JPH07282194A - Character recognizing device - Google Patents

Abstract

PURPOSE:To provide a character recognizing device which can execute a highly precise discrimination in a short processing time by shaping locally only a part showing a feature effective for discriminating from another character by a method suitable to a character kind even if a character pattern is blurred or broken. CONSTITUTION:As for the character pattern whose pattern is judged to need to be shaped among candidate characters outputted from a first discriminating part 32, the shaping of the pattern is executed by a local shaping part 332 on the basis of corresponding shaping information in a shaping information table 333, and afterwards, feature extraction is executed again by a second feature extracting part 335, and detailed discrimination is executed by a second discriminating part 34, and therefore, the discrimination can be executed highly precisely even for the blurred or broken character pattern. Besides, since an input part 334 to enable the shaping information in the shaping information table to be designated is provided the flexible and more highly precise discrimination becomes possible.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition device for recognizing a character based on image data such as a document written on an object to be read.

[0002]

2. Description of the Related Art Conventionally, character recognition devices such as OCR have been described in literature a: "Japanese Patent Publication No. 60-38756", b: "Japanese Patent Publication No.
62393 ”, a document or the like written on an object to be read is optically read and the character portion is binarized and stored in a pattern register. For example, a character line is extracted from a stroke component in each direction. The feature quantity representing the length is normalized by the size of the character to create a feature matrix, which is compared with a dictionary created in advance to perform the first identification, and when matching with this dictionary, two or more similar characters are compared. Is obtained as a candidate, a second identification different from the first identification is performed based on the feature amount that often shows the difference between these similar characters, and an appropriate one is selected from the candidate characters. The method of outputting was adopted.

[0003]

However, in such a character recognition device, when the character pattern has a blur or a collapse for some reason, the accurate detailed characteristic of the character pattern is determined in the second identification. It is not possible to obtain it, which causes the recognition rate to decrease. For example, even if the center line or the contour line of a portion having a character pattern is traced to obtain the curvature or the like, if the center line, the contour line, or the like is missing in the middle, it is difficult to obtain the value for the portion. Further, when trying to obtain the number of loops in a character pattern or the like, if the pattern is faint and a loop is generated, or if the pattern is destroyed and the loop disappears, the original number of loops cannot be obtained.

Conventionally, as measures against such a case, in addition to processing such as uniform noise removal for image data obtained by the reading unit and uniform shaping for a character pattern before feature extraction,
Various measures have been taken, such as creating an algorithm for obtaining detailed features by assuming the effect of faintness or crushing in advance and responding. However, uniform noise removal and shaping processing for image data requires a lot of processing time, and
Not only is an identification algorithm that assumes blurring or blurring very redundant, but it is also not flexible to deal with the blurring or blurring of patterns that change irregularly depending on the characteristics of photoelectric conversion, paper quality, and the state of printing or writing. There was a problem that it could not be done.

Therefore, an object of the present invention is to locally localize only a portion showing a characteristic effective for distinguishing from other characters even if the character pattern is faint or crushed by a method suitable for the character type. An object of the present invention is to provide a character recognition device that can be shaped and can perform highly accurate identification in a short processing time.

[0006]

SUMMARY OF THE INVENTION In order to solve the above problems, the present invention provides a reading unit for obtaining image data of an object to be read, and an identification unit for extracting and identifying a characteristic of a character pattern from the image data. In the character recognition device, the identification unit includes a first feature extraction unit that extracts a feature effective for identification from a character pattern, and a first identification unit that outputs one or more candidate characters based on the feature. A local shaping unit that shapes a specific part in a character pattern, a shaping information table that holds shaping information that includes flag information indicating whether or not there is a need for shaping for each character code, information about the shaping part, and information about the shaping method. An input unit that can specify the shaping information; a shaping unit that includes a second feature extraction unit that extracts a feature from the shaped character pattern; and a candidate character or a candidate character after pattern shaping. Characterized in that a second identification unit for selecting and outputting an optimal character from.

[0007]

According to the present invention, with the above-described configuration, the character pattern that is determined to be required to be shaped in the candidate characters output from the first identification unit is stored in the shaping information table. After the local shaping unit shapes the pattern based on the shaping information of, the second feature extraction unit performs feature extraction again, and the second identification unit performs detailed identification. Patterns can also be identified with high accuracy. Also,
By providing the shaping section with an input section that allows the shaping information in the shaping information table to be specified, the characteristics of photoelectric conversion, paper quality,
It is possible to freely specify the flag information indicating the necessity of shaping for each character, the shaping portion, and the shaping method according to the printing state or the writing state, and thus it is possible to perform flexible and more accurate identification. Therefore, the said subject can be solved.

[0008]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of a character recognition device of the present invention will be described below with reference to the drawings. It should be noted that the drawings are only schematically shown to the extent that the present invention can be understood, and therefore the shapes, arrangements, and connection relationships of the respective constituent components are not limited to the illustrated examples. FIG. 1 is a block diagram showing the configuration of the character recognition device of the present invention. The character recognition device 1 is composed of, for example, an OCR, and is for reading and recognizing characters written on an object to be read and for promptly performing an input operation to a computer or the like.

The character recognition device 1 mainly includes a reading unit 2 for obtaining image data from an object to be read, and an identifying unit 3 for extracting and identifying a characteristic of a character from the obtained image data.
It consists of and. The reading unit 2 captures the reflected light of a character and converts it into an electric signal to obtain, for example, binarized image data, and an image memory for storing the binarized image data. 22 and 22.
In addition, the identification unit 3 extracts a feature necessary for identification from image data, and a first identification unit 32 that outputs a candidate character by obtaining a similarity, for example, from the obtained features.
And a shaping unit 33 that shapes a pattern of a specific part according to the type of the candidate character, and a local shape or topological feature of the center line or contour line of the pattern, for example, is examined to select an appropriate character from the candidate characters. And a second identification unit 34 that selects the one and outputs the result. Further, the shaping unit 33 holds, for each character type, a local shaping unit 332 that expands or contracts a pattern of a region effective for identification according to the type of the candidate character, and information regarding a region and a method for performing the shaping process. Shaping information table 333, an input unit 334 that specifies the shaping site and method, a second feature extraction unit 335 that extracts features from the shaped pattern, and a candidate as to whether or not the pattern needs to be shaped. And a control unit 331 for judging from the character and shaping information table 333.

In order to recognize a character written on an object to be read by using the character recognition device 1, the photoelectric conversion unit 21 converts the reflected light of the character written on the object to be read into an electric signal. The binarized image data is stored in the image memory 22. Then, the feature extraction unit 31 extracts the feature of the character pattern from this image data, and then the first identification unit 32.
Determines one or a plurality of candidate characters by obtaining the degree of similarity by pattern matching, stroke analysis, or the like.

FIG. 2 is a diagram for explaining how patterns are shaped according to the candidate characters obtained by the first identifying section 32, and illustrates the format of the candidate characters and the shaping information table 333. The format of the shaping information table 333 is composed of shaping information records 51 described for each character code. In each record, a character code 52, a shaping flag 53 indicating whether or not to be shaped, and part information 54 indicating a part to be shaped. And, for example, method information 55 indicating a shaping method such as expansion or contraction. For example, 1 is set in the shaping flag 53 if it is necessary to perform shaping, and 0 is set if it is not necessary. The candidate characters 4 are arranged in order from the one having the highest degree of similarity, for example, and are output from the first identification unit 32. Control unit 3
For example, 31 uses the character code of the candidate character 4 as a key item to search the shaping information record 51 in the shaping information table 333, and checks the value of the shaping flag 53 in the corresponding record. Here, when the shaping flag 53 in the shaping information table 333 corresponding to all the characters thus searched is 0, that is, when it is not necessary to shape the pattern, the second
Move to the processing of the identification unit.

On the other hand, if there is a shaping flag 53 of 1, that is, if it is necessary to shape the pattern, the local shaping unit 332 expands, contracts, etc. the specific portion indicated by the portion information 54 to the specific portion indicated by the method information 55. Shape by the method of. The local shaping unit 332 may hold an image memory for pattern shaping. FIG. 3A shows an example of a character pattern [e] that requires shaping processing. When the character pattern 61 before shaping and the shaping portion 62, and the horizontal and vertical coordinates are x and y. Upper left point (x1, y1) of the shaped part
63, a lower right point (x2, y2) 64, and a shaped pattern 65. FIG. 3B is an example of the shaping information record in the case of performing the pattern shaping as shown in FIG. 3A, for example, the character code 71, the shaping flag 72, the coordinates 73 of the upper left point of the shaping site, The coordinates 74 of the lower right point, the code 75 indicating the expansion process, the number 76 of expansion processes, the code 77 indicating the contraction process, and the number 78 indicating the contraction process are included.

For example, if there is a blur such as the pattern 61 in FIG. 3A for some reason, the point at the upper left of the portion to be shaped in the shaping information record storing the information about [e] in the shaping information table. Coordinate 73 and the lower right store coordinate 74 and, for example, a code 75 indicating expansion processing.
And the number of times 76, the code 77 indicating the contraction process, and the number of times 78 are specified from the input unit 334. The local shaping unit 332 performs processing based on the information of this shaping information record and generates a pattern 65. Here, the expansion processing is represented by a binary value of 1 for the character portion and 0 for the background portion, and when the target pixel is 1 when the region to be processed is sequentially scanned from the upper left, for example, all 8 neighboring points are set to 1 Is a process like
The contraction process is a process for setting all 8 neighboring points to 0 when the pixel of interest is 0. The number of times of expansion and contraction processing is the number of times the target area is scanned while performing each processing.

Here, the above-described retrieval processing of the shaping information table 333 may be executed for all candidate characters, or may be executed for only a part of the candidate characters. For example, one or a plurality of characters having a similarity higher than a predetermined threshold may be selected to search the shaping information table, or a predetermined number of characters may be searched from the one having the highest similarity. In addition, "c" and "e" are included in the candidate characters.
The shaping information table may be searched for these characters only when there is a pair of characters that may be erroneously identified, such as. Also, the designation of the shaping part is
A method other than designation by the coordinates of the upper left point and the lower right point may be used. Further, the local shaping section 332 may perform shaping such as expansion and contraction only in a specific direction within the pattern. In this case, the method information 55 has information on the processing direction. In addition, the local shaping unit 332 may perform pattern shaping other than the method described above.

When the shaping of the pattern of one or more candidate characters is completed, the second feature extraction unit 335 extracts features such as the shape of the center line or the contour line of the pattern, the line width, or the phase. The second identification unit 34 uses, for example, a pre-created dictionary having the same format as the features obtained by the first feature extraction unit 31 or the second feature extraction unit 335 for a standard character shape,
For the candidate characters output from the first identification unit 32, the features are collated with the dictionary.

The input unit 334 has characteristics of the photoelectric conversion unit 21,
The contents of the shaping flag 53, the part information 54, and the method information 55 in the shaping information table 333 can be specified according to the paper quality, the state of printing or writing, and the like. For example, the number of expansion processes 76 and the number of contraction processes 78 may be increased or decreased, or the part to be shaped may be changed depending on the degree of blurring or collapse of the character pattern.

[0017]

As described above, according to the character recognition apparatus of the present invention, a reading section for obtaining image data of an object to be read, and an identification section for extracting and identifying a characteristic of a character pattern from the image data. In the character recognition device, the identification unit includes a first feature extraction unit that extracts a feature effective for identification from a character pattern, and a first identification unit that outputs one or more candidate characters based on the feature. A local shaping unit that shapes a specific part in a character pattern, a shaping information table that holds shaping information that includes flag information indicating whether or not there is a need for shaping for each character code, information about the shaping part, and information about the shaping method. An input unit that can specify the shaping information, a shaping unit that includes a second feature extraction unit that extracts a feature from the shaped character pattern, and the candidate character or the candidate character after the pattern shaping. Causing the following effects by providing a second identification unit for selecting and outputting et optimum character. That is, even if the character pattern to be identified has faintness or crushing, after the pattern is locally shaped according to the candidate character as necessary, the shape or position of the center line or contour line, or topological Features and the like can be examined in detail and identified. Further, the shaping site, method, etc. can be freely designated for each character type in accordance with the characteristics of photoelectric conversion, the quality of paper, the state of printing or writing, and the like. Therefore, it is possible to flexibly deal with the quality of the character pattern and improve the recognition accuracy without significantly increasing the processing time.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a character recognition device of the present invention.

FIG. 2 is an explanatory diagram of a shaping information table.

FIG. 3 is an explanatory diagram of local shaping of a character pattern.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 character recognition device 2 reading unit 3 identification unit 3 21 photoelectric conversion unit 22 image memory 31 first feature extraction unit 32 first identification unit 33 shaping unit 34 second identification unit 331 control unit 332 local shaping unit 333 shaping information table 334 input Part 335 Second feature extraction unit

Claims (1)

[Claims] 1. A character recognition device comprising: a reading unit for obtaining image data of an object to be read; and an identification unit for extracting and identifying a characteristic of a character pattern from the image data. A first feature extraction unit that extracts a feature effective for identification, a first identification unit that outputs one or more candidate characters based on the feature, and a local shaping unit that shapes a specific portion in a character pattern. A shaping information table that holds shaping information that includes flag information indicating whether or not shaping is required for each character code, information regarding a shaping site, and information regarding a shaping method, an input unit that can specify the shaping information, and shaping A shaping unit including a second feature extraction unit that extracts a feature from the subsequent character pattern, and a second identification unit that selects and outputs an optimum character from the candidate character or the candidate character after pattern shaping. Character recognition device which is characterized in that there was example.