symbols]

1. Slip 2 Character reading device 3 Recognition processing device

Claims (4)

[Claims] [Claim 1] A character reading device (2) that reads the characters written in each of the plurality of items of the slip (1), and a recognition processing device (2) that determines recognized characters from the character data read by the character reading device (2). 3), and the recognition processing device (3) includes means (4) for specifying, for each item, character candidates to be recognized as the characters described in each of the plurality of items, and means (5) for converting a character candidate to be recognized as a character into a character candidate to be recognized as a character described in a second item related to the first item; and specifying the character candidate to be recognized for each item. The character candidates to be recognized as the characters specified in the second item specified by the means for converting (4) and the character candidates converted by the converting means (5) are compared, and the matching character candidates are converted into the second item. A character recognition device comprising: (6) means for recognizing and confirming characters as described in item (6). [Claim 2] A recognition character confirmation method that reads the characters written in each of a plurality of items of a slip (1), and recognizes and confirms the read characters for each item, as the characters written in each of the plurality of items. Identify character candidates to be recognized for each item (4), and recognize character candidates to be recognized as characters listed in the first item as characters listed in the second item related to the first item. Convert to character candidates (5), compare the character candidates to be recognized as characters listed in the specified second item and the converted character candidates, and convert the matched character candidates to the second item. Recognition and confirmation as written characters (6
) A method for determining recognized characters. Claim 3: The characters listed in the first item are katakana or hiragana, and the characters listed in the second item are kanji corresponding to the characters listed in the first item. The character recognition device according to claim 1. Claim 4: The characters listed in the first item are katakana or hiragana, and the characters listed in the second item are kanji corresponding to the characters listed in the first item. 3. The method for determining recognized characters according to claim 2.