JPH02219190A - Character recognizing method - Google Patents

Abstract

PURPOSE:To contrive the improvement of the recognition rate and the segmenting performance of a character, etc., by comparing the shape feature quantity of an input character with the shape feature quantity stored in advance of a candidate character corresponding to the result of its recognition and determining whether the result of recognition based on the similarity is adopted or not. CONSTITUTION:For instance, it is assumed that an input image is (u) (Japanese), and a shape feature obtained from the input image is 'long longitudinally, and separated into two in the longitudinal direction', and on the other hand, a first place in a first - a tenth places of candidate characters of a result of recognition is (ro) (Japanese). In this case, the shape feature of (ro) stored in advance in a dictionary is 'not long longitudinally nor long laterally and not separated longitudinally nor laterally'. Accordingly, since its result becomes 'NO' by a longitudinally long/laterally long check, the check of the candidate character of a second place is executed. The candidate character of a second place is (u) (Japanese), and the shape feature quantity coincides, therefore, it is adopted as the result of recognition. In such a way, the improvement of the recognition rate and the segmenting performance of a character, etc., can be contrived.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a character recognition method for recognizing characters and marks (also referred to as characters, etc.), and particularly to improvements thereof.

[Conventional technology]

Conventionally, as a method of this kind, for example, after cutting out a character line or character string (hereinafter referred to as a character line) from a document image, what looks like characters is extracted as temporary characters, and a predetermined character size is extracted from these temporary characters. There is a method of cutting out full-width characters (such as Kanji characters) as a standard, and if characters cannot be determined to be full-width characters based on the character size, check whether the characters can be effectively established as half-width characters or not, and if so, recognize them as half-width characters.

However, in such a method, there is a risk that a full-width character may be mistakenly recognized as two or more half-width characters. Therefore, the applicant has proposed the following method (
Japanese Patent Application No. 63-292445; also referred to as the proposed pickling method).

FIG. 4 is a flow chart for explaining the proposed dipping method.

This shows the processing procedure of an image processing device (not shown). First, document image data is input (see ■), and each character line is cut out by taking its horizontal projection value (see ■). , determine the line width dimension and obtain the amount equivalent to the size of a full-width character (character size). Note that although horizontal writing is assumed here, the same applies to vertical writing.

Next, by examining the projection value in the direction perpendicular to each line and taking into account the character size, cut out what appears to be a character, that is, a group of temporary characters, from each line of characters. Select full-width characters using the above font size (see ■).

The conditions for selecting characters as full-width characters are as follows.

b) The font size is full-width by itself, that is, there is no room to combine it with other temporary characters.

(mouth) Punctuation marks (c) Punctuation marks are half-width by themselves, but become full-width when combined with other half-width adjacent characters.

2) Although the size alone is smaller than the full-width size,
There is too much distance between adjacent half-width temporary characters, and if you forcefully combine them, the size will exceed the full-width character size.

After selecting all full-width characters according to the above conditions, combine or separate the remaining temporary characters to create 2 integrated characters and 2 separated characters (see ■), and after selecting these integrated characters and 9 separated characters. is recognized by an OCR (character reading device) using the degree of similarity with a dictionary pattern (see ■).

Next, the following contradiction processing is performed on the recognition result (see ■).

a) For example, even though the object to be recognized is a separate character, if the recognition result by OCR indicates a full-width kanji character, it would be inconsistent with each other, so such a recognition result would not be adopted.

b) Contrary to the above case, even though the object to be recognized is an integrated character, the recognition result by OCR shows half-width size characters such as alphabetic characters and numbers.

Then, regarding the last remaining character, whether it should be used as an integrated character or a separated character is determined by OCR using relative similarity (see ■).

Note that the relative similarity X is defined as the product of the similarity X and a constant (for example, 1024) that is the ratio of the average value m of the similarities. That is, x,=x/mx constant (1024).

[Problem to be solved by the invention]

In the proposed method, recognition results are determined based on the degree of similarity, so for example, very similar words such as ``u'' and ``ro'' or ``te'' and ``chi'' may be recognized by slight differences in the character images. For example, there is a risk of erroneous recognition due to differences in the thickness or font of characters, such as one character being crushed or blurred.

Also, for the same reason, the relative similarity of ``(
” and the “yo” part have a higher relative similarity, which may lead to misrecognition. This point will be explained in more detail.

FIG. 5 shows I! of the input character "wa" when it is processed as integrated character 1 and when it is processed as separated characters 2 and 3. (11 degrees X, relative degree of assimilation The relative similarity obtained by multiplying by a constant number (1024) is
For the integrated character l, r954J is used, and for the separator character 2.3, r1028J and r898J are used, respectively, and the average value is "963". Since 954<963, the input character "wa" is "(" ) and "wa".The constant is of course not limited to r1024J.

[Means to solve the problem]

When recognizing a character by checking at least its similarity with a dictionary pattern, the aspect ratio of the character to be recognized and shape feature values indicating how many parts can be separated in the vertical and horizontal directions are also stored in advance as a dictionary, The shape feature amount of the input character is compared with the shape feature amount stored in advance of the candidate character corresponding to the recognition result, and it is determined whether or not to adopt the recognition result based on the degree of similarity.

[Effect]

In addition to the similarity, we also extract shape features,
Improves the recognition rate of similar characters with different shape characteristics, such as "u" and "ro", "chi" and "te", etc. Furthermore, the character extraction performance is improved by preventing ``wa'' from being misrecognized as ``('' and ``wa'').

〔Example〕

FIG. 1 is a flowchart showing an embodiment of the invention. As is clear from the figure, this embodiment is characterized by the addition of a shape feature matching step (2).

Step (2) is subdivided into [phase] to [phase] as shown in the figure.

That is, in step [phase], it is checked whether the input character is vertically long or horizontally long. Regarding whether the character is vertically long or horizontally long, if the character is a regular character, if the aspect ratio (height 7 width) is, for example, 2 or more, it is considered vertically long, and if it is 1/2 or less, it is considered horizontally long. Furthermore, in the case of long characters or flat bar characters, the characters are converted to the normal characters for determination.

In steps {circle over (2)} and @, it is determined how many vertical and horizontal divisions are to be made, respectively, and whether or not the manner of separation matches the input character and the candidate character resulting from its recognition is checked. This operation is performed until a matching candidate character (1st to 10th) is found. If none of the candidate characters matches, a reject output is output in step [phase].

Note that the other parts are the same as those in FIG. 3, so the explanation will be omitted.

Next, a specific example will be explained.

In Figure 2, the input image (or input character) is ``u'', and the shape feature obtained from the input image is ``vertically long and separated into two vertically'', and the recognition result candidate character This is an example in which the first place among the first to tenth places is "ro". In this case, the shape characteristics of "ro" stored in the dictionary in advance are "neither vertically nor horizontally long, nor separated vertically and horizontally."

Therefore, the step of vertical/horizontal check (Fig. 1 [
Since the answer is "NO", the second candidate character is checked. The second candidate character is "u", and since the shape feature values match, this is adopted as the recognition result.

Figure 3 shows an example where the input character "wa" is processed as an integrated character and when it is processed as a separate character r (J, ryo). In other words, it is processed as an integrated character as shown in (a) of the same figure. In this case, the first candidate "ha" in the recognition result is suitable from the point of view of the shape feature, so "ha" is adopted.Similarly, the separated character "(" is the first candidate as shown in (b) in the same figure. Although it matches the character "(", the separated character "yo" does not match the first candidate character "ha" due to its shape features as shown in the same figure (c), and in the end it is matched with the second candidate character "ma". As shown in the same figure (d), the relative similarity when recognized as "wa" is r954J, and when it is separated into "(" and "ma"), the relative similarity is r954J. "946", and since 946<954, "ha" will be adopted.In other words, the input character will be recognized as "ha", eliminating the possibility that it will be mistakenly recognized as "(" and "ma"). be able to.

〔Effect of the invention〕

According to this invention, in addition to recognition based on similarity, shape time f
! Since the kit (vertically or horizontally, how many parts to separate vertically, how many parts to separate horizontally) can be taken into account, it is possible to correct errors in recognition results based only on similarity, and it is possible to correct characters, etc. recognition rate.

Cutting performance can be improved.

[Brief explanation of the drawing]

FIG. 1 is a flowchart showing an embodiment of the invention, and FIG.
3 and 3 are explanatory diagrams for specifically explaining the present invention, FIG. 4 is a flowchart for explaining the proposed method, and FIG. 5 is an explanatory diagram for explaining an example of erroneous recognition. be. 1... integrated character, 2, 3... separation character. Representative Patent Attorney Akio Namiki

Claims (1)

[Claims] 1) In recognizing characters by image processing a document, cutting out individual characters, and determining the degree of similarity with a dictionary pattern for each character, the aspect ratio and vertical direction of the characters to be recognized are determined. Shape features indicating how many parts each character can be separated in the horizontal direction are also stored in advance as dictionary data, and the recognition results are compared with the shape features of the corresponding candidate character stored in advance. A character recognition method characterized by determining whether or not to adopt a recognition result based on similarity.