JP3344069B2 - Character segmentation device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an optical character recognition device (O
(CR). In recent years, there has been an increasing demand for an optical character recognition device (OCR) as an input peripheral device for handwriting.

In order to realize a high recognition rate of individual characters in a character extracting device, a character extracting process which is a pre-recognition stage is important in terms of its accuracy. In a conventional form or the like, a character whose character is correctly written at a designated position can be recognized with high accuracy. However, a character string written in a free-pitch area without a character frame, particularly including a faint character, has a problem in that the recognition rate is reduced because there is no technology for accurately cutting out characters one by one. Therefore, an accurate character segmentation device for characters written in the free pitch area is required.

[0003]

2. Description of the Related Art FIG. 6 is a diagram for explaining a conventional character segmentation device. FIG. 6 (a) shows an example of the configuration, and FIGS.
1) and (c2) show examples of pattern expansion / contraction.
As a conventional technique for the blurring of the characters, when a black pixel connected with a pattern and a white pixel as a background are used, an expansion process for increasing the width by several pixels in the direction of a white pixel from a boundary pixel of the pattern is performed. A contraction process of deleting several pixels in the direction of a black pixel after filling in a blur of a pattern is generally known.

The expansion processing and the contraction processing are described in, for example, "" Image Recognition Theory ", 6.1. Properties of Square Lattice, 6.1.3 Connection of Broken Lines, P69-P71, Makoto Nagao, Corona, 1983 The description is found in the first print on February 15, the first edition. The principle is that, as shown in FIGS. 6 (a) and 6 (b), when image data is input, for example, As described above, when scanning is performed in a 3 × 3 pixel window and the pattern is connected to black pixels and the background is set to white pixels, the expansion processing unit 7 widens a few pixels in the direction from the boundary pixels of the pattern to the white pixels. After the expansion processing fills in the pattern cracks / pattern blurs, the contraction processing unit 8 performs contraction by deleting several pixels in the direction of black pixels, as shown in FIG. 6 (b). This is a process capable of filling white pixels in the pattern, that is, fading.

[0005]

In the above conventional method,
Since the expansion and contraction processing is performed on the entire image by the window of 3 × 3 pixels, it is necessary to access all the pixels to perform the expansion and contraction processing, and there is a problem that the processing time is long. In addition, since the expansion / contraction processing is performed under the same conditions in all directions, blurred characters and characters written separately (hereinafter referred to as separated characters) can be integrated. ｝) On the other hand, characters written close to each other are incorrectly integrated.｝ See Fig. 6 (c2).
There was a problem.

SUMMARY OF THE INVENTION The present invention has been made in view of the above-mentioned conventional drawbacks, and has as its object to provide a character cutout apparatus capable of reducing the processing time and avoiding the problem of erroneously integrating characters written close to each other. It is assumed that.

[0007]

FIG. 1 is a diagram showing the principle of the present invention, FIGS. 2 and 3 are diagrams for explaining the principle of the present invention, and FIGS. 2 (a1) and (a2) show characters. FIG. 2 shows an example of blurring.
(b1) and (b2) show examples of separated characters, FIG. 3 (a) shows an example of extraction of a specific pattern, and FIGS. 3 (b1) and (b2) show examples of outline calculation. The above problem is solved by the character segmenting device configured as follows.

(1) A character segmenting device for extracting a connected pattern by a labeling process
(Connected pattern extraction unit) 2 and pattern feature calculation / extraction means 3, 4 for extracting only patterns having predetermined features
And a contour line calculating means 5 for calculating a contour line determined by weighted expansion for changing the amount of expansion depending on each part of the extracted pattern, and integrating the two when the calculated contour line is in contact with other neighboring patterns. And an integrated determination means 6 that performs the determination.

(2) A character segmenting apparatus which extracts only a pattern having a predetermined characteristic, and calculates a contour determined by weighted dilation which changes the dilation amount depending on each part of the extracted pattern. Means for extracting a rectangle circumscribing the pattern obtained in the labeling unit (connection pattern extraction unit) 2 for extracting a connection pattern by processing, and calculating an average size of each pattern in a character string based on the extracted rectangle. An average size calculating means 3a for calculating; a pattern extracting means 4a for extracting only a pattern having a size smaller than the calculated average size of the character; and a predetermined expansion width for each part of the extracted pattern. It is configured to include a contour calculating unit 5 that calculates a contour determined by the following.

(3) A character segmenting apparatus which extracts only a pattern having a predetermined characteristic, and calculates a contour determined by weighted dilation that changes the dilation amount depending on each part of the extracted pattern. The patterning unit 3b that calculates the directionality of each pattern obtained by the labeling unit (connection pattern extraction unit) 2 that extracts the connection pattern by processing, and extracts only the pattern that has a specific directionality It is configured to include a pattern extraction unit 4b and a contour calculation unit 5 that calculates a contour determined by a predetermined expansion width for each part of the extracted pattern.

(4) The character extracting apparatus extracts only a pattern having a predetermined characteristic, and calculates a contour determined by weighted expansion in which an expansion amount is changed according to each part of the extracted pattern. It is configured to include a contour calculating unit 5 that calculates a contour by connecting points determined by the weighted expansion amounts of the respective parts of the extracted pattern.

(5) The character extracting apparatus extracts only a pattern having a predetermined characteristic, and calculates a contour determined by weighted expansion in which the amount of expansion is changed depending on each part of the extracted pattern. The expansion width in the direction equal to the direction of the extracted pattern is increased, the expansion width in the direction orthogonal to the direction of the pattern is reduced, and the contour line is connected by connecting a series of points determined by the expansion amount of each part. It is configured to include a contour line calculation unit 5 for calculation.

(6) The character segmenting apparatus extracts only a pattern having a predetermined characteristic, and calculates a contour determined by weighted expansion in which the amount of expansion is changed depending on each part of the extracted pattern. A contour line that increases the expansion width at the center position of the extracted pattern, reduces the expansion width as the distance from the center position of the pattern increases, and connects the point sequence determined by the expansion amount of each part to calculate the contour line It is configured to include the calculation unit 5.

(7) The character cutout device according to the above (5) or (6), wherein the expansion width in the direction equal to the direction of the extracted pattern is increased, and Either reduce the expansion width in the direction perpendicular to the direction, or increase the expansion width at the center position of the extracted pattern, as the distance from the center position of the pattern,
Whether to reduce the expansion width is determined based on the features of the extracted pattern.

[0015]

In the character extracting apparatus according to the present invention having the above-described structure, the following operation is performed.

First, in the case of "character fading", 2
There are two cases. One is a case where a part of a character is accidentally blurred as shown in FIGS. 2 (a1) and (a2). This is when using a writing instrument such as a ballpoint pen,
When the writing pressure of the writer is low, it occurs during the preprocessing process of the image. In this case, countless cases can be considered.
The other case is when the writer writes characters separately from the beginning. For example, as shown in FIGS. 2 (b1) and (b2), there are R, M, K and the like in English characters.

In the case of the example shown in FIGS. 2 (a1) and 2 (a2), various shapes can be considered as the shape of the blurred pattern. Generally, the blurred pattern is often small in size. . In the case of the example shown in FIGS. 2 (b1) and 2 (b2), by examining a large amount of data, it is possible to understand the tendency of characters written separately and the shape of the separated pattern.
For example, in English characters, the above R, M, K, etc. are apt to be blurred characters, and the separated pattern is a vertically long pattern.

Therefore, the character cutout apparatus of the present invention first extracts only a pattern that seems to be blurred in the character in the pattern feature calculation unit 3 and the specific pattern extraction unit 4 shown in FIG. The contour calculating unit 5 calculates a contour determined by the expansion width of each part of the pattern, and the next integration determining unit 6, when the contour and another pattern intersect, calculates both patterns. It is intended to be integrated.

FIG. 3A shows an example of extracting a specific pattern. In order to extract faint characters, a small-sized pattern is extracted from a character string. As the size, the area of a rectangle circumscribing the pattern is considered. Therefore, first, a simple average of the sizes of all the rectangles is calculated, and only rectangles smaller than the average value are extracted. This process is performed by the average size calculation unit 3a and the small-size pattern extraction unit 4a in FIG.

Considering the case of separated characters, in the case of English characters, characters written separately include R, H, M, K, etc., and the separated left pattern (vertical bar) is It has direction. Therefore, in order to extract the vertical bar, the vertical direction is extracted by using the pattern direction calculator 3 in FIG.
Extract with b. Those having the directionality in the vertical direction have a large value in terms of a rectangular aspect ratio (vertical length / horizontal length). Therefore, a pattern having a vertical aspect ratio of FIG. 1 is extracted as a vertical bar in which the aspect ratio of the rectangle is equal to or greater than a certain threshold value.

Next, referring to FIG. 3 (b1), an example will be described in which an outline is calculated for a pattern having a small circumscribed rectangle described in FIG. 3 (a). First, for the small-sized pattern extracted as described above, the expansion width in the direction equal to the direction of the pattern is increased, and the expansion width is reduced in the direction intersecting at right angles. In this case, there is a feature in the outline calculation according to the present invention in that only the expansion of the extracted pattern is considered instead of expanding the image itself as described in the above-described conventional character cutout apparatus. .

Next, referring to FIG. 3 (b2), an example of calculating a contour line for the extracted vertical bar described in FIG. 3 (a) will be described. First, with respect to the extracted vertical bar, the center position of the height is set as the peak of the expansion width, and the left and right sides of the vertical bar are expanded so that the expansion width decreases as the distance from the center increases. This is because, in the case of a blurred character of alphabetic characters, the blurring occurs most frequently in the vicinity of the height center of the vertical bar, and the blurring in other portions is less frequently.

In the case of the example of the letter "H" in FIG. 3 (b2), as is clear from the comparison with FIG. 6 (c1), it can be seen that the method according to the present invention and the conventional method are correctly integrated. However, in the case of the example of the number “12”, FIG.
As is clear from comparison with (c2), the method according to the present invention does not perform the integration process, and the conventional method shows a case where the integration is performed erroneously, and the effectiveness of the integration process according to the present invention is clear. is there. The outline calculation and integration processing are performed by the outline calculation unit 5 and the integration determination unit 6 in FIG.

[0024]

BRIEF DESCRIPTION OF THE DRAWINGS FIG. FIG. 1 described above is a principle configuration diagram of the present invention.
FIG. 3 is a view for explaining the principle of the present invention, and FIGS. 4 and 5 are flowcharts showing the outline calculation processing of the present invention.

In the present invention, only the pattern that seems to be blurred is extracted by the pattern feature calculation unit 3 and the specific pattern extraction unit 4, and then the contour determined by the expansion of the pattern is extracted by the contour calculation unit 5. In the case where the calculated outline and another pattern intersect with each other, the integration determination means 6 for integrating the two is the means necessary to carry out the present invention. Note that the same reference numerals indicate the same object throughout the drawings.

Hereinafter, the configuration and operation of the character segmenting device of the present invention will be described with reference to FIGS. First, as shown in FIG. 1, the character cutout device of the present invention extracts only a labeling unit (connection pattern extraction unit) 2 for extracting a connection pattern by a labeling process, and a pattern having a predetermined characteristic. Pattern feature calculating / extracting means 3, 4; contour calculating means 5 for calculating a contour determined by weighted dilation which changes the amount of expansion depending on each part of the extracted pattern; and other patterns in which the calculated contour is nearby Contact with
An integrated judgment means 6 for integrating the two is provided.

A labeling section for extracting only a pattern having a predetermined characteristic and calculating a contour determined by the weighted expansion of each part of the extracted pattern, for extracting a connected pattern by the labeling process.
(Connected pattern extraction unit) means for extracting a rectangle circumscribing the pattern obtained in 2, average size calculation means 3a for calculating the average size of each pattern in the character string based on the extracted rectangle, A pattern extracting means 4a for extracting only a pattern having a size smaller than the calculated average size of the character pattern; and a contour for calculating a contour determined by a predetermined expansion width for each part of the extracted pattern. A line calculation unit 5 is provided.

Further, a pattern directivity calculation unit for calculating the directionality of each pattern obtained by the connection pattern extraction unit 2 for extracting a connection pattern by the labeling process.
3b, a pattern extracting unit 4b for extracting only a pattern having a certain specific direction, and a contour calculating unit 5 for calculating a contour determined by a predetermined expansion width for each part of the extracted pattern. ing.

The contour calculation unit 5 for calculating a contour by connecting a sequence of points determined by the weighted expansion amount of each part of the extracted pattern has a dilation in the same direction as the direction of the extracted pattern. The width is increased, the expansion width in the direction orthogonal to the direction of the pattern is reduced, and the outline is calculated by connecting the point sequence determined by the expansion amount of each part, or The expansion width at the center position is increased, and as the distance from the center position of the pattern increases, the expansion width is reduced, and a contour line is calculated by connecting a series of points determined by the expansion amount of each part.

The outline calculation operation which is the core of the character extraction of the present invention in the character extraction device configured as described above will be described with reference to FIGS. FIG.
FIG. 4 is a flowchart of a process of extracting a pattern having a small size of a character pattern constituting a character string and calculating an outline of the pattern. In the configuration example of FIG. Small pattern extraction unit 4
b and the operation of the contour calculation unit 5 are shown.

First, regarding the target character pattern,
Extreme tilt and rotation are corrected, preprocessing such as noise removal is performed, and a binary image after the preprocessing is obtained. In order to distinguish individual patterns forming characters, a labeling unit (connected pattern extraction unit) 2 applies a known labeling process to the input image data, for example, 3 × 3.
A line consisting of pixels is sequentially scanned line by line, and a new label or the same label as that of a pixel that has already been assigned is assigned to a pattern connected by any one of eight connections. A pattern composed of the assigned pixels is extracted. At this time, since the size of each pattern obtained by the labeling becomes a problem later, coordinate extraction (upper left and lower right coordinates) of a circumscribed rectangle of each pattern is also performed at the same time as the labeling process. .

Next, the average size calculation section 3a calculates the area of the rectangle circumscribing the concatenated pattern of the same label given to the input character string, and calculates the average size of each pattern. . ｝ See processing step 100 in FIG. 4｝ The size of the circumscribed rectangle is
For example, a pattern of 1/3 or less is extracted, and the inclination of the extracted pattern is calculated.処理 Processing steps 101 and 102 in FIG.
Next, a contour line is calculated based on the calculated direction of the inclination of the pattern. Specifically, the expansion width in the direction equal to the direction of the inclination of the pattern is increased, the expansion width in the direction orthogonal to the inclination of the pattern is reduced, and points determined by the expansion amount of each part are connected to form the contour. Calculate the line. ｛Figure 3 (b
1), and processing step 103 in FIG. 4｝ It is checked whether or not the contour line calculated as described above contacts another pattern, and if so, the two patterns are integrated. If the calculated contour does not contact another pattern, the two patterns are not integrated. << See processing steps 104 and 105 in FIG. 4 >> FIG. 5 shows a case of contour line calculation in the case of extracting a vertical bar which is written separately for a separated character and determining whether or not to integrate it with another pattern. ing.

As described above, the direction of the connected pattern extracted by the labeling unit 2 is calculated by the pattern direction calculation unit 3b of FIG.
The pattern is extracted by the pattern extraction unit 4b having a vertical direction.

Next, the contour calculation unit 5 checks whether the aspect ratio of the circumscribed rectangle of the extracted pattern is, for example, 2.5 or more. For the extracted vertical bar, the expanded width at the center position of the height is set to the maximum value, and as the distance from the center position decreases, the expanded width becomes smaller. Are expanded in the left and right directions, and points determined by the expansion amounts of the respective portions of the vertical bar are connected to calculate a contour line. << Refer to processing steps 200 to 202 in FIG. 5 >> It is checked whether or not the contour line calculated as described above comes into contact with another pattern, and if so, the two patterns are integrated. If the calculated contour does not contact another pattern, the two patterns are not integrated.処理 Processing steps 203 and 2 in FIG. 3 (b2) and FIG.
04 As described above, the character cutout device according to the present invention includes a unit for extracting only a pattern having a predetermined characteristic and a unit for calculating a contour line determined by the weighted expansion of each portion of the extracted pattern. And means for integrating both when the calculated contour line comes into contact with another nearby pattern. Further, as the predetermined feature, a rectangle circumscribing the pattern is extracted based on the connection information, and an average pattern size of each character in the character string is calculated based on the extracted rectangle, and compared with the average size. Small pattern,
Alternatively, the directionality of the pattern is calculated, and only the pattern having a specific directionality (vertical length) is extracted. Also, in order to extract the contour line by weighted expansion of each part of the pattern, when the size of the pattern is smaller than the average size, the expansion width is increased in the direction equal to the direction of the pattern, and the pattern has It is characterized in that the expansion width in the direction orthogonal to the direction is reduced, and when the pattern is vertically long, the center position of the pattern is set as the peak of the expansion width, and the expansion width is reduced as the distance from the center increases.

[0035]

As described above in detail, according to the character segmenting apparatus of the present invention, 1) a character string having a faint character is not expanded over the entire image, but is subjected to a specific pattern only. Is extracted, and the dilation evaluation is performed, the dilated portion to be evaluated is reduced, so that the possibility that characters adjacent to each other are mistakenly integrated is reduced, and the processing can be performed at high speed.

2) By extracting only a pattern that is considered to be small in size as one character and performing dilation evaluation,
The number of patterns to be evaluated for dilation is reduced, and the processing can be speeded up.

3) By extracting only a pattern having a specific direction and performing dilation evaluation, the number of patterns to be evaluated for dilation is reduced, and processing can be speeded up. 4) Rather than considering isotropic dilation for the pattern, weighting each part and performing dilation evaluation can relieve frequently blurred characters, and furthermore, unnecessary dilated parts can be reduced. So there ’s less wrong integration,
Accurate character segmentation is possible.

[Brief description of the drawings]

FIG. 1 is a block diagram of the principle of the present invention.

FIG. 2 is a diagram illustrating the principle of the present invention (part 1).

FIG. 3 is a diagram illustrating the principle of the present invention (part 2).

FIG. 4 is a flowchart showing a contour calculation process according to the present invention (part 1);

FIG. 5 is a flowchart showing a contour calculation process according to the present invention (part 2);

FIG. 6 is a diagram illustrating a conventional character cutout device.

[Explanation of symbols]

1 Image image input unit 2 Labeling unit (Connected pattern extraction unit) 3 Pattern feature extraction unit 4 Specific pattern extraction unit 3a Average size calculation unit 3b Pattern directionality calculation unit 4a Small size pattern extraction unit 4b Vertical directionality Pattern extraction unit 5 Contour calculation unit 6 Integration determination unit 7 Expansion processing unit 8 Shrinkage processing unit 100 to 105,200 to 204 Processing steps

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int.Cl. ⁷ , DB name) G06K 9/00-9/82

Claims (6)

(57) [Claims] 1. A character segmentation unit, a labeling unit for extracting a link pattern by labeling processing, and a pattern feature calculation and extraction means to extract only the pattern having a predetermined characteristic, the pattern feature calculation and extraction Pattern extracted by means
Weighting the amount of expansion for each part of the pattern
Two-dimensionally inflated, the contour line calculation means to calculate a contour determined by the expansion, when the contour the calculated is in contact with the other patterns in the vicinity, other patterns in contact with the expansion previous pattern character extracting device characterized by comprising an integrated determination means to integrate and. 2. A character cutout device, comprising: extracting only a pattern having a predetermined characteristic; and calculating a contour determined by weighted expansion for changing an expansion amount depending on each part of the extracted pattern. means for extracting a rectangle that circumscribes the pattern obtained by the labeling section for extracting a connection pattern by, based on the extracted rectangle, and the average size calculation means to calculate the average size of each pattern in a string the <br/> a pattern extracting means to extract only a pattern of small size compared to the average size of the character pattern issued the calculated provided, and extracted by the pattern feature calculating and extracting means pattern
Weighting the amount of expansion for each part of the pattern
Two-dimensionally inflated, character segmentation apparatus characterized by comprising a contour line calculation means to calculate the contour line defined by said expansion. 3. A character cutout device, which extracts only a pattern having a predetermined characteristic and performs a labeling process when calculating a contour line determined by weighted expansion in which an expansion amount is changed depending on each part of the extracted pattern. a pattern extraction means to extract the directional calculation hand stage of the pattern to calculate the orientation, only the pattern having a specific orientation of having the respective patterns obtained by the labeling unit that extracts a link pattern by the pattern Pattern extracted by feature calculation / extraction means
Weighting the amount of expansion for each part of the pattern
Two-dimensionally inflated, character segmentation apparatus characterized by comprising a contour line calculation means to calculate the contour line defined by said expansion. 4. A character cutout device, which extracts only a pattern having a predetermined characteristic, and calculates a contour determined by weighted expansion that changes an expansion amount according to each part of the extracted pattern. Weight of the expansion amount for each part of the
Only two-dimensionally inflated, the character segmentation apparatus characterized by comprising a contour line calculation means to calculate the contour line connecting the point determined by the expansion. 5. A character segmenting apparatus, wherein a predetermined feature is provided.
Extract only the patterns that have
A ring determined by weighted expansion that changes the amount of expansion depending on the position
When calculating the contour line , increase the expansion amount at the center position of the extracted pattern.
And expands away from the center of the pattern
The volume is reduced and expanded two-dimensionally.
Contour calculation means for calculating contours by connecting round points
A character segmentation device characterized by the following. 6. The character segmenting device according to claim 5,
Te, expansion equal direction as the direction having the pattern of the extracted
In the direction perpendicular to the direction of the pattern.
Reduce the amount of expansion or use the extracted pattern
Increase the amount of expansion at the center position of
As you move away from the
It is specially determined based on the characteristics of the extracted pattern.
Character cutout device to be used.