JPS59148983A - Method for selecting "kanji" recognizing dictionary - Google Patents

Abstract

PURPOSE:To attain ''kanji'' (Chinese character) recognition with high recognition accuracy and superior operability by using correspondence between item names and columns in a business form. CONSTITUTION:A preprocessing logical part 304 reads out the contents of a sequential picture memory 303 under the control of a control part 302, extracts the ruled lines of a table and segments images included in the insides of the A and B columns of each line. A control part 302 selects the internal image of the column A of the 1st line and an educational ''kanji'' dictionary as a specified dictionary from a dicitionary file 306 and transmits the selected results to a recognition logical part 305. The recognition part 305 returns the recognized result to the control part 302. On the basis of the returned result, the control part 302 selects a proper noun ''kanji'' dictionary as a specified dictionary for the column B of the 1st line in the corresponding table from the dictionary file 306 and transmits the selected dictionary to the recongition part 305 together with the image of the inside of the 1st line column B. On the basis of the specified ''kanji'' dictionary, the recognition part 305 recognizes the contents of the column B of the 1st line and transmits the recognized result to the control part 302.

Description

[Detailed description of the invention] [Field of application of the invention] The present invention automatically reads characters on a form.

4. Regarding optical character reading devices (hereinafter referred to as OCR). .

In particular, the present invention relates to a recognition dictionary selection method suitable for reading Japanese information including kanji.

[Prior art]

Read conventional alphanumeric characters, symbols, and kana (with JCB, there are only about 100 types of characters that can be read, so there was no need to consider characters other than those classified as cannons (so-called external characters), but kanji, If you expand the target to hiragana, 1 usually costs 2000~
3000 character types are required, and other external characters must be considered depending on the application name.

For example, in the example of the slip shown in FIG. 51, in the product name column of the first line.

Kanji derived from the unique name, in the second line address field.

Kanji characters unique to place names, unique to one person's name in the third line name column.

The kanji is entered. One of the conventional methods is simply.

Recognizes all necessary external characters using a common dictionary that has been registered in advance.

, and the target character type is unnecessary for the corresponding column.

This has the disadvantage of reducing recognition accuracy.

there was.

Another method that has been used in the past is to specify the format on the slip in detail. The third row, column B, is used to check a person's name dictionary at the recognition stage. Although it is optimized for the dictionary, if there are many types of slip formats, or if there are changes to be made, the format specifications are changed to OCR one by one. I have to give.

Because there is no such thing, it has the disadvantage of being extremely complicated. -Object of the Invention] An object of the present invention is to solve the drawbacks of the above-mentioned conventional methods and to provide five Kanji recognition methods with high recognition accuracy and excellent operability.

[Summary of the invention]

A feature of the present invention is that it is possible to create one name per item by using the correspondence between items/names and entry fields in a form.

A highly versatile kanji dictionary with a relatively small number of characters.

A special kanji dictionary is selectively used for recognition of entry fields based on the recognition results of corresponding items and item names.

It's where you apply it.

The format of normal forms is shown in Figure 1.

As shown, the item names representing the contents of the entry column (column B) are provided adjacently (Δ column). Therefore, as a first step, the Δ column is read, and from this recognition result, a necessary dictionary can be selectively applied when recognizing the corresponding B column. or,
The character types listed in Column A are generally extremely versatile and can be used only in, for example, an educational kanji dictionary.

Furthermore, regarding the diversity of funer mats, for example, as shown in Figure 1, the data structure in one line is one item field (
A') and the entry field (13) can be specified in advance by providing only this multi-line configuration and by recognizing the check lines in the form. Regarding the detection of frames in documents, this can be easily achieved using known techniques.

For example, it is described in the following document.

Information Processing Society of Japan 25th Gold Country Conference 1゜ "Detection of frames in documents" Mata Sakurai Regarding the diversity of item names and corresponding dictionaries.

Create a corresponding table and add this table.

This can be done by simply adding Corresponding tape to Figure 2.

Here is an example of a file. Embodiment of the I5r Invention] An embodiment of the present invention will be described below with reference to FIG. Images on the form are sequentially read by the scanner 301 and stored in the image memo+7303 via the control unit 302.

The preprocessing logic unit 304 is a control unit. 602, the contents of the image memory 303 are sequentially read.

Extract the lines from Table 1 and record the Δ and B columns in each row.

Cut out the image inside.

The control unit 502 first selects the internal image in the Δ column of the first row, and
An educational kanji dictionary is selected from the dictionary file 306 as the designated dictionary and is transmitted to the recognition logic unit 605.・If we take FIG. 1 as an example of a form, the recognition unit 305 sends back the recognition result of “product” to the control unit 302.
First, based on this, from the correspondence table shown in FIG.

As a designated dictionary, use a proper noun kanji dictionary.

Select from the document file 306 and inside the first row, column B.

It is transmitted to the recognition unit 605 along with the image. Recognition part.

605, in the first row, column B, based on the designated kanji dictionary.

and transmits the result to the control unit 302. below,.

In the same way, apply to the second row, column I, and the third row, column A.

The control unit 602 recognizes "address" and "name", and selects a place name kanji dictionary and a person's name kanji dictionary from the correspondence table shown in FIG. The contents of the corresponding column B are recognized based on the Kanji, 4 and 4 character dictionaries, and the results are sent back to the control unit 502.

The above series of processing steps of the control section are shown in a flowchart in FIG.

[Effects of the Invention] 5. According to the present invention, complicated format regulations can be provided even for various document formats.

Rather, the most suitable kanji corresponding to the item field on the form.

High precision and variety by applying a dictionary.

It is possible to realize leisurely reading of kanji types. 18

[Brief explanation of drawings]

Figure 1 shows an example of a form that includes kanji, and Figure 2 shows the correspondence between the name of one item and the corresponding kanji dictionary. 3 is a block diagram of one embodiment of the present invention, and FIG. 4 is a flowchart showing the control and five-part processing procedure of FIG. 3. 301... Scanner 302... Control unit 303
...Image memo IJ 3n4,...Preprocessing logic section 305...Recognition logic section 306...Kanji dictionary file No. / Figure 2 Figure 4 Calendar

Claims (1)

[Claims] 1. Covers forms written in multiple characters including kanji. The optical character recognition device uses the correspondence between the five item names and entry fields in the form, and uses a highly versatile kanji dictionary with a relatively small number of characters to recognize the item names. ,
A kanji recognition dictionary 1 degree calligraphy selection method characterized by selectively applying a dedicated kanji dictionary from the recognition results of corresponding item names for recognition of entry fields.