JPH04218885A - Character recognizing method for optical character reader - Google Patents

Abstract

PURPOSE:To limit a character specification even when a large number of characters exist in the character specification when character recognition is efficiently performed by limiting the character specification at every field of a document. CONSTITUTION:A character specification designation mask table stored in character specification designation mask memory 8 is read out based on a field designation signal 6 which instructs a field performing reading at present outputted from a character recognizing means, and outputs a character specification designation signal 9 in accordance with the field. A character specification selection table stored in character specification selection memory 11 is read out based on a category instruction signal 13, and a character specification selection signal 15 in accordance with the above category is outputted. Also, a character specification selection function part 10 takes the AND of the character specification instruction mask signal 9 and the character specification selection signal 15, and instructs the execution or interruption of matching with a reference pattern, etc., to a character recognizing means based on the above result.

Description

[Detailed description of the invention]

0001

[Industrial Application Field] The present invention relates to a character recognition method in an optical character reading device, and particularly to an optical character recognition method suitable for improving reading accuracy and reading speed when there are many types of characters to be read. The present invention relates to a character recognition method in a digital character reading device.

0002

2. Description of the Related Art Conventional optical character reading devices use a character type selection function, as described in Japanese Patent Publication No. 59-52473, in order to read documents with a large number of character types. The character types for each field were restricted to increase accuracy as much as possible. for example,
When reading the form 20 shown in FIG. 11, the process is as follows. In Figure 11, the fields are 21, 22, 2
3, and field 21 is the number “0”.
It is known that only "1" and "2" are written, only alphanumeric characters are written in field 22, and only alphabetic characters are written in field 23. Therefore, a character type table as shown in FIG. 12 is created in advance. Figure 12 shows three types of characters a and b.
, c are defined. Character type a is in category “0”
1" and "2", the character type b is the numbers "0" to "9", and the character type c is the alphabetic characters "A" to "Z".

[0003] When reading field 21, character type a
"0" to "2" can be read by using only the field 22, alphanumeric characters can be read by using the character types b and c when reading the field 22, and only alphabetic characters can be read by using the character type c when reading the field 23.

By limiting the character types for each field in this manner, it is possible to reduce the number of standard patterns to be matched, and it is possible to improve both processing speed and recognition accuracy.

[0005]

[Problems to be Solved by the Invention] In conventional optical character reading devices, in order to define character types as described above, reading control parameters are created in advance before reading a form, and a character type table and a I was making a correspondence with the field.

[0006] When the number of characters included in a character type is small, such as numbers or letters, it was possible to input from the keyboard, but when you start reading kanji, the number of characters included in the character type increases to more than 2000 characters. Therefore, it is impossible to input each character from the keyboard.

The present invention has been made in view of the problems of the prior art, and it is an object of the present invention to enable the use of a character type selection function even when a character type has a large number of characters.

[0008]

[Means for Solving the Problems] A character recognition method in an optical character reading device of the present invention is based on a field signal that indicates the field currently being read by a photoelectric conversion means, and writes data into the field on a field-by-field basis. A character type specification mask table that defines the character types that may be used is used to define a character type specification mask corresponding to the field, and furthermore, for each character type, character type selection is performed that determines whether or not the character type is used for each category. Using a table, form a character type selection signal corresponding to the above category, perform a logical product of the character type designation mask and the character type selection signal, and instruct execution, cancellation, or execution of matching with the standard pattern based on the result. It is characterized by

Further, the character type selection table may be provided for each column position, or candidate characters obtained as a result of matching may be selected based on the result of the logical product. .

The above character type selection table can be easily created from a word dictionary possessed by the optical character reading device.

[0011]

[Operation] That is, according to the present invention, the above-described character type designation mask and the character type selection signal are ANDed, and based on the result, an instruction is given to execute or cancel matching with the standard pattern, thereby eliminating unnecessary matching processing. The reading speed and reading accuracy can be improved.

Further, the character type designation mask and the character type selection signal are logically ANDed, and candidate characters are selected based on the result of the logical product for the candidate characters obtained as a result of matching. In this case, if it is determined that the category in the candidate characters is not selected as a character type based on the result of the logical product,
By lowering the candidate ranking, only the category in which the character type has been selected can occupy a high rank among the candidate characters.

[0013] In order to improve reading accuracy, optical character reading devices capable of reading kanji generally have the following features:
Equipped with a word dictionary. This word dictionary is used to compare and match words with the reading results and output the most matching word. Since characters that are not included in the word dictionary do not need to be recognized, a character type selection table created from this word dictionary is sufficient.

[0014]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will now be described in more detail with reference to embodiments shown in the accompanying drawings. First, a description will be given of a form, a character type table, a character type selection table, and a character type designation mask table used in the present invention.

As shown in FIG. 1, the form 30 has three fields, and the field 31 has numbers "0", "1", and
2” is written, only alphanumeric characters are written in field 32, and only kanji are written in field 33.However, field 33 is where the prefecture name is written, and the optical It is assumed that a prefecture name dictionary (file name F1) is prepared in advance in the character reading device.

Next, as shown in FIG. 2, a character type table is defined. Character types a to c are the same as those shown in FIG. Character type d is different from character types a to c, and file name F
1 is displayed. To indicate the file name,
An “&” mark is added at the beginning. Here, the file name F1 means a prefecture name dictionary containing prefecture names as described above.

Based on the character type table shown in FIG. 2, a character type selection table shown in FIG. 3 and a character type designation mask table shown in FIG. 5 are created. As shown in FIG. 3, the character type selection table has one-to-one correspondence with all standard patterns used in optical character reading devices. Each category is character type a, b
, c, and d, and if that category is included in the character type, the flag is set to "1". Since character type a uses only the numbers "0,""1," and "2," only "0,""1," and "2" are "1" and all other categories are "0." Regarding character types b and c, only the corresponding numeric and alphabetic categories have flags set to "1".

In the case of character type d, it is created as follows. First, turn off the flags for all categories. Next, since the character type table shown in FIG. 2 stores "&" indicating a file name and the file name F1, all word dictionaries in the prefecture name dictionary indicated by the file name F1 are loaded. Next, a list of character types is created from the characters included in the word. From this list of character types, character type a
Create a character type selection table in the same manner as in ~c. Such a character type selection table can be easily created from a dictionary in an optical character reading device.

A character type specification mask table is created by specifying which character type is to be used for each field. The character type designation mask table has a structure as shown in FIG. Since only character type a is used in field 31, a mask of "1000" is created. In field 32, only character types b and c are used, so “011
A mask of "0" is created. Since only the character type d is used in the field 33, a mask of "0001" is created. Such a character type specification mask can be easily created in an optical character reading device. I can do it.

Next, a case will be described in which the above-described character type selection table and character type designation mask are used to limit the number of matchings between a standard pattern and an unknown pattern, thereby speeding up the processing. FIG. 5 is a block diagram showing one embodiment of the present invention. In the embodiment shown in FIG. 5, 1 is a photoelectric conversion means that optically scans a form and captures the form image into a memory. 2 is a field extraction means, and photoelectric conversion means 1
Based on the field position information given from the output of and predetermined reading control parameters,
Identifies the field currently being read. This field extraction means 2 outputs a field instruction signal 6 indicating the field to be read. Reference numeral 3 denotes a single character cutting means, which cuts out characters in the field one by one. Reference numeral 4 denotes a feature extracting means, which extracts features of the character pattern output from the single character cutting means 3. Reference numeral 5 denotes a determining means, which extracts a large number of standard patterns stored in the standard pattern memory 14 one by one, performs matching with the features of the unknown pattern extracted by the feature extracting means 4, and selects the most similar standard pattern. Output the category for. 7 is a character type designation mask selection means;
Using the field designation signal 6 outputted from the field extraction means 2 as an address, the character type designation mask stored in the character type designation mask memory 8 is read out, and a character type designation mask signal 9 is output. Here, it is assumed that the character type designation mask memory 8 stores a character type designation mask table shown in FIG. 10 is a character type selection function section. This character type selection function section 10 has a standard pattern memory 14.
The character type selection memory 11 is accessed based on the category instruction signal 13 output from the character type selection memory 11. Here, it is assumed that the character type selection memory 11 stores a character type selection table shown in FIG. The character type selection function section 10 reads out the character type selection signal corresponding to the above-mentioned category from the character type selection memory 11, and then selects between the character type designation mask signal 9 outputted from the character type designation mask selection means 7 and the read character type selection signal 15. Take a logical product. If “1” is obtained as a result of the logical product, the character type selection function unit 10 outputs the selection signal 12 as “1”, and the character type selection function unit 10 outputs the selection signal 12 as “1”, and the standard pattern output from the standard pattern memory 14 and the It instructs execution of a judgment process between unknown patterns output from the feature extraction means 4. Further, if “0” is obtained as a result of the logical product, the character type selection function unit 10 outputs the selection signal 12 as “0”.
0'' and instructs the determining means 5 to stop the determination process between the standard pattern output from the standard pattern memory 14 and the unknown pattern output from the feature extracting means 4.

In the embodiment shown in FIG. 5, characters on a form are read as follows. In the following explanation,
An example in which the form 30 shown in FIG. 1 is read based on the character type table shown in FIG. 2 will be described. It is also assumed that the character type selection table is created as shown in FIG. 3, and the character type designation mask table is created as shown in FIG.

The pattern obtained by scanning the form 30 by the photoelectric conversion means 1 is stored in a memory (not shown), and then the field extraction means 2 extracts the field currently being read. The field extraction means 2 outputs a field instruction signal 6 of the extracted field. For example, if the field 31 shown in FIG. 1 is to be read, the field instruction signal 6 is f.
31 is output. Similarly, when fields 32 and 33 are to be read, f31 and f32 are output as field instruction signals 6, respectively.

The character type designation mask selection means 7 reads the character type designation mask corresponding to the field designation signal 6 from the character type designation mask memory 8 based on the field designation signal 6, and outputs the character type designation mask signal 9. In parallel with this operation, the character cutting means 3 cuts out the characters in the field one by one. After that, the feature extracting means 4 extracts features for the extracted character,
It is output to the determination means 5. Next, character type selection function section 1
0 accesses the character type selection memory 11 based on the category instruction signal 13 outputted from the standard pattern memory 14, and reads out data in the character type selection table corresponding to the category instruction signal 13 as the character type selection signal 15. Next, the character type designation mask signal 9 outputted from the character type designation mask selection means 7 and the read character type selection signal 1
Perform a logical product between 5. That is, since there is a one-to-one correspondence between the standard pattern and the character type selection table, which standard pattern should be used can be determined by taking the logical product of the character selection signal 15 and the character type designation mask signal 9. For example, for category "0", the content of the character type selection table shown in Figure 3 is "1100", and field 3
When the logical product is ANDed with the character type designation mask "1000" in 1, it becomes "1000", and since the result is other than "0", the character type selection function unit 10 outputs "1" as the selection signal 12. Therefore, in this case, the determining means 5
Category “0” output from standard pattern memory 14
The standard pattern is matched with the unknown pattern output from the feature extraction means 4. Further, for standard patterns of category "3" or higher, the above-mentioned logical product is "0000", and since the result is "0", the character type selection function section 10 outputs "0" as the selection signal 12. Therefore, in this case, the determining means 5 does not perform matching.

In the above example, by performing a logical product from beginning to end for all categories in the character type selection table, it was determined whether matching based on the standard pattern could be executed. However, even though it is known that, for example, only numbers are read, it is uneconomical to perform a logical product for all categories to determine whether or not to perform matching. In such a case, the range to be investigated can be limited by doing the following.

That is, a table similar to the character type selection table (
Create a category group selection table (named category group selection table). The category group selection table is created by calculating the logical sum of the character type selection tables of the categories included in the category group. Selecting a group of categories is done in the same way as selecting a category.
This is done by performing a logical product between the category group selection table and the character type selection mask. For example, if the logical product of the number category group and the mask is 1, it can be seen that the number is used, and if it is 0, it can be seen that the number is not used. Similarly, by deciding whether to use each category group such as alphabetic characters, kana characters, kanji characters, etc., and selecting character types only for the categories to be used, processing speed can be increased without checking consistency with other categories. FIG. 6 shows an example of the category group selection table.

FIG. 7 is a block diagram showing another embodiment of the present invention. The embodiment shown in FIG. 7 differs from the embodiment shown in FIG. It is. Here, the column position signal 16 indicates the character position within the field. The column position signal 16 uses the position information of the written character in the character type selection table described above,
Used for more efficient matching.

FIG. 8 shows an example of a character type selection table for each column position. In FIG. 9, the columns are 1 to n. Here, n is determined by the maximum length of the word. In the following example, it is assumed that the word dictionary is a prefecture name dictionary. First, a category list is created by collecting only the categories that appear in column 1 from the word dictionary. That is, the categories that appear in column 1 of prefecture names are limited to several dozen types, such as "Higashi" in Tokyo and "Kami" in Kanagawa Prefecture. Using this category list, create a character type selection table for column 1. The character type selection table K1 shown in FIG. 8 is an example of a character type selection table for column 1. Similarly, a category list is created by collecting only the categories that appear in columns 2 to n, and character type selection tables K2 to Kn for columns 2 to n are created as shown in FIG.

The character type selection tables K1 to Kn for each column position shown in FIG. 8 are used as follows. That is, a form is scanned by the photoelectric conversion means 1 shown in FIG. 8, and a form image is captured onto a memory (not shown). The field extraction means 2 extracts the field currently being read from the field position information indicated in the reading parameters, and sends the field instruction signal 6 to the character type designation mask selection means 7.
Output to. The character extraction means 3 extracts characters one by one from the form image output from the field extraction means 2 and outputs them to the feature extraction means 4. The feature extraction means 4 extracts the features of one character and outputs it to the determination means 5.

On the other hand, the character type designation mask selection means 7 accesses the character type designation mask memory 8 in response to the field instruction signal 6 outputted by the field extraction means 2. Then, the character type designation mask selection means 7 outputs a character type designation mask signal 9 read out from the character type designation mask memory 8. The character type selection function 10 selects a character type selection table (K1
~Kn), and further accesses the selected character type selection table based on the category instruction signal 13 output from the standard pattern memory 14, and selects the character type selection signal 15.
and the character type selection mask signal 9. if,
If the AND is not "0", the standard pattern is selected based on the selection signal 12, and if it is "0", it is not selected. The determining means 5 matches only the selected standard pattern with the unknown pattern output from the feature extracting means 4, and outputs the most similar category as an answer.

As is clear from the above explanation, by using the word dictionary, it is possible to know not only the use/non-use information of categories as a whole, but also the use/non-use status of categories for each column position. Furthermore, since the type of category used is different for each column position, reading accuracy can be further improved by selecting a character type for each column position.

FIG. 9 is a block diagram showing another embodiment of the present invention. The embodiment shown in FIG. 9 is different from the embodiment shown in FIG. is input to the candidate selection means 17.

In general, when performing word matching, if the word written on the form is completely included in the word dictionary, recognition can be achieved with fairly high accuracy. However, if the word dictionary is inadequate and the word written on the form does not exist in the dictionary, the word will be compared incorrectly, which may reduce recognition accuracy. Similarly, in the recognition method shown in FIG. 1, the standard pattern to be matched is skipped for categories that are not included in the word dictionary, so the correct answer is not selected as a candidate. The embodiment shown in FIG.
This applies to such cases.

That is, as shown in the embodiment of FIG. 9, the determining means 5 performs matching between unknown patterns and standard patterns for all categories, and character type selection is performed in the candidate selecting means 17 for candidate characters obtained as a result of matching. Let's do it. In FIG. 9, a predetermined number of candidate categories from the top among a plurality of candidate categories obtained as a result of matching are selected by the candidate selection means 17.
is stored in the candidate character memory 18. Here, when the candidate selection means 17 determines based on the selection signal 12 that a category among the candidate characters is not selected as a character type, the candidate selection means 17 lowers the candidate ranking so that only the category for which the character type has been selected occupies a higher rank. can do. Figure 9
In the example shown in , if a category that is not in the word dictionary is written in the field, it will not be ranked high, but it will not be deleted from the candidate characters, but will be included in the candidate characters, so the candidate character display and This can be recognized through selection processing, and misreading can be reliably prevented.

Other embodiments of the invention are shown below. If the word dictionary includes word appearance frequencies, recognition accuracy can be further improved by using the word appearance frequencies. Assuming that the degree of similarity between the unknown pattern and the standard pattern is S, and the frequency of appearance of a category is h, the evaluation function r that takes the frequency of appearance into consideration is expressed by the following Equation 1 or Equation 2.

[0035]

[Math 1]

[0036]

[Math 2]

As is clear from Equations 1 and 2, a category with a high appearance frequency h has a large evaluation value r and is easily recognized.

In order to obtain the appearance frequency h of a category from the word dictionary, proceed as follows. Assuming that category i is included in k words, the frequency of appearance h(i) is defined by the following equation 3.

[0039]

[Math 3]

[0040] When the frequency of appearance of a word is included in the word dictionary, the following procedure is performed. Category i is word 1
, word 2, . . . are included in word k, and the frequency of appearance of each word is T1, T2, . . . Tk. The appearance frequency h(i) of category i is

[0041]

[Math 4]

It is defined as:

The appearance frequency defined in this way is shown in FIG.
Put it in the character type selection table as shown in the figure below. Figure 10
In the example shown in , the appearance frequency is divided into 10 ranks from 0 to 10, with 10 being the highest appearance frequency. Also, 0
indicates that the character is not selected.

The character type selection table for the given category is compared with the character type designation mask, and the frequency of appearance of character types with "1" in the character type designation mask is determined. If the appearance frequency is 0 for all character types, that category is not used. If the appearance frequency of any character type is other than 0, that category is used. If the frequency of appearance is not 0 for multiple character types, the evaluation value r is calculated using the largest value among them.

The frequency of appearance described above can be applied not only to the embodiments shown in FIGS. 1, 5, and 7, but also to the embodiment shown in FIG. That is, in the embodiment shown in FIG. 9, the character type selection function unit 10 and the candidate selection means 17 calculate the appearance frequency for each candidate character, and change the ranking of the candidates in order of the highest appearance frequency, or change the ranking of the candidates in order of the highest appearance frequency, or By ranking candidates according to the size of r,
Characters that appear frequently can be moved to the top.

Note that although the explanation so far has been made using a word dictionary, it is possible to create a character selection table without necessarily using a word dictionary. In other words, if you use a word that is created in the same format as a word dictionary, with each character that should be used one by one, as a one-character word (a word consisting of only one character, not a word obtained from multiple characters). , the same effects as in the above embodiment can be obtained.

As is clear from the above description, according to the above embodiment, the following effects can be obtained.

Since matching is performed using only the selected standard pattern, both recognition accuracy and recognition speed are improved.

Since the character type selection table is automatically constructed from the word dictionary, no key-in operation is required to input the character type selection table.

Since the character type selection table is automatically constructed from the word dictionary, it is also possible to create a character type selection table that takes column positions into account.

[0051] If the word dictionary contains word appearance frequencies, recognition results can be obtained using the appearance frequencies for each category.

[0052] If the word dictionary is not complete, please refer to Figure 9.
As shown in Figure 2, recognition accuracy is improved by incorporating the character type selection function using selection signals into the candidate selection function.
and can be easily modified.

[0053]

According to the present invention, since it is constructed as described above, it has the following effects.

Even when there are a large number of characters in a character type, the character type selection function can be used, which has the effect of improving both recognition accuracy and recognition speed.

[Brief explanation of the drawing]

FIG. 1 is an explanatory diagram showing an example of a form.

FIG. 2 is an explanatory diagram showing an example of a character type table.

FIG. 3 is an explanatory diagram showing an example of a character type selection table.

FIG. 4 is an explanatory diagram showing an example of a character type specification mask table.

FIG. 5 is a block diagram showing an embodiment of the present invention.

FIG. 6 is an explanatory diagram showing an example of a category group selection table.

FIG. 7 is a block diagram showing another embodiment of the present invention.

FIG. 8 is an explanatory diagram showing an example of a character type selection table created for each column position.

FIG. 9 is a block diagram showing another embodiment of the present invention.

FIG. 10 is an explanatory diagram showing an example of a character type selection table that takes into account the frequency of appearance of words.

FIG. 11 is an explanatory diagram showing an example of a form.

FIG. 12 is an explanatory diagram showing an example of a character type table.

[Explanation of symbols]

1 Photoelectric conversion means 2 Field extraction means 3 Single character cutting means 4 Feature extraction means 5 Judgment means 6 Field instruction signal 7 Character type specification mask selection means 8 Character type specification mask memory 9 Character type specification mask signal 10 Character type selection function section 11 Character type selection memory 12 Selection signal 13 Category instruction signal 14 Standard pattern memory 15 Character type selection signal 16 Column position signal 17 Candidate selection means 18 Candidate character memory

Claims (3)

[Claims] Claim 1: Read characters written in fields provided on a form, identify the field on the form that is currently being read, cut out the read characters one by one, and combine the cut out characters with a standard pattern. - an optical character reading device comprising a character recognition means for matching with the character recognition means; and a standard pattern memory storing the standard pattern and sequentially reading out standard patterns corresponding to each category to the character recognition means. , based on the field instruction signal output from the character recognition means that indicates the field currently being read, the field is selected from a character type designation mask table that determines the character types that may be written in the field for each field. A character type designation mask signal indicating a character type designation mask corresponding to the character type is formed, and for each character type, a character type selection signal corresponding to the above category is generated from a character type selection table that determines whether or not the character type is used for each category. the character type designation mask signal and the character type selection signal, and based on the result, instructs the character recognition means to execute or cancel matching with the standard pattern. A character recognition method in a digital character reading device. Claim 2: Read characters written in fields provided on a form, identify the field on the form that is currently being read, cut out the read characters one by one, and determine the column position of each read character. Character recognition means that identifies and further matches cut out characters with a standard pattern, and stores the standard pattern,
In an optical character reading device equipped with a standard pattern memory that sequentially reads standard patterns corresponding to each category to the character recognition means, a field output from the character recognition means that indicates the field currently being read. Based on the instruction signal, a character type designation mask signal indicating a character type designation mask corresponding to the field is formed from a character type designation mask table that defines character types that may be written in the field for each field, and further, for each character type. , a character type selection table is provided for each column position that determines whether or not the character type is used for each category,
A character type selection signal corresponding to the category and column position is formed from the character type selection table provided corresponding to the column position, the logical product of the character type designation mask signal and the character type selection signal is performed, and based on the result, the character type selection signal is 1. A character recognition method in an optical character reading device, comprising instructing a recognition means to execute or stop matching with a standard pattern. Claim 3: Read characters written in fields provided on a form, identify the field on the form that is currently being read, cut out the read characters one by one, and combine the cut out characters with a standard pattern. A standard pattern memory that stores the standard patterns and sequentially reads out the standard patterns corresponding to each category to the character recognition means. In an optical character reading device, the type of character that may be written in each field is determined for each field based on a field instruction signal output from the character recognition means and indicating the field currently being read. A character type designation mask signal indicating a character type designation mask corresponding to the field is formed from the determined character type designation mask table, and the above character type designation mask signal is generated from the character type selection table that determines whether or not the character type is used for each category for each character type. An optical character characterized in that a character type selection signal corresponding to a category is formed, the character type designation mask signal and the character type selection signal are ANDed, and the character recognition means selects a candidate character according to the result. Character recognition method in reading device.