JP2743995B2 - Character reader - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Object of the Invention] (Industrial application field) The present invention relates to a character reading apparatus suitable for reading, for example, characters handwritten on a form. (Prior Art) Conventionally, optical character reading devices have been operated in various business processing fields, and in particular, handwritten katakana,
2. Description of the Related Art Handwritten character reading devices for reading characters such as English characters have become widespread. Normally, such a character reading device reads characters such as katakana and alphanumeric characters written in a character entry frame on a form. By the way, as shown in FIG. 8, when the voiced sound 3 and the semi-voiced sound 4 are to be entered in the character entry frame 2 on the form 1, they are converted into two character parts of the clear sound 5, the voiced dot 6 and the semi-voiced voice 7. It is general to divide and enter in separate character entry boxes 2 respectively. In other words, this is a rule for obtaining a final reading result using the knowledge base by matching the comparison unit of the voiced sound and the semi-voiced sound in the reference word stored in the knowledge base. Then, the characters written on the form 1 are read and recognized as an image for each character entry box 2 by the character recognition device, and the obtained character candidates are stored in the knowledge base by the post-processing device. The final reading result is obtained by comparing with the reference word. However, sometimes, due to a writing error or the like, the voiced sound 3 or the semi-voiced sound 4 in one character unit is written in each character entry frame 2, and as a result, the reading of the character pattern may be impossible. As a countermeasure in such a case, one voiced sound 3 or half voiced sound 4 is added to one reference word together with normal data.
A method is conceivable in which a knowledge base that stores multiplexed data by registering data composed in units of characters is considered. In the case of using such a method, for example, two or more voice sounds 4 such as “Kabushiki Geisha” are used. In addition, for a word having a semi-voiced sound 5, four types of multiple registration are required. Therefore, there is a problem that not only a knowledge base having a larger storage capacity is required, but also post-processing is complicated and efficiency is reduced. (Problems to be Solved by the Invention) The present invention is intended to solve such a problem. Characters having different comparison units output from recognition means can be output without registering a plurality of reference words in a knowledge dictionary. It is an object of the present invention to provide a character reading device capable of determining the validity of a character string having the character string and preventing a decrease in reading efficiency. [Structure of the Invention] (Means for Solving the Problems) In order to achieve the above object, the present invention provides a recognition unit that reads an image of a character written on a form and performs character recognition, and a recognition unit that performs the recognition. A knowledge dictionary in which reference words to be compared when judging the validity of a plurality of candidate character strings obtained as recognition results output by the means are stored in advance, and a plurality of candidate character strings obtained by the recognition means. When any of the candidate characters constituting the first candidate character string is a voiced sound or a semi-voiced sound, the plurality of voiced or semi-voiced sound candidate characters corresponding to positions in the first candidate character string are And a post-processing means for dividing the candidate character of the candidate character string into two character parts of clear sound and voiced sound or semi-voiced sound, and matching the reference character of the knowledge dictionary in units of the divided character parts to obtain a result. It is characterized by comprising. (Operation) In the character reading device of the present invention, the editing processing unit may be configured so that the comparison unit in the character string output from the recognition unit matches the comparison unit in the reference word stored in the recognition dictionary. Because a part of one of the character parts of the reference word is edited, the validity of a character string having a different unit of comparison output from the recognizing unit can be determined without registering a plurality of reference words in the knowledge dictionary. Can be performed, and a decrease in reading efficiency can be prevented. Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. FIG. 1 is a block diagram showing a configuration of a character reading device according to one embodiment of the present invention. In the figure, reference numeral 11 denotes a form having a large number of character entry frames arranged according to a predetermined rule. Reference numeral 12 denotes a reading unit that optically reads an image of a character for each predetermined area handwritten in the character entry frame on the form 11 to obtain a character pattern, and 13 denotes a provisional character pattern obtained by the reading unit 12. A storage unit for temporarily storing characters, a cutout unit for sequentially extracting character patterns stored in the storage unit 13 one by one, and 15 a cutout unit
The recognition unit obtains character candidates by recognizing the character pattern in units of one character output from 14 according to a predetermined recognition rule. That is, a character recognition device 16 is configured to read one-by-one characters handwritten on the form 11 from the reading unit 12, the storage unit 13, the extraction unit 14, and the recognition unit 15 and output a matrix of character candidates. . Reference numeral 17 denotes a knowledge base in which word information is stored in advance. The information indicating the voiced sound and the semi-voiced sound in the reference word stored in the knowledge base 17 is a code indicating the voiced sound, the voiced voice and the voiced voice in each character unit. Are stored separately. Further, 18 performs the matching by a predetermined method between the matrix of the character candidates output from the character recognition device 16 and the word information described in the knowledge base 17 to thereby obtain the most out of the matrix of the character candidates. This is a post-processing device for selecting a character candidate determined to be appropriate. The post-processing device 18 may be configured to select one of the character candidate and the reference word so that the number of character parts, which is a unit of comparison between the voiced and semi-voiced sound in the character candidate and the voiced and semi-voiced sound in the reference word, match. It has a function of editing a part of the character parts of one of the voiced and semi-voiced sounds. Reference numeral 19 denotes a control device for controlling each device. Next, the operation of the character reading device of this embodiment will be described. First, a character handwritten in the character entry frame on the form 11 is optically read by the reading unit 12 in the character recognition device 16 and converted into a character pattern. Thereafter, the character pattern is stored in the storage unit 13, extracted by the extracting unit 14 one by one, and output to the recognizing unit 15. The recognizing unit 15 recognizes the character pattern in units of one character output from the extracting unit 14 based on a predetermined recognition rule, and outputs a matrix including some character candidates to the post-processing device 18. The post-processing device 18 compares the matrix of the character candidates output from the character recognition device 16 with the reference word stored in the recognition base 17 and performs matching to determine the most appropriate among the character candidates. A character string composed of the selected character candidates is selected as a final reading result. Thereafter, the reading result is output to an output device (not shown) via the control device 19, and the reading of characters is completed. In the above operation, as shown in FIG.
For example, when the characters including “22, 22” are entered, the following data editing process is performed. In the knowledge base 17, as shown in FIG. 3, information indicating the voiced sound 21 and the semi-voiced sound 22 of "c, ゜, su, ゛, lu" are respectively stored in the knowledge base 17 as the clear sound 23 and the turbid point 24, as shown in FIG. And a reference word that is divided into semi-voiced points 25 is stored. First, as shown in FIG. 4, the character recognizing device 16 obtains some character candidates including the voiced sound 21 or the semi-voiced sound 22 in the first and second columns, respectively. Thereafter, using any of the first to third processing methods described below, the comparison unit of the character candidate and the voiced sound 21 and the semi-voiced sound 22 in the reference word stored in the knowledge base 17 match. Thus, the data of the voiced sound 21 and the semi-voiced sound 22 of either the character candidate or the reference word are edited. The first processing method is to synthesize a voiced sound 21 and a semi-voiced sound 22 composed of two character units in the reference word stored in the knowledge base 17 and to compare a voiced sound 21 and a half of a character unit which is a comparison unit equal to the character candidate. The main purpose is to make the voice sound 22. FIG. 5 is a diagram showing such a first processing method, in which 26 is a word when only the half-voiced sound 22 is synthesized, 27 is a word when only the half-voiced sound 21 is synthesized, and 28 is a word when the half-voiced sound 22 and 21 are respectively synthesized. The words when combined are shown. That is, when performing the matching, the reference words 26, 27, and 28 of the combination in all such cases are prepared, and the reference words 26, 27, and 28 are matched with the matrix of the character candidates to finally determine the matching. A reading result is obtained. As shown in FIG. 4, the second processing method corresponding to the present invention employs a first candidate “Puzzle” obtained by the character recognition device 16,
Among a plurality of candidates (candidate character strings) such as a second candidate “puzzle”, a third candidate “Hadey”, and a fourth candidate “vejiba”, among characters (candidate characters) constituting the first candidate “puzzle”,
When any of the characters is the voiced sound 21 or the semi-voiced sound 22, the second and subsequent characters corresponding to the character positions of the respective voiced sound 21 or the half-voiced sound 22 are also determined to be the voiced sound 21 or the half-voiced sound 22. . In this case, since the first candidate “puzzle” includes the character “pa” of the semi-voiced sound 22 and the character “z” of the voiced sound 21, a sequence of the character “pa” and the character “ The two candidate characters “ba” and “su”, the third candidate character “ha” and the character “de”, and the fourth candidate character “pe” and the character “ji” are voiced sound 21 or semi-voiced sound 22 On the other hand, the column corresponding to the first candidate character “レ”, that is, the second candidate character “ル”, the third candidate character “イ”, and the fourth candidate character “バ” have only a clear tone. Is determined. Then, as shown in FIG. 6, the candidate character in the row (position) determined to be the voiced sound 21 or the semi-voiced sound 22 is designated as a clear sound.
This is due to the division into two characters, 23 and 24. Then, a matrix of character candidates including the voiced sound 21 and the semi-voiced sound 22 divided into two characters is matched with the word information stored in the knowledge base 17 to obtain a final reading result. In the third processing method, matrices for all cases are created by assuming whether or not a voiced point 24 or a semi-voiced point 25 exists for each column of character candidates obtained by the character recognition device 16, and this matrix. And the knowledge base 17 are matched. That is, FIG. 7 is a diagram showing the third processing method.
FIG. 7A is a matrix when it is assumed that the character candidates in all columns are Kiyone 23, and FIG.
Matrix when assuming that only the column is the voiced sound 21 or the semi-voiced sound 22, FIG. 7 (c) shows the first column and the second column.
9 is a matrix when it is assumed that a column is a voiced sound 21 or a semi-voiced sound 22. As shown in the figure, the character candidates in the column assumed to be voiced sound 21 or semi-voiced sound 22 are divided into clear voice 23 and voiced voice 24 or semi-voiced voice 25, respectively, so that matching with the knowledge base 17 is possible. I have. Then, by matching these matrices with the reference words stored in the knowledge base 17, a final reading result can be obtained. Thus, according to the character reading device of this embodiment, the comparison unit of the voiced sound 21 and the half-voiced sound 22 in the character candidate obtained by the character recognition device 16 and the voiced sound 21 and the half-voiced sound in the reference word stored in the knowledge base 17 Since the function of editing the data of the voiced sound 21 and the semi-voiced sound 22 of one of the character candidate and the word information is provided so that the comparison unit of 22 is matched, multiple words are registered in the knowledge base 17. , The voice sound of each character recognized by the character recognition device 16
It is possible to perform matching of character candidates including 21 and the semi-voiced sound 22, and it is possible to prevent the knowledge base 17 from being enlarged and to prevent a decrease in efficiency in post-processing. By the way, in the present embodiment, the information indicating the voiced sound 21 and the semi-voiced sound 22 are respectively converted into the clear sound 23 and the voiced point 24 in one character unit.
And the knowledge base 17 divided and stored into codes indicating the semi-voiced sound points 25 has been described. However, the present invention is not limited to this. The same can be applied to the case where the knowledge base 17 in which is stored is used. [Effects of the Invention] As is apparent from the above description, according to the present invention, the validity of a character string having a different unit of comparison output from a recognition unit without registering a plurality of reference words in a knowledge dictionary. Can be determined, and a decrease in reading efficiency can be prevented.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram for explaining the configuration of a character reading device according to an embodiment of the present invention, and FIG. 2 is a plane showing characters handwritten in a character entry frame on a form. FIG. 3 is a diagram showing reference words stored in a knowledge base, FIG. 4 is a diagram showing character candidates obtained by a character recognition device, and FIG. 5 is a diagram showing a character reading device according to an embodiment of the present invention. FIG. 6 is a view showing a synthesized reference word for explaining a first editing method, FIG. 6 is a view showing a matrix of divided character candidates for explaining the second editing method, and FIG. FIG. 9 is a diagram showing a matrix of character candidates for explaining the third editing method, and FIG. 8 is a plan view for explaining a method of writing characters on a form. 11 Form 16 Character recognition device 17 Knowledge base 18 Post-processing device 19 Post-processing device 21 Control device 21 Dense sound 23 Semi-dense sound 23 Clean sound 24 Dense point 25

   ────────────────────────────────────────────────── ─── Continuation of front page    (56) References JP-A-61-229177 (JP, A)

Claims (1)

(57) [Claims] A recognition unit that reads an image of a character written on a form and performs character recognition, and a reference word to be compared when determining the validity of a plurality of candidate character strings obtained as a recognition result output by the recognition unit Among a plurality of candidate character strings obtained by the recognition means,
When any of the candidate characters constituting the first candidate character string is a voiced sound or a semi-voiced sound, the plurality of candidate characters corresponding to the position of the voiced or semi-voiced sound candidate character in the first candidate character string A post-processing means for dividing a candidate character in a row into two character parts of clear sound and voiced sound or semi-voiced sound, and matching the divided character parts with a reference word of the knowledge dictionary to obtain a result; A character reading device characterized by the above-mentioned.