JPH02230484A - Character recognizing device - Google Patents

Abstract

PURPOSE:To recognize a character by segmenting correctly even when recognizing a Japanese language text in which a KANJI(Chinese character) of em and an alphanumeric and a mark of en are mixed by judging the character by collating with a separating character dictionary means when the small part of a separating character is inputted as a partial character string. CONSTITUTION:A character recognizing means 24 collates character data segmented by a character segmenting means 23 with a character recognition dictionary means 27, and recognizes the character including the character part of the separating character. A separating character dictionary means 26 stores relation between the part and the whole of the separating character as a separating character dictionary, and a separating character recognizing means 25 judges whether or not an inputted partial character string is the separating character by referring to the separating character dictionary means 26, and outputs the said separating character as a candidate character of first rank with the highest convictionality, and outputs the partial character string as that of second rank. Thereby, it is possible to accurately recognize the Japanese language text in which the KANJI of em and the alphanumeric and the mark of en are mixed.

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention relates to a computer application product that inputs document image data of a Japanese document and performs character recognition, and is particularly applicable to Japanese word processors, type printing, etc. The present invention relates to a character recognition device suitable for recognizing Japanese printed documents created by the Japanese language, which contain a mixture of full-width kanji and half-width numerals and symbols.

[Conventional technology]

Conventionally, the technology commonly used for OCR devices is to print character frames on special paper in advance, and then separate and recognize the characters. A typical example of this type of device is a postal code reading device. on the other hand,
Recently, OCR devices that process general documents without character frames have been developed, and various attempts have been made to develop techniques for cutting out characters without character frames.

Here, the cutting out of lines and characters is realized, for example, by creating His1 Duram by projecting black pixels. In this case, accurate cutting is possible when the character pitch and character size are ideally uniform, such as in some types of printed matter. However, in reality, the pitch of characters is uneven, as in recent phototype printing, and in most cases, even in typesetting, full-width and half-width characters are mixed in a complex and irregular manner, making it difficult to cut out accurately. It was extremely difficult to do so.

Particularly problematic in the case of horizontally written documents are the characters included in full-width kanji such as "ka'", "wa+2rrmon+2ubetsu",
Interest ++, u river JT, LL measurement ++,'l soku n, u inspection 77%
Characters in which a single character is composed of multiple character parts that can be separated vertically (hereinafter referred to as separable characters), such as ``, ri'', and ``ru'''', and II (IPZ I
This is a document containing a mixture of consecutive half-width character strings such as IIHIZ l112+I, rr abcu. These characters are cut out in various ways, such as being separated or fused, depending on the printing conditions and font type. Furthermore, in some cases, a portion of non-separable characters such as "naka u, u ware" may be blurred and separated.

On the other hand, even in the case of vertically written documents, there are horizontally separated characters such as "2n, u3", and half-width characters are often used as they are in vertically written documents, so a similar problem exists. do. However, in general, processing of horizontally written documents is more serious in terms of quantity.

In response to the above situation, methods have been proposed to extract character strings with uneven pitches using the average pitch of characters that have already been extracted. As long as cutting is performed only by width, no matter what method you use, if the pitch or the spacing between the character parts of separated characters happens to be exactly the same as consecutive half-width characters, but it is extremely difficult to distinguish, a cutting error will occur. obtain.

[Problem to be solved by the invention]

As mentioned above, the conventional technology does not give sufficient consideration to the recognition of Japanese documents that contain a complex mixture of full-width kanji and half-width alphanumeric characters and symbols, and the extraction process using the above-mentioned conventional technology is There is a problem in that characters that should not be separated are separated and cut out, and are mistakenly recognized as similar characters, or conversely, consecutive half-width characters are cut out as separated characters and mistakenly recognized.

An object of the present invention is to provide a character recognition device that can accurately recognize Japanese documents containing a complex mixture of full-width Chinese characters and half-width alphanumeric characters and symbols.

[Means to solve the problem]

In order to achieve the second object, the present invention provides image data storage means for storing character image data, means for reading out character image data from the image data storage means, and all characters to be recognized and characters to be recognized. A character recognition dictionary means stores information necessary for character recognition regarding character parts of all separated characters included in the text as a dictionary, and the read character image data is compared with the character recognition dictionary means to identify all characters to be recognized. and a character recognition means for performing character recognition on partial characters of all separated characters included in the character to be recognized;
separated character dictionary means for storing a relationship between a part of a separated character and the whole as a separated character dictionary; activated when a small part of a separated character is input as a partial character string to a character recognition means;
It is determined whether the input partial string is a separated character string by comparing it with a separated character dictionary means, and if it is determined to be a separated character, the separated character is output as the highest candidate character with the highest degree of certainty. The present invention proposes a character recognition device equipped with a separated character recognition means for outputting a partial character string as a second candidate character.

The character image data reading means reads the image data from the image data storage means, and when one character is a separated character consisting of a plurality of small parts, in the case of a horizontally written document, in the vertical direction, and in the case of a vertically written document. may include character cutting means for cutting out character data by cutting out each small part in the horizontal direction, and in the case of a non-separable character consisting of a single small part, cutting out the character data as a whole.

The character image data reading means may further include, in front of the character cutting means, line cutting means for reading image data from the image data storage means and cutting out character line data.

In either case, the separated character recognition means may have a function of outputting both the combined separated character and the separated partial character string as candidate characters as recognition results when recognizing the separated characters.

The separated character recognition means also separates a candidate character as a full-width or double-width character and a candidate character as a half-width character string when a character to be recognized is regarded as one full-width or double-width character by fusion of half-width characters. It is also possible to provide a function to output both.

These character recognition devices can be incorporated into computer systems, optical character recognition devices, image scanners, and the like.

[Effect]

The image data storage means stores image data input from an image input device such as an image scanner that inputs character image data. The line cutting means extracts, from the image data,
Extract character line data.

If one character is a separated character consisting of multiple small parts from this character line data, the character cutting means cuts out each small part in the vertical direction in a horizontally written document, and in the horizontal direction in a vertically written document, and separates each small part into a single character. In the case of a non-separated character consisting of a small part, the entire character is extracted and the character data is extracted. The character recognition dictionary means stores, as a dictionary, information necessary for character recognition regarding all characters to be recognized and partial characters of all separated characters included in the characters to be recognized. The character recognition means compares the character data cut out by the character cutout means with the character recognition dictionary means and recognizes characters including character parts of separated characters. The separated character dictionary means stores the relationship between the part of the separated character and the whole as a separated character dictionary. The separated character recognition means is activated when a small part of the separated character is input as a partial character string to the character recognition means, and refers to the separated character dictionary means to determine that the input partial character string is a separated character. If it is determined to be a separated character, the corresponding separated character is output as the first candidate character with the highest degree of certainty, and a partial character string is output as the second candidate character.

〔Example〕

Hereinafter, a practical example of the present invention will be explained with reference to the drawings.

In FIG. 1, image data input from an image input device/are stored in image data storage means, . 2/ is stored as black and white binary pixel data. Line cutting means. ，! One is an image data storage means using a method such as creating projection data (histogram) in a direction parallel to the rows. 2/
Line data is cut out from the image data storage means, . 2/ as the values of the upper left and lower right pixel addresses of the rectangle representing the row data storage area. , 23.

Character extraction means. , 23 stores image data storage means . Extract the image data corresponding to that line from 2/, and use the same method as line extraction such as histogram to extract the vertical direction for each character, in the case of a horizontally written document.
In the case of a vertically written document, cut out each small part in the horizontal direction,
In the case of a non-separable character consisting of a single small part, the character data is sequentially cut out by cutting out the whole character, and the character data is stored in, for example, an image data storage means. ，! Character recognition means as the values of the upper left and lower right pixel addresses of the rectangle representing the character data storage area within ．． Output to 2 das.

Character recognition means. ．． 2da uses this character data to store image data. ．． The image data corresponding to the character is extracted from 2/, subjected to pre-processing such as normalization and dust removal, and then processed using a pre-prepared character recognition dictionary means. , 27,
Character recognition dictionary means. ．． Character recognition dictionary means performs character recognition processing using a character recognition method such as pattern matching with 27, and determines that the character data is similar to the input character data. ．． 2
Character recognition means for separating character candidates within 7. Output to 26.

FIG. 2 shows an example of the configuration of the separate character dictionary dB for a horizontally written document. This dictionary includes full-width kanji, ″
II, u and IZLL gate r2rr, II, LL interest'',''
A single character is made from multiple character parts that can be separated vertically, such as kawa II, LL measurement n, u so n, u test r+, rrn n, u ni''', '゛ri', and 'ru'. Contains all separators. This dictionary ρB contains a separate character part, 2/0, and a separate character code for each separated character. 2.20 and corresponding character parts. It is composed of 2JO, with a separated character part. 2
/0 makes it possible to search. Separated character part. 2/
0, the leftmost character part of the separator character is stored as a special code that is distinguished from a normal computer character code, such as a shift JIS code, and the separator code. 2
．． In 20, the corresponding separator character is stored as a computer character code, for example, a shift JIS code.
Corresponding character part. For 2JO, the character part on the right following the leftmost part of the corresponding separator character is the separator part. ．． It is stored as a special character code similar to 2/O as shown in the example.

Character recognition dictionary means. ．． 27 is the separated character part of the separated character dictionary. Character part corresponding to 2/0. :l. All character parts included in 30 are stored as the above-mentioned special code, and are distinguished from normal full-width or half-width characters.

Next, the separated character recognition means is shown in FIG. ．． The operation of 2.5 will be explained. Separate or character recognition means. 2.5 indicates that the input character candidates are separated character dictionary means. 2A (step A), and if not registered, the character candidate is output as is as a recognition result (step B-FGE).

If the character candidate is registered in the separated character dictionary means βB, the character candidate is treated as the character part of the separated character and the separated character recognition means. 2 and 5 (steps A to H), and waits for the input of a subsequent character candidate that is not determined to be a separate character part. Separate character recognition means when consecutive character candidates are held as character substrings of separate characters.
．． 2.5 is a separate character dictionary means. 2B is referred to (steps A-B-C), and if the character substring exists in the dictionary, the corresponding separation character is output as the first candidate character for the character substring (step D). . in this case,
The character substring itself is also output as the second candidate character (steps DE). On the other hand, if the character substring does not exist in the dictionary, the character substring is output as is as a candidate character string (step CG-E).

FIG. 4 shows an example of outputting candidate characters in a document containing both separated characters and half-width characters.

In the above embodiment, line cutting and character cutting are executed before proceeding to character recognition, but character recognition may be performed by directly reading character image data, or character recognition may be performed by omitting line cutting. You can also perform character recognition after.

Furthermore, when a separated character is recognized, both the combined separated character and the separated partial character string can be used as candidate characters in the recognition result.

In addition, when the recognition target character is treated as a single full-width or horizontal double-width character by fusion of half-width characters,
It is also possible to output both candidate characters as full-width or double-width characters and candidate characters as half-width character strings.

A system equipped with these character recognition functions can be realized not only as a computer system but also as an optical character recognition device (so-called OCR) or an image scanner.

〔Effect of the invention〕

According to the present invention, even when recognizing a Japanese document containing a mixture of full-width kanji and half-width alphanumeric characters and symbols, separated characters are correctly extracted and recognized as separated characters, and consecutive half-width characters are correctly recognized as half-width characters. It can be cut out and recognized. In addition, even when it is difficult to uniquely perform segmentation and recognition, both the separated character and the partial character string that makes up the separated character are output as candidate characters, improving the recognition rate of the character recognition device. Since the confirmation and correction processing time of character recognition processing can be shortened, the final recognition accuracy is increased.

The method of the present invention is not limited to the recognition of printed character documents, but can be applied to the recognition of handwritten characters, and the same effect can be obtained if a dictionary of separated characters that are likely to occur is prepared.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the overall configuration of an embodiment of a character recognition device according to the present invention, FIG. 2 is a diagram showing an example of the configuration of a separate character dictionary, and FIG. 3 is a diagram showing the processing flow of the separated character recognition means. The flowchart shown in FIG. 4 is a diagram showing an example of outputting candidate characters for a document including separated characters and half-width characters. / - Image input device, 2... electronic computer, . ．． 2/
- Image data storage means, . ．． 2ρ...Line cutting means,
．． ，! 3 character extraction means, . 2 da...Character recognition means, . ，! 6 separate character recognition means; ，! B ・Separate character dictionary means, ,27 ・Character recognition dictionary means, . 2/0・
・Separated character part, . ,2.20...Separator character code,
．． 230 - Corresponding character part.

Claims (1)

[Claims] 1. A character recognition device that recognizes character information from character image data, comprising: image data storage means for storing the character image data; means for reading the character image data from the image data storage means; character recognition dictionary means for storing, as a dictionary, information necessary for character recognition regarding all characters to be recognized and character parts of all separated characters included in the characters to be recognized; character recognition means for performing character recognition on all characters to be recognized and partial characters of all separated characters included in the characters to be recognized by comparing them with a recognition dictionary means; Separated character dictionary means for storing a separated character dictionary, and activated when a small part of the separated character is input as a partial character string to the character recognition means, and is activated to determine whether the input partial character string is a separated character string. is compared with the separated character dictionary means, and if it is determined to be a separated character, the separated character is output as the first candidate character with the highest degree of certainty, and the partial character string is outputted as the second candidate character. A character recognition device comprising a separate character recognition means. 2. In the character recognition device according to claim 1, the character image data reading means reads the image data from the image data storage means, and when one character is a separated character consisting of a plurality of small parts, horizontal writing is performed. Includes character cutting means for cutting out character data by cutting out each small part in the vertical direction in the case of a document and in the horizontal direction in the case of a vertically written document, and cutting out the entire character in the case of a non-separated character consisting of a single small part. A character recognition device characterized by: 3. In the character recognition device according to claim 2, the character image data reading means includes a line cutting means for reading the image data from the image data storage means and cutting out character line data before the character cutting means. A character recognition device characterized by: 4. The character recognition device according to any one of claims 1 to 3, wherein when the separated character recognition means recognizes the separated character, the separated character string is combined with the separated character string. A character recognition device characterized by having a function of outputting both as candidate characters as recognition results. 5. In the character recognition device according to any one of claims 1 to 4, when the character to be recognized is regarded as a single full-width or horizontal double-width character by a fusion of half-width characters, A character recognition device having a function of outputting both candidate characters as full-width or double-width characters and candidate characters as half-width character strings. 6. A computer system comprising the character recognition device according to any one of claims 1 to 5. 7. An optical character recognition device comprising the character recognition device according to any one of claims 1 to 5. 8. An image scanner comprising the character recognition device according to any one of claims 1 to 5.