JPH09218921A - General document reader - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve the recognition rate of a character string part composed of English characters only and to improve the operational reliability of a device by investigating the similarity tendency of capital letters and small letters consisting of an English character string, discriminating the features of character forms of that English character string and properly judging the capital letters and small letters. SOLUTION: A word collating part 6 partitions the character recognized result for each word, collates it with a word dictionary 7 and extracts candidate words, and an output forming part 8 selects any word to be outputted out of a candidate word group. In this case, the capital letter/small letter similarity tendency of respective characters consisting of that word is investigated. Then, it is discriminated whether that word is a word to be composed of capital letters as a whole (1), a word to be composed of small letters as a whole (2) or a word to consist only its head of a capital letter and to consist the other part of small letters (3) and based on this discrimination, the capital letter/small letter can be properly selected for the respective characters of that word so that the recognition rate can be improved.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a general document reading device for recognizing and reading characters constituting a general document.

[0002]

2. Description of the Related Art Generally, in a conventional general document reading apparatus,
A document image to be read is automatically or manually divided into character, table, ruled line, figure, and photograph areas, and character recognition is performed on the character portions of the character area and table area. As a method of improving the reading accuracy of the character recognition result, word matching or linguistic processing is performed using a string of candidate character sets corresponding to the input character string and a word dictionary, and the matching word is used as the answer of the input character string. Post-treatment methods are widely used.

[0003]

However, according to the prior art with the above-described structure, when recognizing an alphabetic character by character recognition, for example, a character having the same uppercase and lowercase letters such as "P" is originally capitalized. There is a problem that it is very difficult to determine whether it is written or written in lower case, and it is difficult to determine the size of a character even if a word inquiry is performed using a word dictionary.

In view of the above problems, the present invention improves the operation reliability of the apparatus by improving the recognition rate of the character string portion which is composed of only English characters among the portions recognized as characters by the general document reading apparatus. The purpose is to increase.

[0005]

In order to achieve the above-mentioned object, the present invention relates to a case where upper and lower case characters in an alphabetic string are similar, and the upper and lower case of each alphabetic character constituting the alphabetic string is similar. By examining the tendency and determining the characteristics of the glyphs of the relevant alphabetic string, the uppercase and lowercase letters can be correctly determined.

That is, according to the present invention, in a general document reading apparatus for recognizing and reading characters forming a general document, a character area after a document image to be read is divided into areas of characters, tables, ruled lines, figures and photographs. And the character part of the front area is scanned by the scanning part, the image signal obtained by photoelectric conversion is transferred to the character recognition part, and it is judged as an alphabetic character string from the set of candidate characters formed by the character recognition part and the distance value. For the result of performing word matching or linguistic processing using a word dictionary on the character string that is generated, and selecting the matching word, refer to the upper-case / lower-case similarity tendency table of the alphabet, Judge whether it is "certainly uppercase" or "certainly lowercase" and add "certainly uppercase" in the word.
Or, investigate how many "certainly lowercase letters" are included, and from the result of the above investigation and the character type of the first character of the word, whether the word should be all lowercase letters or all uppercase letters, It is determined whether only the first letter should be capitalized and the other letters should be lowercased, or mixed case, and based on the above determination, whether each character in the word is uppercase or lowercase, and if necessary, capitalized Replaces and lowercase letters and outputs.

[0007]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing a configuration of an embodiment of the present invention. In the figure, 1 is a form to be read, 2 is an image input device, 3 is a layout analysis unit, 4 is a scanning unit, 5 is a character recognition unit, 6 is a word matching unit, and 7 is a word matching unit.
Is a word dictionary used in the word matching unit, and 8 is an output character string forming unit.

FIG. 2 is an explanatory diagram showing an example of the result of character recognition of the character string on the form 1. FIG. 3 is an explanatory diagram showing an example of a candidate word group obtained by collating the character recognition result of FIG. 2 with a word dictionary. FIG. 4 is an explanatory diagram showing a result when an output character string is selected from the candidate word group of FIG. 3 according to the conventional technique.

FIG. 5 is an explanatory diagram showing an example of a table for determining the tendency of uppercase letters and lowercase letters of each alphabetic character in this embodiment. FIG. 6 refers to the table of FIG.
5 is an explanatory diagram showing an example of determining the character type of the output word in FIG. FIG. 7 is an explanatory diagram showing an example of a table for determining the feature of the word from the similar character tendency of the characters forming each word in the present embodiment.

FIG. 8 is an explanatory view showing an example in which an output character string is selected from the candidate word group of FIG. 3 according to the present embodiment. The operation of this embodiment will be described below. The document image input from the image input device of FIG. 1 is automatically or manually operated by the layout analysis unit 2 of FIG.
Divide into figure and photo areas.

The scanning unit scans the character portion in the area determined by the layout analysis unit and the character area in the front area, and transfers the image signal obtained by photoelectric conversion to the character recognition unit.
The character recognition unit calculates the distance between the glyph of the input character and the standard glyph of each character, and forms a set of candidate characters and distance values arranged in ascending order of distance (that is, in the order of similar glyphs),
Output to the word matching unit. FIG. 2 shows a recognition target and an example of character recognition.

In the word collating unit, the input character recognition result is delimited at an appropriate position, and a word dictionary is searched for from a character string obtained by combining the above-mentioned set of candidate characters and distance values into a word dictionary. Extract what exists in. When the cut-out character string is an English character string, the word dictionary uses the English word dictionary. All registered words in the word dictionary are written in upper case, and the character recognition result is once converted to upper case before being collated with the registered content in the word dictionary. FIG.
Is an example in which the character recognition result is divided into words and collated with a word dictionary to extract candidate words.

The output forming unit selects a word to be output from the candidate word group extracted by the word matching unit. FIG.
Shows an example in which the words with the highest average candidate rank are selected with priority, that is, the selection results according to the prior art are shown. At the time of word matching in FIG. 3, the character recognition result, which has been converted into all uppercase letters, is converted into the character type at the time of character recognition again in the output forming unit. When uppercase and lowercase letters of the same English appear in the candidate character at the character position, the upper candidate character is output. As an example of this, “o” and “O” are listed as the candidate characters at the character position of “o” of “ToM” in FIG. 4, but since “o” is the higher recognized character, “O” is recognized. Output as "o" instead of "."

In the word matching using the word dictionary, it is possible to correct the arrangement of letters, but it is not possible to judge whether "o" or "O" is correct. Therefore, the judgment at the time of character recognition will be given priority. In the present invention, for the alphabetic character string from the character recognition result, after extracting the candidate word in the output forming unit, the uppercase / lowercase similarity tendency of each character forming the word is examined, and the word is
It is determined whether the word should be all uppercase letters, all lowercase letters, or only uppercase letters and the other parts should be lowercase letters. The correct recognition of uppercase / lowercase can improve the recognition rate.

The output character selection procedure according to the present embodiment will be described below with reference to FIGS. First, for each character forming the word, it is determined whether the character is certainly an uppercase letter or certainly a lowercase letter by referring to the uppercase / lowercase similarity tendency table shown in FIG. If the character belongs to a dissimilar character, it is determined that the character is definitely uppercase or lowercase. If the character belongs to a similar character, it is determined that the character may have mistakenly recognized uppercase and lowercase letters. FIG. 6 shows the output formation result of FIG.
This is an example of determining a character that can be determined to be an uppercase letter or a lowercase letter.

Next, it is examined how many characters each of the words can be surely determined to be uppercase letters or lowercase letters. However, since the probability that the first letter of a word is written in capital letters is higher than other letter positions, it is excluded from the check target. The “total” column of each table in FIG. 6 is the total number of characters judged to be definitely uppercase or lowercase for each word.

Next, referring to the "total" in FIG. 6 and the first character of each word in the output result in FIG. 4, the feature of each word is determined using the table for determining the feature of the word in FIG. , If there is a character in the word that is against the feature, correct it. FIG.
Is an example of the output character string determined from FIGS. 4, 6, and 7. In "HapPy" of FIG. 4, the first character "H" is excluded from the check target as described above, and "apPy" is checked. Here, as shown in FIG. 6 (1), there is one "certainly small letter" in "apPy", and there is no "certainly uppercase letter". In addition, since the first character "H" that is not a check target is determined to be a capital letter, it is determined from the table in FIG. 7 that this word is a word written only in the capital letter and the other letters in lower case. Then, since "P" of "HapPy" is against this feature, "P" is corrected to "p" and "Happ"
y ”is output.

In "bIrThdaY" of FIG. 4, the first character "b" is not checked as described above, and "IrT
Check "hdaY". Here, as shown in Fig. 6 (2), "certainly small letter" is 4 in "IrThdaY".
There is no "certainly uppercase". In addition, since the first character "b" that is not a check target is determined to be a lowercase letter, it is determined from the table in FIG. 7 that this word is written in all lowercase letters. Then, "bIrThdaY"
Since "I, T, Y" of this is contrary to this feature, "I, T, Y"
Correct "Y" to "i, t, y" and change to "birthday"
Is output.

In "tO" of FIG. 4, the leading character "t" is not checked as described above, and "o" is checked. Here, as shown in FIG. 6 (3), “o” is neither “certainly lowercase” nor “certainly uppercase”. In addition, since the first character “t” that is not a check target is determined to be a lowercase letter, it is determined from the table of FIG. 7 that this word is written in all lowercase letters. Then, since "O" of "tO" is contrary to this feature, "O" is corrected to "o" and "t" is corrected.
Output as "O".

In "ToM" of FIG. 4, the leading character "T" is excluded from the check target as described above, and "oM" is checked. Here, as shown in FIG. 6 (4), "certainly lowercase letters" do not exist in "oM", and there is one "certainly uppercase letters". In addition, since the first letter “T” that is not checked is determined to be a capital letter, it is determined from the table of FIG. 7 that this word is written in all capital letters. Then
Since "o" of "ToM" is contrary to this feature, "o" is corrected to "O" and output as "TOM".

In the above embodiment, the device for recognizing the image read character is shown, but the present invention is also applicable to the device for recognizing the pen input character.

[0022]

As described above in detail, according to the present invention, in the general document reading device for recognizing and reading the characters constituting the general document, the document image to be read is displayed as characters.
The scanning unit scans the character area and the character area of the table area after dividing into the table, ruled line, figure, and photograph areas, transfers the image signal obtained by photoelectric conversion to the character recognition unit, and forms it by the character recognition unit. The result of selecting a matching word by performing word matching or linguistic processing using a word dictionary on the character string that is determined to be an alphabetic character string from the set of candidate characters and distance values that are selected. Refer to the lowercase similarity tendency table to determine whether each character that constitutes the word is "certainly uppercase" or "certainly lowercase", and which of the words is "certainly uppercase" or "certainly lowercase" Whether the word should be all lowercase letters or all uppercase letters, or only the first letter should be capitalized from the result of the above survey and the character type of the first letter of the word It judges whether it should be lowercase or mixed case, and based on the above judgment, it judges whether each character in the word is uppercase or lowercase, and switches the uppercase and lowercase as necessary and outputs it. , For letters with similar upper and lower case letters in an alphabetic string, by examining the similarities of the uppercase and lowercase letters of the letters that make up the alphabetic string, and distinguishing the glyph features of the alphabetic string, Can correctly detect lowercase letters.

As a result, it is possible to improve the recognition rate of the character string portion composed of only alphabetic characters, which has the effect of increasing the operational reliability of the device.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of an embodiment of the present invention.

FIG. 2 is an explanatory diagram showing an example of a result of character recognition of a character string on a form.

FIG. 3 is an explanatory diagram showing an example of a candidate word group obtained by collating the character recognition result of FIG. 2 with a word dictionary.

FIG. 4 is an explanatory diagram showing a result when an output character string is selected from the candidate word group of FIG. 3 according to a conventional technique.

FIG. 5 is an explanatory diagram showing an example of a table for determining similarity between uppercase letters and lowercase letters of each English character according to the embodiment of this invention.

6 is an explanatory diagram showing an example of determining the character type of the output word in FIG. 4 with reference to the table in FIG.

FIG. 7 is an explanatory diagram showing an example of a table for determining the characteristics of a word from the similar character tendency of the characters forming each word according to the embodiment of the present invention.

FIG. 8 is an explanatory diagram showing an example in which an output character string is selected from the candidate word group in FIG. 3 according to the embodiment of the present invention.

[Explanation of symbols]

 1 form 2 image input device 3 layout analysis unit 4 scanning unit 5 character recognition unit 6 word matching unit 7 word dictionary 8 output character string formation unit

Claims (1)

[Claims] 1. A general document reading device for recognizing and reading characters constituting a general document, wherein a document image to be read is divided into character, table, ruled line, figure, and photo areas, A character string that is scanned by the scanning unit, transfers the image signal obtained by photoelectric conversion to the character recognition unit, and is a character string that is determined to be an alphabetic character string from the set of candidate characters formed by the character recognition unit and the distance value. For word matching or linguistic processing using a word dictionary and selecting matching words,
By referring to the lower case similarity tendency table, it is judged whether each character forming the word is "certainly uppercase" or "certainly lowercase", and which of the words is "certainly uppercase" or "certainly lowercase" is determined. Whether the word should be all lowercase letters or all uppercase letters, or only the first letter is uppercase and the others are lowercase, based on the result of the above survey and the character type of the first letter of the word It is necessary to judge whether each character in the word is upper case or lower case based on the above judgment, and switch the upper case and lower case, and output it. Characteristic general document reading device.