JP2753094B2 - Word segmentation device - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a word extraction device for extracting words from image information such as a European document.

(Prior Art) Conventionally, as a technique in such a field, Japanese Patent Application Laid-Open No. 62-133
There was one described in JP-A-585.

In general, when a device that processes a document image handles European languages or the like, it is often necessary to perform not only processing in units of characters but also processing in units of words. For example, in an optical character reading device (OCR), by performing a word recognition process after performing a character recognition process, even if a recognition result in a character unit is incorrect, the wrong character is corrected by the word recognition process. Becomes possible. At this time, it is necessary to cut out character lines from the image information of the target document, cut out characters from each character line, and also cut out words.

As a conventional word extracting device of this type, there is one described in the above-mentioned document. In this word cutout apparatus, based on the marginal distribution of the image information of the character string along the character string direction, a histogram (histogram) of the width of the break of the peripheral distribution is used.
ogram; a single random variable occurrence frequency distribution diagram) is created. Then, a threshold for determining the gap between words is determined according to the maximum width corresponding to the peak of the histogram, and a break in the peripheral distribution having a width equal to or greater than the determination threshold is determined as a gap between words, and words are cut out. .

(Problems to be Solved by the Invention) However, the apparatus having the above configuration has the following problems.

(A) In the conventional device, when the number of words included in the character string is small, it is difficult to obtain a peak corresponding to the inter-word gap in the created histogram, so that it is difficult to determine the determination threshold of the inter-word gap, as a result,
Words cannot be cut out correctly.

(B) In the conventional apparatus, since it is necessary to create a histogram in order to determine the threshold value for determining the gap between words, there has been a problem that the apparatus configuration is complicated and the processing speed is reduced.

The present invention is to provide a word segmentation device which solves the problems of the conventional technology that the word segmentation cannot be performed accurately and that the device configuration becomes complicated and the processing speed decreases. is there.

(Means for Solving the Problems) In order to solve the problems, claim 1 of the present invention.
In the invention according to the first aspect, in the word extraction device, preprocessing means for inputting image information of a document such as a European language and extracting a character string pattern from the image information of the document, and a peripheral distribution of the character string pattern in the character string direction Pitch estimating means for detecting an estimated character pitch based on the estimated character pitch, a determination threshold is set in accordance with the estimated character pitch, a blank width larger than the determined threshold is detected, and the word of the document is determined by the position of the blank width. Cutting means for cutting out.

In the invention according to claim 2, in the word segmenting apparatus according to claim 1, the pitch estimating means obtains position information of a partial graphic constituting a character string based on a peripheral distribution in the character string pattern, and And the minimum width of the blank area adjacent to the partial graphic having the maximum width, the estimated character pitch is detected.

(Operation) According to the present invention, since the word segmentation device is configured as described above, the preprocessing unit extracts a character string pattern from image information of a document such as a European language, and supplies it to the pitch estimation unit. The pitch estimating means detects an estimated character pitch based on the peripheral distribution of the character string pattern in the character string direction, and supplies the estimated character pitch to the cutout means. The cutout means sets a determination threshold based on the estimated character pitch, detects a blank width larger than that based on the determined threshold, and cuts out a word based on the blank width position. Thus, regardless of the number of words included in the character string, words can be cut out with high accuracy, and the creation of a histogram as in the conventional case is not required, thereby simplifying the apparatus configuration and increasing the processing speed. I can do it. Therefore, the above problem can be solved.

(Embodiment) FIG. 1 is a block diagram showing a configuration of a word segmentation apparatus according to an embodiment of the present invention.

This word segmentation apparatus includes a preprocessing unit 100 for obtaining character string pattern data S103, and character string pattern data S103.
Pitch estimating means 11 for detecting an estimated character pitch based on
0, a determination threshold value TH according to the estimated character pitch, a cutout unit 120 that detects a blank width larger than the determination threshold value TH in the character string pattern data S103, and cuts out a word based on the blank position.

The preprocessing means 100 has an image reading unit 101 that reads black and white binary image information S101 from a recording medium such as a European print document. The image reading unit 101 is composed of an image scanner or the like, and on the output side, an image storage unit 102 such as an image buffer for storing image information S101, and a character string extracting unit 1
03 is connected. The character string extraction unit 103 has a function of extracting character string pattern data S103 from the image information S101 stored in the image storage unit 102 and providing the extracted character string pattern data S103 to the pitch estimation unit 110.

The pitch estimating unit 110 includes: a block extracting unit 111 that detects position information S111 of a partial graphic (which is called a black block) forming a character string based on a peripheral distribution of the character string pattern data S103 along the character string direction; Black block position information S
Pitch calculator 1 that detects the estimated character pitch based on 111
12, and the cut-out means 120 is connected to the output side.

The extracting unit 120 includes a threshold calculating unit 121 that calculates a determination threshold TH based on the estimated character pitch, and a word extracting unit 122. The word extraction unit 122 detects that the blank width between adjacent black blocks calculated based on the black block position information S111 is larger than the determination threshold TH, and indicates that a word is cut out based on the blank position. It has a function of outputting a cutout signal S122.

FIG. 2 is a diagram for explaining an example of the word extraction operation of FIG. 1, and the operation of the apparatus of FIG. 1 will be described with reference to FIG.

In FIG. 2, 111a is a peripheral distribution along the character string direction in the character string pattern data S103, B1 to B22 are black blocks, W1 to W4 are blank areas, and H is a black block B8 and a blank area W1.
Is the combined width.

First, record information such as a European print document is stored in the image reading unit 101.
And stored in the image storage unit 102 as monochrome binary image information S101. Next, the character string extracting unit 103 extracts character string pattern data S103 (for example, word seg) corresponding to one line from the binary image information stored in the image storage unit 102.
mentation method) and extract it from the block extraction unit 11
Give to one.

The block extraction unit 111 creates the peripheral distribution 111a along the character string direction of the character string pattern data S103 by calculation, and further extracts each of the continuous areas where the peripheral distribution 111a has a positive value as black blocks B1 to B22. I do. Then, the position information S111 of each of the black blocks B1 to B22 including the start position and the end position of each of the black blocks B1 to B22 along the character string direction is calculated, and the calculation result is stored.

Next, the pitch calculator 112 calculates the position information S111 of the black block.
The sum of the maximum width of the black block and the minimum width of the blank area adjacent to the black block having the maximum width is calculated based on the (starting position and end position), and the sum is output to the threshold calculation unit 121 as the estimated character pitch. . That is, the start position and the end position along the character string direction with respect to the i-th black block are S _i and E _i (where S _i <E _i <S _{i + 1} , i = 1, 2, 3,...). , N,
(N is the number of black blocks), the estimated character pitch is expressed by the following equation.

Here, z represents i when the value (E _i −S _i ) takes the maximum value, and min (A, B) represents the minimum value of A and B. The estimated character pitch obtained in this manner has a value approximately equal to the character pitch in each word.

That is, in the pitch calculating unit 112, as shown in FIG.
A black block B8 having a maximum width is detected from the black blocks B1 to B22, and a blank area W1 adjacent to the black block B8 is detected.
And a blank area W1 having the minimum width among the blank areas W2 and W2, and calculates a width H including the black block B8 and the blank area W1 and outputs the calculated width H to the threshold value calculating unit 120 as an estimated character pitch.

Based on the estimated character pitch (= H), threshold calculating section 121 calculates a determination threshold TH regarding a blank width for cutting out a word according to the following formula, and outputs the result to word extracting section 122.

TH = a × where a is a constant, and in this embodiment, for example, a = 0.25
And

The word extracting unit 122 determines the width of the blank area (for example, W3 and W4) between adjacent black blocks based on the position information S111 of each black block stored in the block extracting unit 111.
If it exceeds TH, a word cutout signal S122 indicating that a word is cut out is output based on the blank position. That is, when the value (S _{i + 1} −E _i −1) is larger than the value TH,
The word cutout signal S122 is output between the (i + 1) -th black block and the (i + 1) -th black block as a word break (i = 1, 2,..., N−1). Therefore, in this embodiment, each time a blank width exceeding 0.25 times the estimated character pitch is detected, the cut-out position (start position and end position) of one word is determined.

For example, in FIG. 2, black blocks B1 to B4 ("wor
d "), the black blocks B5 to B16 (" segmentation "), and the black blocks B17 to B22 (" method ") form a word cutout signal S122 indicating that each word is formed. Such a word cutout signal S122 is used. When output from the word extraction unit 122, the word extraction processing for the character string pattern data S103 is completed.

As described above, this embodiment has the following advantages.

(A) The estimated character pitch of the character string pattern data S103 is detected, a blank width larger than a determination threshold TH determined according to the estimated character pitch is detected in the character string pattern data S103, and a word pattern is cut out based on the blank position. With the configuration, words can be cut out with high accuracy regardless of the number of words included in the character string.

(B) The pitch estimating means 110 outputs the character string pattern data S10
3, the position information S111 of the black block is obtained based on the peripheral distribution 111a in the character string direction, and the sum (H) of the maximum width of the black block and the minimum width of the blank area adjacent to the black block having the maximum width is obtained. , Because it is configured to detect the estimated character pitch, compared to the conventional word segmentation device that needed to create a histogram of the width of the break of the marginal distribution 111a
The apparatus configuration is simplified, and the processing speed can be further increased.

Note that the present invention is not limited to the illustrated embodiment, and various modifications are possible. For example, there are the following modifications.

(I) In the above embodiment, the pitch estimation means 110
Based on the peripheral distribution 111a of the character string pattern data S103 in the character string direction, the position information S111 of the black block is obtained, and the sum of the maximum width of the black block and the minimum width of a blank area adjacent to the black block having the maximum width is obtained. According to (H), the configuration is such that the estimated character pitch is detected, but this may be another configuration. For example, based on the marginal distribution 111a in the character string direction of the character string pattern data S103, the position information of the black block is obtained, and the estimated pitch is detected by the sum of the maximum width of the black block and the minimum width of the blank area. The estimated character pitch can be detected in various ways.

(Ii) Each block in FIG. 1 may be constituted by software processing using a microprocessor or the like in addition to being constituted by individual circuits.

(Effects of the Invention) As described in detail above, according to the first and second aspects of the invention, the estimated character pitch is detected by the pitch estimating means based on the character string pattern extracted by the preprocessing means, and then cut out. By the means, the determination threshold is set according to the estimated character pitch, a blank width larger than the determination threshold is detected, and the word pattern is cut out based on the blank position. Therefore, regardless of the number of words included in the character string, And can extract words with high accuracy.

In addition, since the estimated character pitch is detected based on the marginal distribution of the character string pattern in the character string direction, it is not necessary to create a histogram as in the related art, which simplifies the apparatus configuration and increases the processing speed. The effect of conversion can also be expected.

[Brief description of the drawings]

FIG. 1 is a block diagram showing the configuration of a word extracting apparatus according to an embodiment of the present invention, and FIG. 2 is a diagram for explaining an example of the word extracting operation shown in FIG. 100 pre-processing means 101 image reading unit 102 image storage unit 103 character string extracting unit 110 pitch estimating unit 111 block extracting unit 112 pitch calculating unit 12
0 ... cutout means, 121 ... threshold value calculation unit, 122 ... word extraction unit.

──────────────────────────────────────────────────続 き Continuation of the front page (56) References JP-A-62-133585 (JP, A) JP-A-63-158678 (JP, A) JP-A-2-255995 (JP, A) (58) Field (Int.Cl. ⁶ , DB name) G06K 9/00-9/82

Claims (2)

(57) [Claims] 1. A pre-processing means for inputting image information of a document and extracting a character string pattern from the image information of the document, and detecting an estimated character pitch based on a peripheral distribution of the character string pattern in a character string direction. Pitch estimating means, and a cutout means for setting a judgment threshold value according to the estimated character pitch, detecting a blank width larger than the judged threshold value, and cutting out words of the document based on the position of the blank width. A word segmentation device characterized by the following. 2. The word segmenting apparatus according to claim 1, wherein said pitch estimating means obtains position information of a partial figure constituting a character string based on a peripheral distribution in said character string pattern, and A word segmentation device wherein the estimated character pitch is detected by the sum of a large value and a minimum width of a blank area adjacent to a partial graphic having the maximum width.