JPS6318484A - Method for recognizing printed character - Google Patents

Abstract

PURPOSE:To rapidly recognize various kinds of characters with the small number of templates by correcting the height of a detected character to a character pattern height in a character dictionary, and at every generation of a reject, referring a template with a different font and a general template. CONSTITUTION:A character feature pattern is extracted from a character string read out by a scanner 11 in a optical input device 2 by a character extracting unit 12 and a feature extracting unit 13 and matched with the content of a template memory 15 by a feature matching unit 14 controlled by a processor such as a CPU to recognize the character. At that time, the character height in the feature pattern is corrected to the character pattern height of the character dictionary, templates respectively corresponding to various fonts are successively referred at every generation of a reject, and a reject is still generated, the general template including characters with low using frequency is referred. Thereby, various kinds of characters can be recognized with the small number of templates and characters in a sentence can be rapidly recognized.

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field] The present invention relates to character recognition technology, and particularly to technology for recognizing printed characters such as Japanese and English including various character types.

[Prior Art] Conventionally, in order to read a wide variety of characters, the following procedure was used. Recognition is performed, and if the font is rejected, it is compared with a template (dictionary) of a different font. That is, as shown in FIG. 6, when one character string is recognized, in step 1 51, a blank space of a certain width or more is detected and a character pattern is cut out, and in step 52 it is determined whether character cutting is completed. If the process has been completed, the process ends (END).If the six character extraction has not been completed, the process moves to the first stage N53.

The character pattern cut out in this step 53 is subjected to feature extraction and matching processing (in FIG. 6, the flow for feature extraction processing is omitted because a normal flow is used).
Then, in step 54, it is determined whether the matching result is a reject, and if the result is a reject (Y
ES), the template is changed in step 55, and if it is not rejected (No), the matching result is output in step 56 and the process returns to step 51.

As a result of changing the template in step 55, it is determined in step 57 whether or not there is any template to be changed, and if there is no template to be changed (YES).

Output a reject code in step 58 and return to step 51. If there is a template to be changed (No), return to step 53. Try changing the font in this way until matching is achieved or maximum matching is achieved. I was searching for candidate templates until now.

However, when there are many types of characters as in printed matter, all the fonts must be prepared, so the template (dictionary) becomes enormous.

Therefore, there was a problem in that it took too much time to recognize one character.

[Objective] An object of the present invention is to provide a technology that can recognize sentences including various character types using a small number of templates (dictionaries).

Another object of the present invention is to provide a technique that can quickly recognize sentences containing various character types.

The above and other objects and novel features of the present invention will be explained by the following description of the specification and the accompanying drawings.

[Structure] The present invention is a printed character recognition method for recognizing printed characters using font-specific templates corresponding to various fonts and general-purpose templates including character patterns that are rarely used. a step of detecting the height of the detected character, a step of correcting the height of the detected character to the height of the character pattern of each of the dictionaries, and each time a rejection occurs,
The present invention is characterized by the step of referring to templates for different fonts and, if the template is still rejected, recognizing the character using a general-purpose template.

[Example] Hereinafter, an example of the present invention will be specifically described using the drawings.

Note that throughout the description of the embodiments, parts having the same functions are given the same reference numerals, and repeated explanations thereof will be omitted.

FIG. 1 is a block diagram showing a schematic configuration of an OCR input device according to a printed character recognition method according to an embodiment of the present invention, and FIG. 2 shows multifunctional printed character information using the OCR input device shown in FIG. FIG. 3 is a block diagram showing a schematic configuration of the processing device. FIG. 3 is a flowchart of a printed character recognition method according to an embodiment of the present invention.

In the printed character information processing apparatus to which the printed character recognition method of this embodiment is applied, as shown in FIG.

In addition to inputting a single character, keyboard 1 can also be used in various modes (kana-kanji conversion, kanji-kana conversion, OCR character recognition, English text, etc.)
Used to specify. The OCR input device 2 optically reads and inputs a document. The processing device 3 includes a keyboard 1
The input information from the OCR input device 2 is processed according to the specified mode, and outputted to the output bag W14. The output device 4 is a general term for a display device, a printer, etc. It is stored in the program memory (ROM) 5 necessary for the processing of the processing device 3, and includes post-processing such as kana-kanji conversion by keyboard input, OCR character recognition,
A common algorithm is used as much as possible for the kana-kanji and kanji-kana conversions of character strings input by OCR.

A data memory (RAM) 6 is used to store data and parameters that are being processed by the processing device 3. The word dictionary memory 7 stores a word dictionary to which reading orthography correspondence data is added.

The OCR input device 2 shown in FIG. 2 is the OCR input device 2 shown in FIG.
Image information such as characters on a document is read and input using an optical scanner 11 comprising a light source, a charge-coupled device (CCD), and the like. The input image information such as a kana character string, a kana-kanji mixed character string, or an alphabetic character string is cut out character by character by a character cutting unit 12, as shown in FIG.

A feature extraction unit 13 extracts the features of the extracted characters. This extracted data is stored in a template memory (ROM or RAM) 1 in a feature matching unit 14.
Matching is performed with the feature dictionary data (font-specific templates, general-purpose templates) stored in 5. If matching is achieved, the input characters are recognized and sent to the processing device 3. If the matching is rejected, a reject signal is generated from the feature matching unit 14 and sent to the character cutting unit 12.

Next, the processing process of the character recognition method for an OCR input device of this embodiment will be explained according to the flowchart shown in FIG.

First, when recognizing a character string as shown in FIG. 4, a blank space having a width equal to or greater than the first stage 101 is detected, a plurality of character patterns within the range of ■ are cut out, and then divided into individual character patterns. A projection method is generally used as this character extraction method. In the character string pattern in the range of ■, HO (height of the most frequent character) = 30. H, (height of letters with next frequency)=22. P.

(Width of most frequent character)=19. Suppose that P, (width of the character with the next frequency)=15 is extracted. Here, the unit of numerical value is one pixel.

Next, in step 1 102, it is determined whether character cutting is completed,
If the process has ended, the process ends (END).

If character segmentation has not been completed, the process moves to step 1 103.

At this stage 103, the first pattern capital letter T is used for feature extraction.
Matching processing is performed (feature extraction is omitted in Figure 3).At this time, the dictionary required for matching is the height of one character, H,, Hl, and the width (pitch) of character, P,, P. Yes, these and extracted information) low h>* pay
Pi and the extracted information) lee
) ltt Pa, Pl is Ho in the dictionary.

If it is within the range of H, , P, , Pl, subsequent matching is executed. If it is outside the range, proceed to step 104. At step 104, it is determined whether or not the matching result is a reject. If the result is reject (YES), at step 105, if the matching result is in the dictionary containing H and P. If the candidate is not rejected (NO), the matching result is output in step 106 and the process moves to step 107.

The post-matching process uses a conventional method.

Here, the font-specific templates and general-purpose templates stored in the template memory 15 are the fifth
As shown in the figure, the dictionary identification code 21 is placed first, and then the heights H, , H, 22 of the characters of this dictionary are placed. Next, the character width (pitch) p, .

The general template has font code and character height H
, it is assumed that a 0 code is included in the width P portion of the character.

In the same way, the part marked ■ shown in Fig. 4 is recognized.

Next, recognize the part marked ■. The above-mentioned method may be used for the first six patterns in the part (3), but matching fails in the seventh and eighth patterns, and step 104 of the flowchart
Rejection may occur. At this time, if there is no next candidate to be changed in step 105, step 10
8, it is determined whether there is a dictionary to be changed to the primary candidate, and if there is no dictionary to be changed, the process moves to step 109, and using the extracted information has h1+ Pat pt,
Templates containing this are sequentially selected and ChEEH, pC
P).

Repeat this until you find a suitable font.

Frequently used fonts are usually converted into dictionaries (templates), so the found fonts can be recognized.

When there is an exceptional character pattern, such as the part marked ■ in Figure 4, the character pattern is normalized (for example, the same height as the general template).

After converting to a size with a uniform width), the font code is 0.
At step 110, matching with the general template (dictionary of normalized images) is attempted using the template.

It is determined whether or not the matching result is rejected in step 111, and if it is not rejected, the result is 9 (NO).
, move to stage FI1107, and if it is a project (YES
), proceed to step 112.

A reject code is output and the process moves to step 1 107.

In step 107, the next image pattern of interest is cut out from the image information pattern, and the process proceeds to step 113.In step 113, it is determined whether or not there is the next image pattern of interest. ), return to step 104 1
If there is no next image pattern of interest (No), step 101
6. The process of the character recognition method for the OCR input device of this embodiment is ended.

The present invention has been specifically explained above using examples.
It goes without saying that the present invention is not limited to the embodiments described above, and can be modified in various ways without departing from the spirit thereof.

〔Effect of the invention〕

As described above, according to the present invention, there are a step of detecting the height of a character to be recognized, a step of correcting the height of the detected character to the height of the character pattern of each dictionary, Every time a rejection occurs.

The system includes a step of referring to templates for different fonts and recognizing characters using a general-purpose template if the template is still rejected, so that sentences containing a variety of character types can be recognized with a small number of templates. It is possible to quickly recognize sentences containing a variety of character types.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing a schematic configuration of an OCR input device according to a printed character recognition method according to an embodiment of the present invention. FIG. 2 is a block diagram showing a schematic configuration of a multifunctional printed character information processing device using the OCR input device shown in FIG. 1. FIG. 3 is a flowchart of a printed character recognition method according to an embodiment of the present invention, and FIG. 4 is a diagram showing an example of a pattern of a plurality of cut out characters. FIG. 5 shows the contents of the template memory according to this embodiment.
FIG. 6 is a flowchart for explaining a conventional character recognition method for an OCR input device. In the figure, 3...processing device, 11...scanner 1
2... Character extraction unit, 13... Feature extraction unit, 14... Feature matching unit, 15...
This is template memory.

Claims (2)

[Claims] (1) A printed character recognition method that recognizes printed characters using font-specific templates that support a variety of fonts and general-purpose templates that include character patterns that are less frequently used, and which detects the height of characters to be recognized. a step of correcting the height of the detected character to the height of the character pattern of each dictionary, and a step of referencing a different template for each font each time a rejection occurs, and if a rejection still occurs, a general template is used. A printed character recognition method characterized by comprising a step of recognizing characters using the method. (2) The printed character recognition method according to claim 1, wherein each dictionary has a typical character size (height, width) in addition to its identification code.