JP3197441B2 - Character recognition 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 recognition apparatus used for an optical character reader (OCR) and the like, and more particularly to a character recognition apparatus for cutting out characters one by one from an input document and recognizing the characters.

[0002]

2. Description of the Related Art In general, an optical character reading apparatus reads a document image such as printed characters printed on a document or the like, and performs character recognition from the image data. In such a character recognition procedure, first, it is necessary to read a document image from a document and extract (cut out) an image forming each character from the document image.

Conventionally, as such a character recognition device, for example, as shown in Japanese Patent Application Laid-Open No. 5-143776, general characters having a uniform character width, such as kanji and kana characters, have special characters such as English characters. When characters are mixed, there is a character recognition device that cuts out characters based on the shape position of the circumscribed rectangle of the black pixel connected component.

[0004]

However, in the above-described conventional character recognition device, only the shape of the circumscribed rectangle of the black pixel connected component constituting the character is used, and a general character string or a special character such as an English character is used. For example, if the shape of the circumscribed rectangle is similar between the general character string and the special character string, the character string may be incorrectly determined and character extraction performed with high accuracy. Was difficult.

[0005] In view of the above, it has been desired to realize a character recognition device capable of performing high-speed and high-precision character extraction even when ordinary characters and English characters are mixed.

[0006]

According to the character recognition device of the present invention, in order to solve the above-mentioned problem, a character cut-out unit includes:
Character extraction is performed by using an extraction dictionary composed of a separated character dictionary and a half-width character dictionary. Here, a half-width character dictionary is a dictionary in which characters whose size in the character string direction is described as being smaller than a character having a normal size are registered. The circumscribed rectangular block at the head is a dictionary in which characters similar to the characters registered in the half-width character dictionary are registered among the characters that can be separated into a plurality of circumscribed rectangular blocks of the pixel block. In addition, the character cutout unit includes a number determination unit that determines whether a plurality of circumscribed rectangular blocks are included between the cutout start candidate position and the cutout end candidate position. If the number determined by the number determination unit is plural, the image of the leading circumscribed rectangular block among the plural circumscribed rectangular blocks and the image of the leading circumscribed rectangular block are registered in the separation character dictionary .
Sentences that match characters and are registered in the separate character dictionary
If it is determined that it is not a character, it is checked against a half-width character dictionary,
When it is determined that the character is registered in the half-width character dictionary, the first circumscribed rectangular block is cut out as a half-width character, and a new cut-out candidate start position is set between the next circumscribed rectangular block. If it is determined that the character is registered in the separated character dictionary , or if it is not a character registered in the separated character dictionary ,
When it is determined that the character is not a character registered in the square character dictionary, a cutout determination unit that cuts out the plurality of circumscribed rectangular blocks as separated characters is provided.

[0007]

In the character recognition device according to the present invention, when performing character segmentation, first, the number determination unit includes a plurality of candidate circumscribed rectangular blocks between the segmentation start candidate position and the segmentation end candidate position. Is determined. Next, if the number determined by the number determination unit is plural, the cutout determination unit compares the image of the first circumscribed rectangular block with the character registered in the separated character dictionary.
Is not a character registered in the separate character dictionary
If it is, it is compared with the half-width character dictionary, and if it is determined that the character is registered in the half-width character dictionary, the first circumscribed rectangular block is cut out as half-width character, and a new cut-out candidate is inserted between the next circumscribed rectangular block. Set the start position. On the other hand, if it is determined that the character is registered in the
If, or, it is a character registered in the separator dictionary
If it is determined that the character is not a character registered in the half-width character dictionary , a plurality of circumscribed rectangular blocks are cut out as separated characters.

[0008]

Embodiments of the present invention will be described below in detail with reference to the drawings. << Embodiment 1 >> FIG. 1 is a configuration diagram showing Embodiment 1 of the character recognition device of the present invention. The apparatus shown in the figure is composed of an image input unit 1, a character area extraction unit 2, a line extraction unit 3, a circumscribed rectangle extraction unit 4, a character extraction unit 5, a character recognition unit 6, a result output unit 7, a cut-out dictionary 8, and a recognition dictionary 9. Become.

The image input section 1 is an apparatus which comprises an image scanner or the like and optically reads character information on a document. Further, the character area extraction unit 2 extracts a character area from the image read by the image input unit 1. The line cutout unit 3 has a function of extracting the character area extracted by the character area extraction unit 2 in units of lines, and the circumscribed rectangle extraction unit 4
A circumscribed rectangular block of a character pixel block (black pixel connected component) is extracted from the line extracted by the line extraction unit 3.

The character cutout unit 5 determines an area constituting one character of each character based on the circumscribed rectangular block extracted by the circumscribed rectangle extraction unit 4.
1 and a cutout determination unit 52. Number determination unit 51
Has a function of determining whether or not a plurality of candidate circumscribed rectangular blocks are included between the extraction start candidate position and the extraction end candidate position. In addition, the cutout determination unit 5
When the number of candidate circumscribing rectangular blocks determined by the number determining unit 51 is plural, images of these candidate circumscribing rectangular blocks and characters registered in a separated character dictionary 81 of the cut-out dictionary 8 which will be described later. And if it is determined that the character is registered in the separated character dictionary 81, cut out as a separated character, and if it is not determined that the character is registered, The image of the rectangular block is compared with the characters registered in the half-width character dictionary 82 described later in the cut-out dictionary 8, and
If it is determined that the character is registered here, it has a function of cutting out as a half-width character.

The character recognition unit 6 has a function of performing character recognition on the image of the character area determined by the character cutout unit 5, and the result output unit 7 outputs the result of the character recognition by the character recognition unit 6. It has a function to convert to character code.

The cut-out dictionary 8 is a dictionary for performing character cut-out in the character cut-out unit 5, and includes a separated character dictionary 81.
And a half-width character dictionary 82. Separate character dictionary 8
1 is a character whose circumscribed rectangular block of a character pixel block can be separated into a plurality of characters in a character string direction (horizontal direction in a horizontal writing sentence). This is a dictionary in which similar characters are registered. The half-width character dictionary 82 is a dictionary in which characters whose size in the character string direction is described as being smaller than that of a character having a normal size are registered.

Next, the overall operation of the character recognition device having the above configuration will be described. From the image data input from the image input unit 1, the character region extracting unit 2 extracts only information on the character region by a generally well-known method, for example, a region dividing method using a density distribution of black pixels. The extracted character area information is subjected to line extraction by a line extraction unit 3 which is also a known method, for example, a method of creating a histogram in the line direction and extracting line information. The circumscribed rectangle extracting unit 4 creates a histogram in the vertical direction with respect to the row direction from the line cutout information, detects a black pixel block in the horizontal direction, and furthermore, detects a black pixel block in the horizontal direction in the vertical direction. The upper and lower ends of the black pixels are obtained, and a circumscribed rectangular block is detected. The circumscribed rectangular block is a circumscribed rectangular block of a single character or, in the case of a separated character, a circumscribed rectangular block constituting a part of a character such as a partial portion or a side portion.

The character cutout unit 5 uses a cutout dictionary 8 to perform a well-known pattern matching on a black pixel block in the circumscribed rectangular block or a black pixel block obtained by integrating a plurality of circumscribed rectangular blocks. Using a method, identify whether it is a separation character or a half-width character such as English characters,
Cut out characters one by one. Further, the character recognition unit 6 uses the recognition dictionary 9 to recognize the characters from the cut-out characters using a pattern matching method, which is also well known, and passes the recognition result to the result output unit 7. Result output unit 7
Converts the recognition result into a character code that can be handled by a computer and outputs it to a display device or the like.

Next, the operation of the character segmenting section 5 which is a feature of the present embodiment will be described in detail. FIG. 2 is a flowchart showing the operation of the circumscribed rectangle extraction unit 4 and the character cutout unit 5. FIGS. 3A and 3B are explanatory diagrams of each state. FIG. 3A shows an image cut out into one line and position information, and is a result of line cutout performed by the line cutout unit 3.
(B) shows a histogram of black pixels used in the circumscribed rectangle extracting unit 4, and (c) shows a circumscribed rectangle obtained by the above. (D) shows the description of the cutout operation in the character cutout unit 5.

First, the circumscribed rectangle extracting unit 4 creates a histogram of black pixels in the direction perpendicular to the row direction as shown in FIG. 3B from the image for one row (step S).
1). Next, the left end and the right end of the horizontal black run (the length of the continuous black pixels) are obtained from the histogram, and the upper end and the lower end of one line of the image corresponding to each black run are obtained. A circumscribed rectangular block (hereinafter, simply referred to as a circumscribed rectangle) of a black pixel block as in the example of (1) is extracted (step S2). Next, the character cutout unit 5 detects a character pitch using a generally well-known method, for example, a distance between centers of circumscribed rectangles (step S3). Then, the extraction start candidate position and the extraction end candidate position of one character in the horizontal direction are determined using the detected pitch, and a candidate circumscribed rectangle included in the candidate is detected (step S4).

This will be described with reference to the example of FIG. 3 (d). If the extraction start candidate position s is set to the left of the circumscribed rectangle b6, the extraction end candidate position e is determined using the character pitch. I do. Then, candidate circumscribed rectangles included between s and e are determined as b6 and b7.

Further, the number judging section 51 in the character segmenting section 5 judges whether or not two or more (plural) candidate circumscribed rectangles are included between the segmentation start candidate position and the segmentation end candidate position. If two or more rectangles are included, the process proceeds to step S6. If only one candidate circumscribed rectangle is included, the process proceeds to step S10. In the example of FIG. 3D, the circumscribed rectangle candidates are two of b6 and b7.
Therefore, the process moves to step S6.

As described above, among the characters that can be separated into two or more rectangles in the horizontal direction, the leftmost rectangular block is a half-width character dictionary 82 in the separation character dictionary 81 included in the cut-out dictionary 8. A character similar to the character registered in is registered. For example, the left part of "i" is "L"
And [(], and the left part of "ru" is similar to "/" and ")". Further, the portion to the left of the "mark" is similar to "E". In this way, by limiting and registering the leftmost rectangular block only to the separation characters similar to the half-width characters, the capacity of the dictionary is small, and a high-speed and high-precision separation character dictionary for matching is configured. .

In step S6, all the candidate circumscribed rectangles included from the extraction start candidate position to the extraction end candidate position are integrated, and the separated character dictionary 81 is used to separate the separated character using a well-known pattern matching method. It is determined whether or not the character is a separating character included in the dictionary 81. If the character is a separating character, the process proceeds to step S9. If the character is not a separating character, the process proceeds to step S7. In the example of FIG.
The rectangle “0” of b6 and the rectangle “2” of b7 are integrated,
The pattern "02" is created and matched with the separated character dictionary 81. If the pattern "02" is not registered in the separated character dictionary 81, the process proceeds to step S.
Move the process to 7.

The half-width character dictionary 82 included in the cut-out dictionary 8 includes, as half-width characters, characters that are described with a character width smaller than that of a normal-size character, such as English characters, numerals, and symbols such as parentheses. In step S7, the circumscribed rectangle closest to the cutout start candidate position is included in the half-width character dictionary 82 using this half-width character dictionary 82 and also by using a well-known pattern matching method. It is determined whether the character is a half-width character. If it is a half-width character, the process proceeds to step S8. If it is not a half-width character, the process proceeds to step S9. In the example of FIG.
If it is determined in step S6 that the character is not a separated character, matching is performed with the half-width character dictionary 82 for the circumscribed rectangle b6 closest to the extraction start candidate position s, and the pattern “0” is registered in the half-width character dictionary 82. If step S
Move the processing to 8.

In step S8, characters are cut out with the closest circumscribed rectangle from the cutout start candidate position as one character.
In the example of FIG. 3D, “0” of the circumscribed rectangle b6 is cut out as one character.

In step S9, all the circumscribed rectangles included from the extraction start candidate position to the extraction end candidate position are integrated, and a character is extracted as one character. Step S1
In the case of 0, a character is cut out using a circumscribed rectangle included from the cutout start candidate position to the cutout end candidate position as one character.

In step S11, it is determined whether or not the end of the line at which the extraction of all circumscribed rectangles ends, and if not, the new circumscribed rectangle is inserted between the immediately preceding circumscribed rectangle and the next circumscribed rectangle. Set the extraction start candidate position,
The process moves to step S4. On the other hand, if it is at the end of the line, the character extraction ends. In the example of FIG. 3D, a new clipping candidate position is set between b6 and b7.

As described above, according to the first embodiment, as the cut-out dictionary 8, the separated character dictionary 81 and the half-width character dictionary 82
By using the shape of the black pixel in the circumscribed rectangle instead of cutting out the character only with the conventional circumscribed rectangle of the black pixel, the difference between the bias of the character and the half-width character Can be performed in consideration of the character, and the performance of extracting characters can be improved. In addition, the characters registered in the cut-out dictionary 8 can be limited to English characters, numbers, and symbols, and a very small number of kana and kanji characters that are separated into right and left.
High-speed matching can be performed, and high-speed and high-performance character segmentation can be performed.

Next, a description will be given of a character recognition apparatus according to a second embodiment in which a half-width reject dictionary in which a part of the kanji is registered is added to the first embodiment. << Embodiment 2 >> In Embodiment 2, the cut-out dictionary 8 shown in FIG. 1 of Embodiment 1 is used together with the separation character dictionary 81 and the half-width character dictionary 82 to detect a character other than half-width characters. The half-width reject dictionary 83 (see FIG.
, And the other configuration on the drawing is the same as that of the first embodiment, so that the illustration is omitted.

The half-width reject dictionary 83 registers the following characters. That is, in the half-width character dictionary 82 included in the cut-out dictionary 8, half-width characters to be cut out as half-width characters with symbols such as English characters, numbers, and parentheses are registered.
A Japanese character such as "Akira" may be erroneously detected as a numeral "8" or the like, which may cause a kanji to be separated and cut out. Here, all the kanji of Japanese characters are separated character dictionary 8
1 can be registered, but the separation character dictionary 81
This is not a very effective method because the capacity of the data becomes large and the matching speed decreases. Therefore, by limiting the characters to the characters similar to the characters registered in the half-width character dictionary 82, such as the date deviation, and registering them as half-width reject characters in the half-width reject dictionary 83, the increase in the dictionary capacity is small and the speed is increased. High-precision cutting can be performed.

When the number determined by the number determination unit 51 is plural, the extraction determination unit 52 of the character extraction unit 5 compares the number with the character registered in the separation character dictionary 81, and If it is determined that the character is registered in the dictionary 81, it is cut out as a separated character, and if it is not determined that the character is registered, it is registered in the half-width reject dictionary 83 and half-width character dictionary 82. Matches the character that is If it is determined that the character is registered in the half-width reject dictionary 83, the character is cut out as a separation character, and the half-width character dictionary 8
If it is determined that the character is a character registered in No. 2, it is configured to cut out as a half-width character.

Next, the operation of the second embodiment will be described.
FIG. 4 is a flowchart illustrating the operation of the second embodiment.
The difference from the operation of the first embodiment is that the determination of half-width characters is performed using the half-width character dictionary 82 and the half-width reject dictionary 83, and the other operations are the same.
Only the parts that differ from the operation are described. That is, FIG.
Are the same as those in the first embodiment. In step S7, in the first embodiment, it is determined only whether or not the character is a half-width character included in the half-width character dictionary 82. In the second embodiment, however, the half-width character is further determined using the half-width reject dictionary 83. It is determined whether the character is a half-width reject character or a character not registered in either of them. If it is determined in step S7 that the character is a half-width character, step S7 is executed.
If the character is a half-width reject character or a character not registered in either of them, the process proceeds to step S9.

As described above, according to the second embodiment, by adding the half-width reject dictionary 83 to the cut-out dictionary 8 and registering the component elements of the partial characters similar to the half-width characters, No need to increase capacity,
It is possible to improve half-width character determination performance and perform high-speed and high-performance character segmentation.

In each of the embodiments described above, character extraction is described in the case of a horizontal writing sentence. However, the present invention is not limited to this, and the same can be applied to a case of a vertical writing sentence. In this case, the half-width character dictionary 82 registers characters in which the size of characters in the vertical direction is described as being smaller than that of a normal size. Among the characters that can be separated into a plurality of rectangular blocks in the vertical direction, the uppermost circumscribed rectangular block is one in which characters similar to the characters registered in the half-width character dictionary 82 are registered. Also, the half-width reject dictionary 83
Is a character in which the circumscribed rectangular block of the character pixel block can be separated into a plurality of pieces in the vertical direction, and the character component included in the uppermost circumscribed rectangular block is similar to the character registered in the half-width character dictionary 82. It becomes a dictionary that registers what you have.

[0032]

As described above, according to the character recognition device of the present invention, a half-width character in which a character whose size in the character string direction is smaller than a normal size character is registered. In the dictionary and the character string, circumscribed rectangular blocks of character pixels can be separated into a plurality of characters. Among the characters, the first circumscribed rectangular block registers characters similar to the characters registered in the half-width character dictionary. If a plurality of candidate circumscribed rectangular blocks are included between the extraction start candidate position and the extraction end candidate position, the image of the first circumscribed rectangular block is registered in the separated character dictionary.
Matches with the character that is
If it is determined that it is not a character, match it with a half-width character dictionary
And, if it is determined that the character registered in byte character dictionary, the cut the head of the circumscribed rectangular block as a single-byte characters, and sets a new cut-out candidate start position between the next circumscribed rectangular blocks, whereas, If it is determined that the character registered in the separated character dictionary, or registered to a separation character dictionary
It is not a recorded character and it is registered in a half-width character dictionary
If it is determined that the character is not a character that has been written , multiple circumscribed rectangular blocks are cut out as separate characters, so high-speed and high-precision character cut-out can be performed even for sentences in which general characters and English characters are mixed. It can be carried out. Also,
Characters registered in the separation character dictionary and half-width character dictionary can be limited to English characters, numbers, and symbols, and a very small number of kana and kanji characters that can be separated to the left and right. Character can be cut out.

[Brief description of the drawings]

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

FIG. 2 is a flowchart illustrating an operation of the character recognition device according to the first exemplary embodiment of the present invention.

FIG. 3 is an explanatory diagram of the operation of each unit of the character recognition device of the present invention.

FIG. 4 is a flowchart illustrating an operation of the character recognition device according to the second exemplary embodiment of the present invention.

[Explanation of symbols]

 2 Character area extraction unit 3 Line extraction unit 4 Bounding rectangle extraction unit 5 Character extraction unit 51 Number determination unit 52 Extraction determination unit 6 Character recognition unit 8 Extraction dictionary 81 Separated character dictionary 82 Half-width character dictionary 83 Half-width reject dictionary

────────────────────────────────────────────────── ─── Continuation of the front page (56) References JP-A-3-291778 (JP, A) JP-A-2-230484 (JP, A) (58) Fields investigated (Int. Cl. ⁷ , DB name) G06K 9/00-9/82

Claims (2)

(57) [Claims] 1. A circumscribed rectangle extracting unit for extracting a circumscribed rectangle of each character pixel block in a character area in an input image, wherein a character size in a character string direction is described as being smaller than a character having a normal size. And a half-width character dictionary in which a circumscribed rectangular block of a character pixel block can be separated into a plurality of characters in the character string direction, and a first circumscribed rectangular block is a character registered in the half-width character dictionary. A separated character dictionary in which similar characters are registered, and a number determining unit that determines whether or not a plurality of the circumscribed rectangular blocks are included between a character extraction start candidate position and a character extraction end candidate position; If the number determined by the number determination unit is plural, the image of the first circumscribed rectangular block among the plurality of circumscribed rectangular blocks and the character registered in the separated character dictionary
Sentence registered in the separated character dictionary
If it is determined that it is not a character, it is checked against the half-width character dictionary
And, if it is determined that the character that is registered with the half-width character dictionary, along with cutting the top of the circumscribed rectangular block as a single-byte characters, and sets a new cut-out candidate start position between the next circumscribed rectangular block, On the other hand, if it is determined that the character is registered in the separated character dictionary, or if it is not a character registered in the separated character dictionary and
If it is determined that the character is not a character registered in the half-width character dictionary, a cutout determination unit that cuts out the plurality of circumscribed rectangular blocks as separated characters, and a character that performs character recognition based on the determination result of the cutout determination unit A character recognition device comprising a recognition unit. 2. The character recognition device according to claim 1, wherein a character component included in a leading circumscribed rectangular block among characters in which a circumscribed rectangular block of a character pixel block can be separated into a plurality in the character string direction. A half-width reject dictionary in which the character components similar to the characters registered in the half-width character dictionary are registered, and if the number determined by the number determination unit is plural, the half-width reject dictionary , The image of the first circumscribed rectangular block and the characters registered in the separated character dictionary
Sentence registered in the separated character dictionary
If there was decipher the non-character, matches it to a character component character stored in said byte character dictionary said registered in byte reject dictionary, image of the head of the circumscribed rectangle blocks, wherein or determines that the character elements registered in byte reject dictionary, or, prior to SL if it is determined not to be registered in the half-width character dictionary and the half-width reject dictionary, a plurality of circumscribed rectangles block separator A character recognition device comprising: a cut-out determination unit that cuts out a character.