JPH0792818B2 - Pattern cutting and recognition method and its system - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a pattern cutting and recognizing method in an optical character reader (hereinafter referred to as "OCR") and its system, and particularly to cutting a pattern on a form written under natural writing conditions, BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a pattern cutting and recognition method suitable for recognizing it and a system thereof.

[0002]

2. Description of the Related Art Conventionally, the characters read by the OCR are:
As shown in FIG. 1A, it was necessary to write correctly in the character frame 11 set for each character. In that case, some protrusion from the frame is allowed, but as shown in Fig. 1 (b), the extent is about 1.0 to 1.5 mm in the vertical direction, and the adjacent frame in the horizontal direction. There was not much. By the way, in order to further spread OCR, as described above,
As shown in Fig. 2 (a) or (b), the characters and numbers are not written in the OCR unique character frame, and the character frame is not so particularized, and it is natural to write as usual. It is necessary to enable writing conditions. As is clear from a comparison between FIG. 1 and FIG. 2, the conventional character frame has a large size and the inter-character frame gap 5 is 0.5 to 1.0 mm. The frame is 1 in FIG.
As shown in Nos. 2 and 13, as the dimensions decrease,
The gaps 6 and 7 between the character frames are 0 mm.

[0003]

As a result of this, the amount of characters protruding from the frames 12 and 13 becomes large, and
There arises a problem that the characters overlap each other in the vertical direction or the characters easily contact each other. Further, when the character pattern components are separated, for example, in the case of the number "5" in FIG. 2 (a) or (b), most of the components may be in the adjacent frame, and the character reading Moreover, it causes difficult problems. In such a case, the conventional technology has not been able to correctly cut out characters or read them correctly. The present invention has been made in view of the above circumstances, and an object of the present invention is to solve the above-described problems in the related art, to largely protrude from a character frame, or to be in contact with an adjacent character. About the characters written under different writing conditions, especially the entered 1
A pattern cutout and recognition method capable of combining a plurality of partial patterns into one character when the unit image pattern is determined to be a part of a pattern representing one category and performing character recognition on the combined character. To provide that system.

[0004]

The above objects of the present invention are as follows:
Enter the no form of the character frame of one character unit, electrical
From the two-dimensional image pattern on the form converted into a signal
The step of cutting out separately for each unit image pattern,
The cut out unit image pattern is used as one unit image pattern.
Or a combination of multiple unit image patterns
The recognition target video pattern is divided into target video patterns.
Compare each with each pattern in the pattern dictionary,
The character pattern corresponding to the unit image pattern is recognized.
Of the recognition target video pattern
The recognized recognition results are used to recognize adjacent recognition target video patterns.
Refer to the result and write according to a predetermined rewriting rule.
2D image pattern on the form.
It is achieved pattern cut and recognition wherein the this <br/> to comprehensively judge the character string pattern contained in over ting by the system.

[0005]

Function: Pattern cutting and recognition method according to the present inventionWhen
In that system, the pattern in the character reader is
In the cutout process, the cutout pattern (this
" Unit video pattern " Called)), one or more
Image pattern to be recognized when " Video to be recognized
pattern " Is not a complete pattern, but a single sentence
When it is judged that it is a part of a character (partial pattern),
The recognition result is used to recognize adjacent recognition target video patterns.
Refer to the result and write according to a predetermined rewriting rule.
By changing it, I decided to make a comprehensive judgment
Character recognition for characters written under natural writing conditions.
The effect of being able to improve the accuracy of knowledge
It is what is done.

[0006]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The principle of the present invention will be described below, and then embodiments will be described in detail with reference to the drawings. The principle of the present invention lies in the following two points. That is, (1) When ambiguity occurs in the cutout of a pattern, the cutout unit makes a plurality of hypotheses and recognizes a unit pattern including a partial pattern according to each hypothesis without forcibly making a judgment. Send to the department. (2) The recognition unit discriminates the unit pattern described above, and checks the validity of the cutout from a comprehensive judgment to eliminate the ambiguity. Hereinafter, this will be described more specifically. Figure 3
It is a figure which shows the various states of an adjacent character pattern. Figure 3
In (a), the patterns 31 and 32 overlap in the vertical direction. In this case, a continuous black area can be cut out as a pattern component. The method of extracting a continuous black area as a pattern component is well known in the art, and not only when the characters are correctly written in the frame, but also when the text is simply overlapped, the black area is extracted. Since it is possible to extract even the part that is outside the frame along the
The unit pattern can be cut out correctly. Regarding this, for example, A. Rosenfeld et.al., “Sequ.
Priority Opations in Digital Picture Processin
g ”(J. ACM, vol.14, No.4, Oct., 1966, pp.471-494)
Can be referred to.

Next, in FIG. 3 (b), the pattern is a portion 33.
And 34, and most of the separated components 34 are contained in adjacent frames. If it is unknown whether the pattern 34 belongs to the frame 21 or the frame 22, it is assumed that both are “possible cases” and a multiple hypothesis is created. Then, both cases are sent to the recognition section separately, and which hypothesis is correct is determined from the recognition result. 3C is a case where the separated character patterns are in contact with each other, and FIG. 3D is a case where the separated patterns are in contact with each other. In the case of FIG. 3C, the separation pattern is only the number “5”, whereas in the case of FIG. 3D, both the numbers “5” and “7” are the separation patterns and Are in contact. FIG. 3E shows a case where the complete patterns are in contact with each other. That is, the patterns are not separated, but the adjacent patterns are in contact with each other. A recognition method for the cases of FIGS. 3B to 3E will be described below with reference to FIGS. Figure 4
3A is an operation explanatory diagram when the cutout unit makes a plurality of hypotheses, and the separated pattern components 3 as shown in FIG.
No. 4 describes the recognition method when it is in the adjacent frame.

In FIG. 4, 51 and 53 are two hypotheses output by the clipping unit, 200 is a recognition unit, 100 is a pattern dictionary, and 101 to 104 are partial dictionaries in the pattern dictionary 100. The recognition unit 200 inputs the above-mentioned two hypotheses 51 and 53 to perform character recognition, and outputs recognition results 52 and 54 for each. That is, the first
The recognition results for Hypothesis 51 are (SP.5) and (RJ.6). Here, (SP.5) means “5 subpattern”, which is recognized by referring to the partial dictionary 101, and (RJ.6) is “reject (unread). But
The candidate is six ”. Furthermore, the second hypothesis 5
The recognition results for 3 are (AC.5) and (AC.6),
Both of them mean acceptance as the number "5" or "6". Therefore, the second hypothesis is valid, and the recognition results are the numbers “5” and “6”. The partial dictionaries 101 to 103 provided in the above-mentioned pattern dictionary 100 are newly provided in the present invention, and conventionally, only the dictionary 104 having a normal pattern is provided.

The partial dictionary 101 is a dictionary storing partial patterns, the partial dictionary 102 is a dictionary storing patterns in which a partial pattern contacts another character, and the partial dictionary 103 is a contact character. It is a dictionary that stores patterns. Next, as shown in FIG. 3C, the recognition method when the separated pattern component is in contact with an adjacent character is
A description will be given based on FIG. In this case, as shown in FIG.
The cutout result is 55, and the recognition result 56 is (SP.
5) and (SC.5, 6). Here, (SC.5, 6) means that the partial pattern of the numeral "5" and the numeral 6 are in contact with each other, and is recognized by referring to the above-mentioned partial dictionary 102. From this result, the read character is
It can be determined that the numbers are “5” and “6”. Next, FIG.
A recognition method when the separated pattern components are in contact with each other as shown in (d) will be described with reference to FIG. In this case, as shown in FIG. 6A, two hypotheses 57 and 59 stand and recognition results 58 and 60 are obtained. Further, in this case, as shown in FIG. 6B, in particular, the subpattern 61, that is, 38 in FIG. 3D is individually recognized, and as a result, the recognition result 62 (SS. ) Is obtained.

The above-mentioned hypothesis 57 is a hypothesis 59 when the sub-pattern 38 is assumed to be added to the right side.
Is a case where the sub-pattern 38 is assumed to be added on the left side. In addition, recognition results 58 (SP.5) and (R
J.? ) Is “5 sub-pattern” and “reject (totally unknown)”, and similarly, (RJ.?) And (S
P.7) is "reject (totally unknown)" and "7 sub-pattern". Further, (SS.5, 7) of the recognition result 62 means "a pattern in which the sub-pattern of the numeral 5 and the sub-pattern of the numeral 7 are in contact". These are partial dictionaries 1 of patterns in which partial patterns and other characters are in contact.
No. 02 is obtained. By summing up these results, the answer is judged to be the numbers "5" and "7". Next, a recognition method in the case where perfect patterns are in contact with each other as shown in FIG. 3E will be described with reference to FIG. In this case, as shown in FIG. 7, the entire dictionary is sent to the recognition unit without being forcibly divided, and the partial dictionary 103
Look for and recognize the same. Here, as a result, (CC.5, 6) is obtained, which means that the numbers "5" and "6" are in contact with each other.

As described above, in the present invention,
The feature is that the final result is given by synthesizing the recognition results. Actually, for example, it is realized by processing according to the following rules. The processes performed on FIGS. 3A to 3E are summarized as follows. (a) (AC.5) (AC.6) → (AC.5) (AC.6) (b) (SP.5) (RJ.6) (AC.5) (AC.6) → (AC .5) (AC.6) (c) (SP.5) (SC.5,6) → (AC.5) (AC.6) (d) (SP.5) (RJ.?) (RJ. ?) (SP.7) (SS.5,7) → (AC.5) (AC.7) (e) (CC.5,6) → (AC.5) (AC.6) The recognition result code for each hypothesis on the left side of is rewritten to the recognition result code on the right side. A generalization of these will be called Rewriting Rules.

In the cutting method shown in this embodiment, the rewriting rule is as follows.

[Outer 1] Rule R1 corresponds to (a) and (b) shown on page 7. Even if there is a place that does not accept (recognize) a and b, there is another place that accepts it. That is to say that it was recognized.

Rule R2 corresponds to the same (c), and
When the contact between the sub-pattern of a and the pattern of b is recognized while the sub-pattern of 1 is recognized, it means that a and b can be recognized. Rule R3 is
It corresponds to the same (d), and a sub-pattern of a is recognized and a candidate of an arbitrary value other than accept is given, for example, while a sub-pattern of b is recognized and a candidate of an arbitrary value other than accept is given. Is given, it indicates that only subpatterns that are separated are to be recognized. Further, the rule R4 also corresponds to the same (d), and shows the case of comprehensive recognition including the recognition result of only the sub-pattern processed by the rule R3. That is, when three patterns, that is, a pattern that cannot be recognized as the sub-pattern of a, a pattern that cannot be recognized as the sub-pattern of b, and a pattern in which the sub-pattern of a and the sub-pattern of b are in contact, are recognized by comprehensive recognition, a Accept and b accept. Rule R5 corresponds to the same (e), and the pattern in which a and b are in contact is a accept or b accept.

FIG. 8 shows a block diagram of a character reading apparatus which is an embodiment of the present invention. The character reading device according to the present embodiment includes a pattern observing unit 800 and a pattern cutting unit 90.
0, form format dictionary 950, pattern recognition unit 20
0, the pattern dictionary 100, the recognition result final determination unit 400, and the recognition result rewriting rule dictionary 300. The operation of this device will be described below. On the form 75 to be read, characters are written under natural writing conditions as shown in FIG. The form 75 is the pattern observation unit 80.
When it is input to 0, and subjected to photoelectric conversion and preprocessing (binarization, form skew correction), the two-dimensional image pattern is sent to the pattern cutout unit 900 as an electric signal. In the pattern cutout unit 900, the form format dictionary 950
With reference to the frame position parameter from, the patterns judged to correspond to one character are cut out one by one from the image of one sheet and sent to the pattern recognition unit 200. In the pattern recognition unit 200, the input pattern for one character,
The respective patterns stored in the pattern dictionary 100 whose contents are illustrated in FIG. 4 are compared and collated, and the recognition result is sent to the recognition result final determination unit 400.

Needless to say, the processing in the pattern recognition section 200 includes the sub-pattern and the pattern of two characters that have touched as described above. Further, in the present embodiment, the recognition result output from the pattern recognition unit 200 is symbolized for each input pattern and sent to the recognition result final determination unit 400. The recognition result final determination unit 400 sequentially applies the rewriting rules in the recognition result rewriting dictionary 300 to the received recognition result until there are no applicable rewriting rules, and performs the rewriting process. That is, from the conditions of the above-mentioned rewriting rules R1 to R5, the one in which the above-mentioned symbolized recognition result matches this is selected and applied, and the result is adopted. The process of the pattern cutout unit 900 and subsequent processes will be described in more detail below. FIG. 9 exemplifies a flowchart of the above-described cutout processing and recognition processing and the content of data corresponding to this. In step 701, the video pattern 712 for one line is cut out from the video pattern 711 for one form.
In step 702, the continuity of the black background pattern is used to
After extracting the pattern component for each black background and performing ordering in the horizontal direction, a component list 713 is created.

Further, the attribute of each component is calculated and a component attribute list 714 is created. Here, the component attributes include the upper and lower ends of each component, the coordinates of the left and right ends, the total length of the contour, and the like. Next, in step 703, a hypothesis of a boundary between characters is set from the component attribute list 714 created above and the information in the form format dictionary 950 to create a character list 715. In this character list 715, each character pattern is
In FIG. 9, it is assumed that the first hypothesis is composed of one character in order 1, 2, and 3, and one character 4 and 5 are combined into one character. In the above, an example is shown in which it is assumed that the sequences 1 and 2 are combined to form one character, the sequence 3 is defined as one character, and the sequences 4 and 5 are combined to form one character. The above is the processing of the pattern cutout unit 900. Step 704 is processing of the pattern recognition unit 200. In step 704, the component list 713, the component attribute list 714, and the character list 715 described above are input, the components shown in the character list 715 are collected and pattern matching is performed, and the result is written in the result list 716.

The result code representing the matching result is the above-mentioned result code.
(SP.a), (SC.a, b), (SS.a, b), (CC.a,
b), (AC.a), (RJ.a), etc. The meanings of these are, as described above, “sub-pattern of category a”, “sub-pattern of category a and sub-pattern of category b”, and “sub-pattern of sub-categories a and b contact each other”, respectively. , “Category a and category b are in contact with each other”, “category a pattern”, and “candidate a is rejected”. Step 705 is processing of the recognition result final determination unit 400. Here, with respect to the above-mentioned result list 716, all rules in the rewriting rule dictionary 300 are referred to, they are sequentially applied until there are no applicable rules, and processing according to the finally obtained result is performed. In step 706, it is determined whether the processing has been completed for all the lines on the form 75. If not completed, the procedure returns to step 701, and the above-described processing is repeated until the processing is completed.

According to the above embodiment, when the recognizing unit determines that the inputted one-unit video pattern is a part of the pattern representing one category, a plurality of partial patterns are combined to form one character. Therefore, it is possible to perform character recognition on the combined characters, and it is possible to obtain the effect of improving the reading accuracy of characters written under natural writing conditions in the character reading device. It is needless to say that the above embodiment shows one example of the present invention, and the present invention should not be limited to this.

[0019]

As described above in detail, according to the present invention, a character written under natural writing conditions, such as a large protrusion from a character frame or a contact with an adjacent character,
In particular, when it is determined that the input 1-unit video pattern is a part of the pattern representing one category, it is possible to combine a plurality of partial patterns into one character and perform character recognition on the combined character. It has a remarkable effect that a pattern cutting and recognition method and its system can be realized.

[0020]

[Brief description of drawings]

FIG. 1 is a diagram showing an example of a character frame and a handwritten character of a conventional OCR form.

FIG. 2 is a diagram showing an example of a character frame and a handwritten character of a conventional OCR form when a writing condition is relaxed.

FIG. 3 is a diagram illustrating various states of adjacent character patterns.

FIG. 4 is a diagram (No. 1) for explaining the principle of the present invention.

FIG. 5 is a diagram (part 2) explaining the principle of the present invention.

FIG. 6 is a diagram (part 3) explaining the principle of the present invention.

FIG. 7 is a diagram (part 4) explaining the principle of the present invention.

FIG. 8 is a block diagram of a character reading device according to an embodiment of the present invention.

FIG. 9 is a diagram exemplifying a flowchart of cutout processing and recognition processing and a content of data corresponding to the processing in the character reading device according to the embodiment.

[Explanation of symbols]

75: form to be read, 100: pattern dictionary,
101 to 103: partial dictionary, 104: normal pattern dictionary, 200: pattern recognition unit, 300: recognition result rewriting rule dictionary, 400: recognition result final determination unit, 800: pattern observation unit, 900: pattern cutout unit, 950: form Format dictionary.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Shozo Kadota 2880 Kunizu, Odawara, Kanagawa Stock company Hitachi Ltd. Odawara factory (72) Inventor Kiyomi Kurino 2880, Kunizu, Odawara, Kanagawa Hitachi Ltd. Odawara factory (56) Reference JP-A-59-91582 (JP, A) JP-A-58-165179 (JP, A)

Claims (11)

[Claims] 1. A form that does not have a character frame for each character is entered.
2D image on the form converted into electrical signals
Separate from the image pattern for each unit image pattern and cut out
Step and one of the cut-out unit image patterns
Unit video pattern or combination of multiple unit video patterns
The recognition target video pattern that is the
Each pattern in the pattern dictionary for each elephant video pattern
Character pattern corresponding to the unit image pattern
A step of recognizing a turn and the recognition target image pattern
The recognition result obtained for the
Pre-defined rewriting with reference to the pattern recognition result
And a step of rewriting according to the rule,
Comprehensive character string patterns included in 2D video patterns
Pattern cutting and recognition characterized by
Knowledge method. 2. The recognition result is the recognition target image pattern.
A character that represents a character pattern that is recognized as a character
The pattern (category) name and the recognition target video pattern are
Is a complete pattern of character patterns or characters
It also includes information indicating whether the pattern is a partial pattern.
The pattern cutout according to claim 1, wherein
And recognition methods. 3. The rewriting rule in addition to the steps
The recognition target video pattern rewritten according to
When it is determined that it is a partial pattern of a character pattern
Is a plurality of unit image patterns including the recognition target image pattern.
For the turn combinations, again, in the pattern dictionary
A plurality of unit image patterns are compared with a pattern.
Recognition of character patterns corresponding to the combination of
3. The method according to claim 1, further comprising a step of
Described pattern cutting and recognition method. 4. One of the cut-out unit image patterns
Unit video pattern or set of multiple unit video patterns
When dividing into recognition target video patterns that are combinations,
If there is ambiguity between the
Create multiple hypotheses at the boundary between the
Recognition target video pattern obtained by performing classification corresponding to each
Recognition processing is performed on the
Contract characterized by selecting a single hypothesis from the theory
Pattern cut and according to any one of Motomeko 1 3
Recognition method. 5. The two-dimensional image pattern on the form
The step that separates each video pattern and cuts it out
Then, from the 2D image pattern on the form, one character line
The 2D image pattern is cut out at once, and the unit image pattern is extracted.
As a turn, the cut-out two-dimensional image for one character line
Characterized by extracting continuous black pixel areas from the pattern
The pattern cutout according to any one of claims 1 to 4.
And recognition methods. 6. The pattern dictionary includes one character pattern.
Character pattern together with the complete pattern
Pattern that constitutes a part of
Memorizes the contact pattern where the character and another character pattern contact
6. The method according to claim 1, wherein
Pattern cutout and recognition method. 7. A form that does not have a character frame for each character is entered.
From the two-dimensional image pattern that has been converted into an electrical signal
Input means for obtaining a unit image pattern, and by the input means
One unit from the obtained two-dimensional image pattern on the form
Video pattern or combination of multiple unit video patterns
Pattern cutout hand that cuts out the recognition target video pattern that is
Step and recognition target cut out by the pattern cutting means
Each of the video patterns, each pattern in the pattern dictionary
In comparison with the character pattern corresponding to the unit image pattern,
Pattern recognition means for recognizing a pattern and the pattern recognition hand
For each of the recognition target video patterns depending on the stage
The obtained recognition result is used to recognize adjacent unit image patterns.
Refer to the result and write according to a predetermined rewriting rule.
By replacing the two-dimensional image pattern on the form
Judgment that comprehensively judges the character string pattern included in
And a pattern cutout characterized by having means and
Recognition system. 8. The pattern recognition means is the recognition target.
Displays the character patterns recognized as corresponding to the video pattern.
The name of the character pattern (category) and the recognition target video pattern
Whether the turn is a complete pattern of letter patterns,
Is information indicating whether it is a partial pattern of a character pattern and
8. The pattern switching according to claim 7, wherein
Placement and recognition system. 9. The comprehensive judgment means complies with the rewriting rule.
Therefore, the recognition result of the rewritten recognition target video pattern
However, it is still judged to be a partial pattern of the character pattern.
If a plurality of recognition target video patterns are included,
For the combination of unit image patterns,
By comparing each pattern in the dictionary,
Character patterns corresponding to combinations of image patterns
9. The method according to claim 7 or 8, characterized in that
Pattern cutting and recognition system. 10. The pattern cutting means is configured to perform the recognition.
When cutting out the target video pattern, one character pattern
There is ambiguity in the separation of the recognition target video patterns that make up the
In some cases, there may be a plurality of borders between the unit image patterns.
A hypothesis is created, and the pattern recognition means uses the plurality of temporary
The recognition target video pattern corresponding to each theory
Recognition processing is performed by the
The feature is to finally select a single hypothesis from among the hypotheses.
The pattern cutting according to any one of claims 7 to 9
Placement and recognition system. 11. The pattern dictionary includes one character pattern.
Along with the complete patterns that make up the
Partial pattern and one character pattern that make up a part
Memorize the contact pattern where the character and another character pattern contact
11. The method according to any one of claims 7 to 10, characterized in that
Pattern cutting and recognition system.