JPS6079486A - Character recognizing system - Google Patents

Abstract

PURPOSE:To reduce a cost of an OCR by generating an OCR use standard pattern from a character generator. CONSTITUTION:An observing part 21 converts an input character pattern 20a to an electric signal 21a. This electric signal 21a is converted to a unit character pattern 22a by a pre-processing part 22, and also converted to a feature 23a by a feature extracting part 23. On the other hand, a feature is extracted by a feature extracting part 25, generating successively a font pattern 24a by a character generator 24, and an OCR use standard pattern 25a is obtained. Subsequently, in a matching part 26, similarity between the feature 23a and each standard pattern 25a in evaluated. Next, in a deciding part 27, basing on an evaluated value of similarity, whether an answer category 27a is outputted or an output of the answer category is refused 27a is determined.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Application of the Invention] The present invention relates to a character recognition system, and particularly to a standard pattern creation and reference method in the character recognition process.

[Background of the invention]

Character recognition process (for a given printed or written character pattern, identify the character category IJ' of that character pattern)
The standard pattern memory (for character 1 and It is common to prepare both a standard standard pattern memory for OCR (hereinafter referred to as standard pattern memory for OCR) and a pattern memory for character display and printing (character generator, hereinafter also referred to as character display and print memory). However, like kanji, there are many character categories (Nariai (commonly used kanji with 194
5 categories, JIS, 2965 categories in 'Le 1 level! J
iJIS No. 1. There was a problem in that the cost would be high if both memory devices (6349 categories) were implemented in both devices at the second level.

As one method to solve this problem, conventionally, OCI
A method has been proposed [Application A 49-262'18] in which only a standard pattern memory for "" is installed in the device, and a font for displaying and printing characters is created from a standard pattern for OCR. The invention has a drawback that if the expression (feature expression) of the head weight pattern for OCR is degenerate, the font block character shape used for character display or printing cannot be restored from the standard pattern.The character shape can be restored. Examples of feature representations that cannot be performed include a feature representation called the 0-marginal distribution, ■ a feature expression called the stroke density function, ■, < l) A feature table called ferential features: L@Of which (C) is the first-order peripheral feature. (d) is called the second-order peripheral feature.

The features of ■~@ have been shown to be effective in recognizing printed or handwritten Chinese characters, and the above-mentioned drawbacks have been a major hindrance in realizing a Chinese character recognition device, which is an important objective of the present invention.

The characteristic expression of ■ is [Chunichi, Nakano et al., Recognize
io口of （：：hinese Character
vs , Pros, of the Conference
on machine perception of
patterns and pictures hel
d at the NPL%Teddington,
It is described in detail in [April 1972].

The feature expressions of ■ and ■ are based on "Hagita et al., Classification of hand-written kanji using three outline features, Theory of IEICE, Vol. 63D, A.
12, PP1096-1101, [December 1980].

[Purpose of the invention]

SUMMARY OF THE INVENTION An object of the present invention is to eliminate the drawbacks of the conventional inventions and provide a standard pattern creation and reference system in the character recognition process.

[Summary of the invention]

In order to achieve the above object, the present invention does not have a standard standard pattern memory for OCR, but only a pattern memory (character generator) for character display and printing is installed in the device,
OCR from font generated by character generator
For reference while creating standard turns,
Enables character recognition. As a result, the drawbacks of the conventional invention are solved, and degenerate feature expressions that cannot restore character shapes are eliminated.
The standard standard pattern for CR can now be used.

Furthermore, by eliminating the need for a standard pattern memory device for OCR, the memory capacity can be kept the same as before. Furthermore, when recognizing printed characters (printed characters) in turn 1, the OCLt standard turns created by the method of the present invention from the font that is the source of the printed character pattern can be used to convert the conventional printed It is characterized by high character quality compared to standard patterns for OC and FT created after conversion to signals.

There is also the effect that character recognition degree (particularly for high-quality character patterns) can be increased by reflecting this feature.

The basic procedure of the TV according to the present invention is shown in FIG.
ramAnalysis l)iagram).

The following are four precautions regarding this PND.

■ Feature f refined in step (2) and n obtained in step (3-1) O(, feature g of standard pattern for R,...
...+glj+...+ g IJ (1) must be in a corresponding format that allows for consistency in step (3-2). For example, when f and gz(1×J(su)) are expressed as vectors, it is necessary to correspond the number of dimensions and the range of each dimension.

(2) Preprocessing (mainly normalization of size, slope, etc.) in step (3-1) is often unnecessary. A typical example is the case where it is possible to normalize the unknown character pattern U to match the font pattern P+ in step (2).

■ As a special case of step (2), do nothing to U f
is U itself. There are cases.

■ As a special case of step (3-1), there is a case where nothing is applied to Pl, g++-P'', PO)-1.

[Embodiments of the invention]

Examples of the present invention will be described below.

FIG. 3 is a diagram showing an embodiment of the present invention.

In the figure, the human character pattern 20a first appears on the observation unit 2.
1 is man-powered. The observation unit 21 receives the input character pattern 202
is converted into a drowsiness signal 21a. The drowsiness signal 21a is converted into a unit character pattern 22a which has been subjected to one or both of preprocessing and cutting. Preprocessing may also be performed within the feature extraction unit 23. The unit character pattern 22a is converted into a feature 23a by the feature extractor 23. In this way, the input character pattern 20a becomes the feature 23a.
is converted to

On the other hand, while the character generator 24 generates the first-order font pattern 24ai, the font pattern 24a is subjected to preprocessing (mainly normalization) and feature extraction by the font 25, and is then used as an OCR standard pattern 25a.

The font pattern 24a is subjected to preprocessing and feature extraction.
, an example of obtaining the CR standard pattern 25a will be described with reference to FIG. 3(a).

31 font patterns z (X, y), X=1,...
...+ Nl + )' = 1 + ......,
Named N2. This font pattern "'(x+y
), (X@) marginal distribution 32P + (X) and (y axis)
The marginal distribution 33 P20') is calculated using the following formula, and OCR
standard standard pattern 25a.

jP2(Y)-Σ: z(X+Y) *y==l
l+・Song, N2! Territory! Further, as a method for preprocessing and feature extraction to obtain a standard pattern for 0CFt from the font pattern 24a, a modest method known in the field of character recognition can be applied.

Next, the matching unit 26 evaluates the similarity between the feature 23a and each standard pattern 25a (feature expression). Next, in the determination unit 27, based on the similarity evaluation value,
Decide whether to output the answer category 27 ai or to reject the output of the answer category (reject 27b).

The control unit for signals such as 20a, 21a+, etc.
This is done by the control unit 28. In Figure @3, the suppression signal line has been omitted to avoid cluttering the diagram. 21°22,...
. . , 28 processing units and storage units include a photoelectric conversion device, μ
It can be realized by m processors, memory, etc. Non-volatile memory is used for the 24 character generators.

FIG. 4 is a diagram showing another embodiment of the present invention.

This embodiment is a word processor or character printing device) 41
This is a case where the character generator 43 is shared by a device that is integrated with a character recognition device t42i.

The character recognition device 42 performs preprocessing, feature extraction, etc. on the font pattern of the character generator 43 to create a standard pattern, and refers to the standard pattern.

A word processor (or character printing device) 41 causes a character generator to generate a font, and uses it for displaying or printing characters.

〔Effect of the invention〕

According to the present invention, since it can be created from the character generator of the standard standard pattern price for OcR and referred to, 0
The price of CR (especially Kanji 0CR) can be lowered. Furthermore, cost performance can be further improved by using the character generator for creating standard patterns for OCR in combination with character display and printing.

FIG. 2 is a diagram showing the P'A D for carrying out the present invention, FIG. 3 is a diagram showing the configuration of a first embodiment of the present invention, and FIG. 4 is a diagram showing the configuration of another embodiment of the present invention. It is a diagram.

Claims (1)

[Claims] 1. Character recognition that evaluates the class [personality] between an unknown character pattern and a number of standard character patterns and determines the output of an answer category based on the evaluation value of its robustness. In processing,
A character recognition method characterized in that the standard character pattern is obtained by performing preprocessing, feature extraction, or other conversions on the above-mentioned font patterns or ondo patterns generated by a character generator. 2. A character recognition method as set forth in claim 1, characterized in that the font pattern generated by the character generator is also used for displaying characters, stamping, etc.