JPH05282484A - Optical character reader - Google Patents

Abstract

PURPOSE:To provide the superior optical character reader which can read characters at a high recognition ratio by performing some postprocessing for improving the recognition ratio of characters without using a 'word' dictionary. CONSTITUTION:An optical system 2 reads characters entered into a document 1. In general, a character recognizing means 8 reads KANA (Japanese syllabary) characters and 'Roman characters'. An area of plural successive characters is specified by using an FC table 7, etc., and data specifying the kind of the read characters in this area is stored; and then, the certain postprocessing is specified for a group of areas consisting of KANA character areas and Roman character areas specified as areas having the same pronunciation. A KANA- Roman character matching processing part 13 collates and compares the KANA character recognition result with Roman character recognition result as to the group of specified areas to perform the certain postprocessing.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an optical character reader for optically reading a character written on a form and outputting the character code to a file or a printing machine.

[0002]

2. Description of the Related Art Conventionally, as an optical character reader,
There are devices having a configuration as shown in FIG. This device will be described below.

First, in FIG. 5, reference numeral 1 is a form to be read, and characters to be read are filled in by a method such as handwriting or printing. The optical character reading device for reading the form 1 is roughly divided into a pre-processing unit for inputting and storing data and a post-processing unit for recognizing the stored data. There is.

Among these, the preprocessing unit scans the characters on the form 1 to obtain an image data in the form of an optical signal, an optical system 2, a photoelectric conversion unit 3 for converting the optical signal into an electric signal, and a digital electric signal. A / D converter 4 for converting into a signal, and an image memory 5 for storing image data converted into a digital signal
Are sequentially connected and configured.

On the other hand, the post-processing section includes a character image cutout means 6 for sequentially cutting out character image data of each character from the form image data stored in the image memory 5, preset form format data (hereinafter, (Referred to as FC data)
Is registered and is connected to the character image cutting means 6
The C table 7, the character recognition means 8 for recognizing the character image data cut out by the character image cutout means 6, and the standard character pattern image data of the character to be recognized are stored in advance together with the corresponding character code. A standard character pattern storage unit 9 connected to the means 8 and a recognition result editing means 10 for editing the recognition result of the character image data.

The operation of the optical character reader having the above construction is as follows. First, the characters written on the form 1 are scanned by the optical system 2 to obtain an optical signal. Then, this optical signal is sent to the photoelectric conversion unit 3 and converted into an electric signal. Further, this electric signal is converted into a digital signal by the A / D converter 4 and stored in the image memory 5 as image data.

Next, post-processing is performed using the form image data stored in the image memory 5. Typically, when the accumulation of the image data of one form in the image memory 5 is completed, the character image cutout unit 6 extracts the characters from the accumulated form image data according to the FC data in the FC table 7. The position is calculated, and character image data of each character is sequentially cut out and sent to the character recognition means 8. The character recognizing means 8 compares and collates the standard character pattern image data in the standard character pattern storage unit 9 with the input character image data, and outputs the character code of the character pattern determined to be the closest. After this recognition, the recognition result editing means 10 edits the recognition result for each character to obtain the final recognition result of the form.

[0008]

However, in the conventional optical character reading device as described above, since the recognition processing is individually performed for each individual character, it is difficult to obtain a high recognition rate. It was As a method other than such individual character recognition method, a method of reading a plurality of consecutive characters and matching them with a character string registered in a dictionary as a so-called "word" and outputting the closest word as a result is also available. Although this method exists, this method can only recognize words registered in the dictionary, and the matching process will take longer if the size of the dictionary increases, or a large external memory will be required to hold the dictionary. Space is needed,
There was a problem such as.

The present invention has been proposed in order to solve the problems of the prior art as described above, and its purpose is to:
Providing an excellent optical character reader that can read characters with a high recognition rate by performing some post-processing that can improve the character recognition rate without using a "word" dictionary It is to be.

[0010]

The present invention focuses on the case where the same reading data is entered in two types of "kana" and "romaji". In such a case, the same reading is performed. It is configured to perform the matching process of "kana" and "romaji".

That is, the optical character reader according to the present invention is an optical character reader for optically reading characters, in which a kana character recognizing means for reading a "kana" character and a roman character recognizing means for reading a "roman character" are provided. An area consisting of multiple consecutive characters is specified, and the recognition character type specification means that stores the data that specifies the read character type of this area, and the kana character area specified by the recognition character type specification means to read the "kana" character. , To perform a certain post-processing on a set of areas that are designated as the same reading data as this Kana character area and that are designated as "Roman character" reading. The kana character recognizing means selects the post-processing specifying means and the kana character area and the roman character area forming the set of areas specified by the post-processing specifying means. It is characterized in that a Roman character matching process section - and recognized Kana character recognition result, by matching comparing the recognized Romanized recognition result Kana performing certain post-processing by the Roman alphabet recognition means.

More specifically, an FC table that also serves as the recognition character designating means and the post-processing designating means is used. In this FC table, FC data corresponding to the form to be read is written for each character. Or a region of continuous characters is registered as a unit, and as a part of this region information, a set of Kana and Romaji regions designated to have the same reading is registered. It is possible to store command information for causing the kana-romaji matching processing section to process the recognition result in the set of the character and the romaji area.

Further, the kana character recognizing means and the roman character recognizing means sort the characters having the highest degree of similarity by performing similarity degree calculation, and output a fixed number of plural candidates for each character as a recognition result. Can be configured. In this case, the Kana-Roman character matching processing unit specifies the same Kana character among the Kana candidate and the Roman character candidate output from the Kana character recognizing unit and the Roman character recognizing unit by the area information of the recognition character designating unit. The Kana candidate and the Roman character candidate corresponding to the pair of the and the Roman character area are used, the Kana candidate character string and the Roman character candidate string are generated according to the designation of the post-processing designating means, and the matching result is processed to edit the recognition result. It is desirable to be configured.

Typically, the Kana-Roman character matching processing unit generates a Kana candidate character string from the Kana candidate output by the Kana character recognition means, generates a Roman character candidate string from this Kana candidate character string, and generates a Kana candidate. The evaluation points of the Kana candidate character strings are calculated by matching the character strings and Kana candidates, and the evaluation points of the Roman character candidate strings are calculated by matching the Roman character candidate strings and Roman character candidates, and the Kana character recognition means outputs the Kana character recognition means. The evaluation points are assigned in the order of candidates, and the evaluation points for each set of candidate strings of Kana and Roman characters are calculated,
Finally, it is desirable that the set having the highest evaluation score is output as the recognition result.

[0015]

In the optical character reading device of the present invention having the above-mentioned structure, the "kana" and "romaji" written in the same reading are converted into kana character recognizing means and The Romaji recognition unit individually reads out the recognition results and outputs the recognition results, and the Kana-Romaji matching processing unit determines a Kana character recognition result for the Kana character region and the Romaji region that form the set of regions designated by the post-processing designation unit. A certain amount of post-processing can be performed by collating and comparing with the Roman character recognition result. That is, in the present invention, by performing the matching process of the kana character recognition result and the roman character recognition result for a set of areas designated as having the same reading, the recognition rate of characters in these areas can be improved. ..

[0016]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the optical character reader according to the present invention will be described with reference to FIGS. The same parts as those in the conventional example shown in FIG. 5 are designated by the same reference numerals.

In FIG. 1, the form 1 has an OCR reading entry frame printed in dropout color in advance as shown in FIG. On this form 1, O
Characters to be read by the optical character reading device of this embodiment are written by a method such as handwriting or printing according to the CR entry frame. The optical character reading device for reading the form 1 is roughly divided into a preprocessing unit for inputting and storing data and a device for recognizing the stored data, as in the above-described conventional example. It is composed of a post-processing unit.

Of these, the preprocessing unit is first shown in FIG.
As in the conventional example shown in FIG. 1, an optical system 2 that scans characters on the form 1 to obtain image data in the form of optical signals, a photoelectric conversion unit 3 that converts the optical signals into electric signals, and converts the electric signals into digital signals. The A / D conversion unit 4 and the image memory 5 that stores the image data converted into digital signals are sequentially connected. In this case, as the photoelectric conversion unit 3, for example,
A CCD sensor or the like is used. Further, the A / D conversion unit 4 performs sampling and quantization, for example, 8 lines /
It is configured to obtain mm binary image data. In addition to the above-described configuration, the preprocessing unit of the present embodiment includes a form conveyance unit 11 for conveying the form 1 and a conveyance control unit 12 that outputs a control signal to control the form conveyance unit 11. Have

On the other hand, as the post-processing section, first, similarly to the conventional example shown in FIG. 5, the OCR stored in the image memory 5 is used.
The character image data is obtained from the image data, and the character image cutout unit 6 sequentially cuts out the character image data for each character, and the FC table 7 connected to the character image cutout unit 6 in which preset FC data is registered. Character recognition means 8 for recognizing the character image data cut out by the character image cutout means 6, standard character pattern image data of the character to be recognized, and a corresponding character code are stored in advance.
A standard character pattern storage unit 9 connected to the character recognition unit 8 and a recognition result editing unit 10 for editing the recognition result of the character image data are provided.

In this case, the FC table 7 corresponds to a unit that serves as both the recognized character designating unit and the post-processing designating unit in the present invention. That is, in this FC table 7, as a plurality of FC data corresponding to various forms to be read, for example, the position, size, character frame pitch, and character type of the OCR character entry frame with the upper left corner of the form as the origin. Data such as is registered for each character or for each field in units of continuous characters. Furthermore, as a part of such field information, a pair of Kana-character and Romaji fields designated as having the same reading is registered. Command information for causing the Romaji matching processing unit 13 to perform processing is stored.

The character recognizing means 8 includes at least two kinds of characters to be read, that is, "kana" character and "roman character", and the recognition functioning as both kana character recognizing means and roman character recognizing means in the present invention. It corresponds to the means. That is, the character recognizing means 8 is configured to calculate the similarity using, for example, the composite similarity method, sort from the character having the highest similarity, and output three candidates for each character. ..

The post-processing unit of this embodiment has a Kana-Roman character matching processing unit 13 between the character recognizing unit 8 and the recognition result editing unit 10 in addition to the above configuration. The Kana-Roman character matching processing unit 13 uses the recognition result of the character recognition unit 8 and the field information in the FC table 7 as a Kana candidate character string of the Kana-Roman character field pair designated as the same reading. It is configured to generate a Roman character candidate sequence, perform a matching process on these, and edit the recognition result.

The operation of the optical character reader of the present embodiment having the above construction is as follows. First, the form conveying unit 11 is controlled by a control signal output from the conveyance control unit 12 to convey the form 1. In this conveyance state, the optical system 2 scans the characters written on the form 1 for each line of the printed OCR character entry frame to obtain an optical signal. Then, this optical signal is converted into an electric signal by the photoelectric conversion unit 3, and further subjected to sampling and quantization by the A / D conversion unit 4 to obtain binary image data of 8 lines / mm, and this image data is obtained. Are stored in the image memory 5.

Next, from the OCR image data stored in the image memory 5, the character image cutting means 6 cuts out character image data for each character. That is, the position, size, pitch of the character frame of the OCR character entry frame registered in the FC table 7 by the character image cutting means 6 for each character or for each field in units of continuous characters,
The character image data for each character is cut out according to the FC data such as the character type, and the cut-out character image data is sent to the character recognition means 8.

Then, in the character recognition means 8, the sent character image data is stored in the standard character pattern storage section 9
The character code of the character pattern determined to be the closest to the standard character pattern image data is output. That is, the similarity is calculated using the composite similarity method, the character having the highest similarity is sorted, and three candidates are output for each character. The recognition result is, for example, a field unit and a kana-romaji character is output. It is sent to the matching processing unit 13.

Next, the Kana-Romaji matching processing unit 13
The received recognition result is collated with the field information in the FC table 7, and the kana character and the roman character are used by using the recognition result of the corresponding field for the field for which the matching process of the kana character and the roman character is designated. Edit the recognition result of. An example of this editing work will be described in more detail below with reference to FIGS. 3 and 4. Here, FIG. 3 is a diagram showing an example of the recognition result by the character recognizing means 8. When the kana character "Takahashi" and the roman character "Takahashi" are read, each kana character image data is read. It is a table (A) showing Kana characters for each given 3 candidates, and a table (B) showing Romaji for each 3 candidates given for each Roman character image data. Further, FIG. 4 is a flowchart showing an example of a processing flow of the Kana-Roman character matching processing unit 13 set to edit the recognition result.

That is, in editing the recognition result shown in FIG. 3, first, steps 1 to 6 as shown in FIG.
Then, the evaluation score per character of the first kana-Roman character candidate string pair is calculated and stored. Specifically, as shown in step 1, only the first candidate is selected from the kana candidates to generate the first kana candidate character string.
In this case, in the example of FIG. 3, it is “Takahatsu”. next,
As shown in step 2, the kana candidate character string "Takahatsu" is converted into Roman characters, and the first Roman character candidate string "TA" is displayed.
A roman character candidate string "KAHATSU" is generated. As a result, the first kana-Roman character candidate string pair “Takahatsu”-“TAKAHATSU” is obtained.

Subsequently, as shown in steps 3 and 4, the kana candidate character string "Takahatsu" and the Roman character candidate string "Takahats" which form the first kana-Roman character candidate string pair "Takahatsu"-"TAKAHATSU" generated.
U ”and the recognition results of the Kana candidate and the Roman character candidate shown in FIG. 3 are compared and collated to calculate the evaluation points of the Kana candidate character string and the Roman character candidate string. For example, if the character exists in the recognition result candidates for Kana and Roman characters respectively, the first candidate has 5 points, the second candidate has 3 points, the third candidate has 1 point, and there is no further candidate. Has
The total sum of all characters is calculated by allocating points such as 0, and the evaluation points of the kana candidate character string and the roman character candidate string are obtained.

In this case, in the example of FIG. 3, the four characters forming the kana candidate character string "Takahatsu" are all first candidates, and 5 points are assigned to each of them, so that the evaluation score as a column is , [5 points × 4 = 20 points]. Also, among the nine characters that make up the Roman character candidate string “TAKAHATSU”, the first character “T” is the second candidate, so three points are assigned, and the subsequent five characters “AKAHA” are all the first candidates. Therefore, 5 points are assigned, but the remaining 3 letters "TS
Since “U” is not a candidate, 0 point is assigned, and as a result, the evaluation score as a column is [3 points × 1 + 5 points × 5 = 2
8 points]. Therefore, the evaluation score of the entire first kana-Roman character candidate string pair is [20 points + 28 points = 48].
Point].

Further, as shown in step 5, the evaluation score of the first Kana-Roman character candidate thus obtained is divided by the total number of characters to obtain an evaluation score per character. In this case, the first kana-Roman character candidate string pair has an evaluation score of 48 points as described above, and the total number of characters is 4 characters + 9 characters = 13 characters. Therefore, the evaluation score per character is [48 Point / 13 = 3.69 ...], which is about 3.7 points. Then, as shown in step 6, this evaluation point (3.7 points) is stored as the evaluation point of the first kana-Roman character candidate string pair “Takahatsu”-“TAKAHATSU”.

After the evaluation score per character of the first kana-Roman character candidate string pair is calculated and stored as described above, the second kana-Roman character candidate character string pair is similarly calculated. The evaluation score per character is calculated and stored. That is, only one character of the kana candidate is selected from the second candidates to generate the second kana candidate character string (step 1), and similarly, a series of steps 2 to 6 is performed to make the second character. Obtain the evaluation score of the Kana-Romaji candidate string pair. By repeating this procedure, the evaluation score per character is calculated and stored for all kana-romaji candidate string pairs.

Then, the final kana candidate character string is generated,
When the evaluation score per character of the last kana-romaji candidate character string pair is calculated and stored, in other words, as shown in step 7, when a new kana candidate character string cannot be generated, step 8 is executed. Proceeding to step 2, the kana-roman character candidate string pair having the highest evaluation point is selected from all the kana-roman character candidate string pairs, and is output as the recognition result.

Furthermore, the kana-romaji matching processing section 13 obtains the kana obtained by the recognition result editing means 10.
The recognition result of the Roman character field pair and the recognition result of the other fields are edited and the final recognition result is output.

As described above, in the present embodiment, the Kana-Roman character matching processing unit 13 specifies the same reading without using the "word" dictionary as in the prior art. By performing the matching process of "kana" and "romaji" in the field pair, the recognition rate of characters in these fields can be improved.

The present invention is not limited to the above-described embodiment. For example, the number of candidates output by the character recognizing means 8 can be changed as appropriate, and the similarity calculation of candidates can be performed by various methods. It is possible to do. In addition, specific kana
The method of roman character matching processing can also be freely selected. For example, in the above embodiment, instead of evaluating all the kana-romaji candidate string pairs, the first candidate and the second candidate are included in a certain proportion or more. It is also possible to extract and evaluate only some of the candidate column pairs that exist.

[0036]

As described above, according to the present invention, the kana-character recognition result recognized by the kana-character recognizing means and the roman character recognition result recognized by the roman-character recognizing means are processed by the kana-roma character matching processing section. To provide an excellent optical character reading device capable of reading characters with a high recognition rate without using a "word" dictionary as in the past by performing collation comparison and performing a certain post-processing. You can

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of an optical character reader according to the present invention.

FIG. 2 is a schematic diagram showing an example of an entry frame of a form to be read by the apparatus of FIG.

FIG. 3 is a diagram showing an example of a recognition result by the character recognition means of FIG. 1, in which a kana character “Takahashi” and “TAKAH” are set for a set of areas designated to have the same reading.
(A) showing Kana characters for each of the three candidates given to each Kana character image data when the Roman character "ASHI" is read, and three candidates given for each Romaji image data (B) showing the Roman letters of.

FIG. 4 is a flowchart showing an example of a processing flow of a Kana-Roman character matching processing unit in FIG.

FIG. 5 is a block diagram showing an example of a conventional optical character reading device.

[Explanation of symbols]

 1 ... Form 2 ... Optical system 3 ... Photoelectric conversion unit 4 ... A / D conversion unit 5 ... Image memory 6 ... Character image cutout unit 7 ... FC table 8 ... Character recognition unit 9 ... Standard character pattern storage unit 10 ... Recognition result edit Means 11 ... Form transportation means 12 ... Transportation control section 13 ... Kana-Roman character matching processing section

Claims (5)

[Claims] 1. An optical character reader for optically reading characters, wherein kana character recognizing means for reading "kana" characters, roman character recognizing means for reading "romaji", and an area consisting of a plurality of continuous characters are designated. Then, the recognition character type designating means that stores the data that specifies the reading character type of this area, and the kana character area designated to read the "kana" character by the recognition character type designating means and the reading of the same reading as this kana character area. A post-processing designation means for designating a fixed post-processing for a set of areas which are designated as data and which are designated as "Roman characters" and which are designated as Roman characters; The kana character recognition result recognized by the kana character recognition means for the kana character area and the roman character area forming the set of areas specified by the specifying means, A Roman alphabet recognition result recognized by the recognition unit by matching comparison make certain postprocessing Kana - optical character reading apparatus characterized by having a Romaji matching process unit. 2. An FC table, which also serves as the recognition character designating means and the post-processing designating means, is used.
In the C table, FC corresponding to the form to be read
Data is registered for each character or for each area in units of continuous characters, and as part of this area information, a set of kana and romaji areas designated to have the same reading is registered. 2. The optical character according to claim 1, further comprising command information for storing a recognition result in the combination of the Kana character and Roman characters in a Kana-Roman character matching processing unit. Reader. 3. The kana character recognizing means and the roman character recognizing means sort the characters having the highest degree of similarity by performing similarity degree calculation, and output a fixed number of plural candidates as a recognition result for each character. The optical character reading device according to claim 1, wherein the optical character reading device is configured. 4. The kana-Roman character matching processing unit specifies that the Kana character recognition unit and the Roman character candidate output from the Roman character recognition unit have the same reading by the area information of the recognized character designating unit. A kana candidate and a roman character candidate corresponding to a pair of a kana character and a roman character area are used to generate a kana candidate character string and a roman character candidate string according to the designation of the post-processing designating means. The optical character reader according to claim 3, wherein the optical character reader is configured as described above. 5. A kana-romaji matching processing unit generates a kana candidate character string from the kana candidate output by the kana character recognition means, generates a roman character candidate string from this kana candidate character string, and generates a kana candidate character string. The kana candidates are matched to calculate the kana candidate character string evaluation score, and the romaji candidate string and the romaji candidate string are matched to calculate the romaji candidate string evaluation score. Further, kana candidate candidates output by the kana character recognition means are calculated. The evaluation points are assigned in order, the evaluation points for each pair of candidate strings of Kana and Roman characters are calculated, and finally,
The optical character reading device according to claim 4, wherein the set having the highest evaluation score is configured to be output as a recognition result.