JP2795222B2 - Character extraction method and character extraction device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character cutout method and a character cutout device in character reading for optically reading characters or the like printed on paper.

[0002]

2. Description of the Related Art Heretofore, this type of character cutout device cuts out characters as follows. First, a line orthogonal to the line is scanned in the direction of the character string to detect the shading of the line constituting the character. This is commonly called projection. Then, during scanning, the part of the cut where the detected density disappears is used as a character delimiter, the horizontal line from the break to the next break is used, and the character string with the height of the character as a vertical line is included. The coordinate information of the circumscribed rectangle to be extracted is extracted. Then, for example, the circumscribed rectangular block is divided using the character height as one unit, and a block of character candidates in units of one character is output.

Here, in the blocks obtained by the above, it may be difficult to obtain a block of a correct character candidate. That is, there are cases where one character is separated like a partial and is output as another block, or where adjacent characters are in contact and one block is output with a plurality of characters. However, in general, in the case of print type, the pitch of characters is almost constant, and
Even if a block of characters is separated from one side,
The two blocks are integrated so as to satisfy the estimated pitch of the character. As a result, a block of character candidates for one character can be obtained relatively easily and correctly.

[0004]

However, conventionally, there has been a problem that one character cannot be cut into one block in the following situation. That is, "river"
If one character separates into three and some of the characters touch or are very close to adjacent characters, such as
“River” cannot be output to one block due to the combination of blocks. In addition, even in block division based on pitch estimation, since the width of a portion close to the adjacent character is small, there is a problem that it is difficult to cut out only this portion and integrate it with another separated portion of "river". Similar characters include "small" and the like, because the directions in which the constituent parts easily contact each other differ depending on the shape of the character, and there has been a problem that it is difficult to cut out a correct character candidate.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-described problems, and has as its object to accurately cut out each character.

[0006]

According to the character extracting method of the present invention, a collating condition comprising a size of a circumscribed rectangle of each part constituting another part except for a part of a predetermined character and a positional relationship between the parts is defined. , Registration information composed of the processing contents when the matching condition is satisfied. Then, first, the character string captured as the image data is divided into blocks so as to be separated at the breaks of the image, and the blocks are extracted. Next, it is checked whether or not there is any one of the extracted blocks that matches the registration information. If there is a match that matches the registration information, the blocks are integrated in accordance with the processing content of the registration information. Then
From the solution obtained by dividing the horizontal length of these blocks by a natural number, the character pitch closest to the height of the character string is used as the character pitch, and the integrated block is divided for each character pitch to extract characters. Is performed. For this reason,
Even if it is one character, even if it is divided and extracted as a block, the divided block is integrated as one character.

The character extracting apparatus according to the present invention further comprises a block extracting means for extracting a block by dividing a character string taken as image data into blocks where the image is broken, and a part of a predetermined character. Registration information storage including registration information consisting of the size of the circumscribed rectangle of each part constituting the other part except for the above, the collation condition consisting of the respective arrangement relations, and the treatment content when the collation condition is satisfied Means, and matching means for checking whether or not there is a match in the registered information among the blocks extracted by the block extracting means, and if there is a match in the matching, according to the action content of the registered information. In the block integration unit that integrates the blocks and the solution obtained by dividing the horizontal length of the block integrated by the block integration unit by a natural number, Character pitch estimating means for setting the character pitch closest to the column height to character pitch, and block dividing means for cutting out characters by dividing the integrated block for each character pitch. . Thus, even if one character is divided and a block is extracted, the divided block is integrated as one character.

[0008]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a configuration diagram showing a configuration of a character cutout device according to one embodiment of the present invention. In the figure, 1 is a character string image input means for inputting image data of a character string, and 2 is a projection of the image of the input character string, and a block of images that do not touch each other is coordinate information of a circumscribed rectangle called a block. The extracting block extracting means 3 is a collating means for collating the extracted block with the registered information of the block pre-registered in the registered information storing means 3a. The collating means 4 is a block in a designated direction when the condition is satisfied by the collation. 5 is a character pitch estimating means for estimating the pitch of the blocks constituting the character string, and an average character pitch based on the height of the character string. 6 is a block for dividing the block by the estimated character pitch. The dividing means 7 is a block output means for outputting the divided blocks as character candidates.

The operation of the character segmenting apparatus according to this embodiment will be described below with reference to the flowchart of FIG. First, character string image data is input from the character string image input means 1 (step S1), and the block extracting means 2
Is performed to separate the blocks into blocks of images that do not touch each other, and this coordinate information is extracted (step S2). Here, in the description of this embodiment, the extraction of characters from character string image data as shown in FIG.

When the character string image is divided into blocks as described above, the result is as shown in FIG. In this block division, a portion in contact with the input character string is extracted as one block. Here, in a character whose constituent part is separated, such as “river”, a part of the character is included in an adjacent block, and another part is a single block.

Next, a start block to be compared is specified (step S3), and the matching unit 3 compares the registered information stored in the registered information storage unit 3a with the extracted block (step S4). FIG. 4 is an explanatory diagram showing the configuration of the registration information stored in the registration information storage unit 3a of FIG. FIG. 4 shows registration information relating to a character whose assumed object (assumed category) is “river”, registration information relating to a character whose assumed object is “small”, and another registration information relating to a character whose assumed object is “small”. It shows the state where it is. For example, in the registration information with the assumed character “kawa”, as indicated by the character “kawa”, one character is composed of three parts, which are separated, and the left part is close to the left character. It is assumed that you are in the state.

This assumption is based on the assumption that the left part of the character “kawa” is divided into blocks including the left character, and the central part of the character “river” and the right part of the character “river” are one block each. This is the case. Since the character "river" has a curved portion on the left side, if the character on the left side approaches, it is easily recognized as being included in that block. On the other hand, the right part of the “river” is composed of straight lines, so that even if it is close to the character on the right, it is easily separated and recognized.

[0013] Further, regarding the character "small", both the left part and the right part constituting it are easily recognized as being included in the block when another character is approaching. For this reason, registration information assuming that the right part is included in the right character block and registration information assuming that the left part is included in the left character block are required. .

Next, a detailed configuration of the registration information will be described. The registration information includes an assumed category indicating an assumed target character, a target block, a condition to be matched, and processing to be performed when the condition is satisfied. The conditions for collation are composed of block 1 information and block 2 information, and relative position information defining the relative positional relationship between these two blocks.

For example, block 1 of the assumed category "river"
The information is the height H1 and width W1 of the target block 1
Is used to determine whether the values match the prescribed values h1 and w1. Block 2
The information is a block 1 that is checked against the block 1 information
Is to determine whether the height H2 and the width W2 of the target block 2 match the prescribed values h2 and w2.

In the relative position information, the relative positional relationship between the block 1 and the block 2 is defined. That is, the distance G1 between the block 1 and the left block and the distance G2 between the block 1 and the block 2 are defined values g
It is determined whether or not the value matches 1 and the prescribed value g2. In addition, Block 1 and Block 2
Is to determine whether or not the vertical positional relationships A1 and A2 with the specified values a1 and a2 match. In the above description, as shown in FIG. 4, the specified value has a width, but this unit is the number of pixels, which is about 6%.

In the collation in step S4 (FIG. 2), the above-mentioned registration information is used, and it is determined whether or not those conditions are satisfied (step S5). Here, whether or not there is a block satisfying the registration information is checked for all the extracted blocks. Here, when there is a block satisfying the registration information (FIG. 3C),
The blocks are integrated in the direction indicated by the corresponding action of the registration information (step S6).

For example, in the case of the example shown in FIG. 3, there is a block which satisfies the matching condition in the registration information of the assumed category "river" in FIG. Therefore, as shown in the corresponding action, the block 1 including the central portion of the “river” and the block 2 including the right portion are integrated into the block to the left of the block 1. Then, based on the determination in step S7, it is determined whether there is another block that satisfies the registration information (step S8), and the blocks are integrated as described above (FIG. 3D).

As a result, when the collation is completed for all the blocks and the blocks are integrated (step S
8) The character pitch estimating means 5 estimates an average character pitch (step S9). The estimation of the average character pitch is a value in which the width of the block of the character string is an integral multiple thereof, and is a value closest to the height of the character string. That is, for a certain block, its width is divided by a natural number, and the value closest to the height of the character string in the obtained solution is adopted. Finally, as shown in FIG. 3E, the block is divided according to the estimated character pitch (step S10), and FIG.
As shown in (1), it is output as a character candidate block (step S11).

As described above, as shown in FIG. 3B, even if the character "kawa" has not been extracted as a single block, the collation using the registration information shown in FIG.
It is integrated into one block (FIG. 3D). And
When the state where one character is divided into different blocks is resolved, the character is divided into individual characters from the estimated character pitch, so that correct character candidates can be output.

[0021]

As described above, according to the present invention, the collation conditions based on the size of the circumscribed rectangle of each part constituting the other part except for one part of the predetermined character and the respective arrangement relations are obtained. , And the registration information composed of the processing contents when the collation condition is satisfied. Then, if there is a match with the registered information, the blocks are integrated according to the treatment of the registered information. For this reason, even when a part is included in another block and the other part is extracted as one block, such as the character “river”,
As a result, there is an effect that the character can be cut out as one character by block integration.

[Brief description of the drawings]

FIG. 1 is a configuration diagram illustrating a configuration of a character cutout device according to an embodiment of the present invention.

FIG. 2 is a flowchart showing the operation of the character segmenting device of the present invention.

FIG. 3 is an explanatory diagram showing an example of character string image data from which characters are cut out.

FIG. 4 is an explanatory diagram showing a configuration of registration information stored in a registration information storage unit 3a of FIG. 1;

[Explanation of symbols]

1 ... Character string image input means, 2 ... Block extraction means, 3 ...
Collation means, 3a registration information storage means, 4 block integration means, 5 character pitch estimation means, 6 block division means, 7 block output means.

──────────────────────────────────────────────────続 き Continued on front page (58) Field surveyed (Int.Cl. ⁶ , DB name) G06K 9/34

Claims (2)

(57) [Claims] 1. A character extracting method for capturing a character string as an image and extracting image data corresponding to one character from each of the captured images, wherein each of the characters constitutes another part except a part of a predetermined character. The character string captured as image data is provided with registration information consisting of matching conditions consisting of the size of the circumscribed rectangle of the part and the positional relationship of each part, and the treatment details when the matching conditions are satisfied. The block is divided so as to be separated where it is extracted, and a block is extracted.In the extracted block, it is checked whether or not there is a match with the registration information. Blocks are integrated according to the treatment of the registration information, and the solution that is closest to the height of the character string is the solution obtained by dividing the horizontal length of the block by a natural number. Character segmentation method for the the character pitch, and performs clipping of a character by dividing the block was integrated for each of the character pitch. 2. A character cutout device for taking in a character string as an image and extracting image data corresponding to one character from each of the captured images, wherein the character string captured as image data is separated at a break in the image. Extracting means for extracting blocks by dividing blocks as described above, collating conditions consisting of the size of the circumscribed rectangle of each part constituting other parts except for a part of a predetermined character and their respective arrangement relations, A registration information storage unit having registration information composed of the contents of a treatment when a matching condition is satisfied; and checking whether there is a block that matches the registration information among the blocks extracted by the block extraction unit. If there is a match found by the matching means and the matching by the matching means, the block is integrated in accordance with the action of the registered information. And a character pitch estimating unit that sets a character pitch closest to the height of the character string in a solution obtained by dividing the horizontal length of the block integrated by the block integrating unit by a natural number. And a block dividing means for dividing the integrated block for each character pitch to extract characters.