JPH0644406A - Method and device for processing image - Google Patents

Abstract

PURPOSE:To enhance the precision for segmenting a character, and to decrease an erroneous recognition by tracking vertically a white picture element from a dividable position in image information containing plural characters. CONSTITUTION:First of all, in order to segment a first character, a method for taking a projection is used. As a result, with respect to the whole inputted image, a projection is taken in the line direction and a line rectangle is segmented, and at every segmented line rectangle, a projection is taken in the direction being vertical to the line, and character images 301-304 are segmented. Subsequently, a dividable position of the character image is derived. Thereafter, lines 305-308 are drawn at height of half of a character rectangle, and a middle point of the part in which a white picture element continues on the line becomes a divibable position. Next, a second character segmentation processing is executed. In this case, a divided position is searched vertically from the obtained dividable position and the processing for segmenting a character is finished. That is, the character rectangle 302 is divided into three of 309-311 and the character segmentation processing is finished.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image processing method and apparatus, and more particularly to an image processing method and apparatus for cutting out characters one by one from input image information.

[0002]

2. Description of the Related Art FIG. 13 shows an example of character recognition processing in a conventional optical character recognition device.

First, an original is read using an image scanner or the like (S301), and then a character image for one character is cut out from the input image (S302). Next, the features of the cut-out character image are extracted (S303), and the similarity is calculated using the standard features of each category that are facilitated in advance (S304). As a result, the category with the highest degree of similarity becomes the recognition result and is displayed on the CRT or the like (S3).
05).

There are some known methods for cutting out a character from an input image, but a method for projecting a character image, which is one of the most general methods, will be briefly described.

Taking a horizontal character string as shown in FIG. 14 as an example, by taking a projection 402 on a horizontal projection plane 401, a line rectangle 403 whose height is the length h of the projection is obtained. Further, by projecting the obtained line rectangle in the vertical direction, projections 405 to 410 are obtained on the projection surface 404, and characters are cut out.

[0006]

However, in the above conventional example, as shown in FIG. 15, "ko", "uu" and "de" are used.
If the two overlap, the projection of the image is taken, and as shown by the frame 15, three characters are combined and “kode” is cut out as one character. Therefore, it is necessary to judge a rectangle that needs to be further divided from the once cut-out rectangles based on the ratio of the height to the width, and perform the character cutting process again for the rectangle that needs to be further divided.

However, as shown in FIG. 16, when an object to be cut out is a character string such as an English sentence whose character width greatly differs depending on the character, it cannot be cut out with the average character width.
Further, if "AWA" is cut in a vertical line, one character cannot be obtained correctly, and subsequent recognition cannot be performed correctly.

Further, as shown in FIG. 17, when adjacent characters are in contact with each other, there is a drawback that the two characters cannot be separated even by the method of tracing the boundary of the character image.

[0009]

In order to solve the above problems, the present invention obtains a dividable position in image information containing a plurality of characters, tracks white pixels above and below the dividable position, and traces the white pixels. There is provided an image processing method and device, characterized in that the image information is divided by the dividing line obtained.

In order to solve the above problems, preferably, the image information containing the plurality of characters is one image information frame obtained by performing character cutting processing from the document image information.

In order to solve the above-mentioned problems, it is preferable that the dividable position is on a white pixel continuous area on a line segment set in the image information including the plurality of characters.

In order to solve the above problems, it is preferable that the line segment set in the image information is at a position half the height of the image information.

In order to solve the above-mentioned problems, the present invention derives a division candidate position in image information containing a plurality of characters, recognizes the image information divided at the derived division candidate position for each divided area, and resembles each other. An image processing method and apparatus, characterized in that a degree is calculated and the division position is determined according to the calculated similarity.

In order to solve the above-mentioned problems, it is preferable that the image information containing the plurality of characters is one image information frame obtained by performing character cutting processing from the document image information.

[0015]

【Example】

[Embodiment 1] FIG. 1 is a block diagram showing the configuration of an image processing apparatus according to this embodiment. In FIG. 1, 101 is a ROM 1
A CPU for inputting images, cutting out character images, and controlling the entire apparatus according to a control program stored in 04, a keyboard (KB) 102 for inputting and correcting characters, and a mouse 103. Pointing device (PD), 104 is a ROM for storing a control program for the processing executed by the CPU 101, which will be described later with reference to the flowchart, 105 is a RAM for storing character images, character cutout results, recognition results, etc., and 106 is a cutout. A similarity calculation unit that calculates the similarity between each character image and each character, 107 is an image scanner (SCA) that reads the image.
N), 108 is an interface (I / F) of the imageless scanner, and 109 is a display for displaying a character recognition result and the like.

This embodiment shows a method for obtaining character cut-outs of English sentences having different character widths as shown in FIG. 16 by tracing the dividable points. An outline of the entire character segmentation process of the present embodiment executed by the character recognition device configured as shown in FIG. 1 will be described with reference to the flowchart of FIG.

First, in step S201, a first character is cut out by taking a conventional mapping. It is determined whether the cut-out character image can be further divided (S20).
2) In the case of division, the position is obtained as a dividable position (S203). According to the dividable position obtained in S203, it is determined whether or not the second character cutting process is performed (S
204), the second character is cut out. For more detailed description, each step will be described using the character image of the alphabet of proportional pitch in FIG. FIG. 3A shows the input character image, and character cutting is performed on this image information. First, in S201, a method of taking a projection is used as the first character cutout. In this, a mapping is first performed in the row direction and a line rectangle is cut out for the entire input image, and then a mapping is performed in a direction perpendicular to the line for each cut out line rectangle, and a character image is cut out. As a result, four rectangles 301, 302, 303, 304 are obtained as shown in FIG.

Here, it is determined whether or not each rectangle obtained in S201 is further divided (S202). As a method of determining this, for example, each rectangle (301, 302, 303, 304)
It can be determined from the ratio of the height to the width of the sheet, or can be determined from the width being equal to or larger than (equal to or less than) a reference value. Further, this step S202 may be omitted, and the dividable position may be immediately obtained in step S203.

Next, the dividable position of the character image is obtained (S203). As shown in FIG. 3C, lines 305, 306, 307, and 308 are drawn at the height of a half of the character rectangle, and the midpoint of the part where white pixels are continuous on the line is set as the dividable position. In S204, each character rectangle (301, 302,
(303, 304) whether or not the dividable position obtained in S203 exists, and if there is, a second character cutout process S205 is performed. In the case of this example, since there are dividable positions (indicated by x in FIG. 3C) in the character rectangles 301 and 302, the second character cutout process is performed on the two rectangles. A method for further dividing the rectangle by tracing up and down from the dividable position obtained in S203 will be described with reference to the flowcharts of FIGS. 4 and 5. FIG. 4 is a flowchart of the character segmentation processing (S205) for tracing in the upward direction and FIG. 5 for tracing in the downward direction. The coordinate axes are x-axis in the horizontal direction and y-axis in the vertical direction.

First, let Ps be the obtained dividable position,
The tracking point P for searching for the division position up and down from the dividable position Ps is set at Ps (S401). Next, P is shifted upward by one pixel in order to attempt the upward division (S40
2). If P is not a black pixel, the pixel is gradually moved upward until P reaches the upper part of the character rectangle (end of division of the character rectangle in the upward direction) (S403). When P is on the black pixel during the dividing line tracking (S404), the boundary line of the (black pixel area) of the character image is tracked. A method of tracking the boundary line of the black pixel region in the clockwise direction will be described by taking the case where the tracking moves from "8" to "P" as shown in FIG. For the 8 pixels around the tracking point P, the position of the previous P is "8".
The black pixels are sequentially examined from the right adjacent to, ie, “1”, and the point where the black pixel is present for the first time, that is, “5” is advanced as the next movement destination of the tracking point P. In the case of counterclockwise tracking, the counterclockwise check is performed in S412 to S418. To trace the boundary line, the coordinate of P is first stored in Pm (S4
05). Next, find a black pixel on the next boundary line clockwise and set P
(S406). If P reaches the upper portion of the character rectangle, the upward division is completed (S407). Also,
If the y coordinate of P becomes equal to the y coordinate of Ps, it is determined that the tracking is proceeding downward, and the clockwise tracking is stopped (S
408). When the x-coordinate of P becomes equal to the x-coordinate of Ps as a result of advancing the tracking, the tracking by the boundary line ends (S409). It is a point where P has passed, and the clockwise tracking cannot be continued and is stopped (S410). This is shown in FIG. If the pixel above P is a black pixel, the boundary line is still continuing, and therefore the process returns to S406 again (S411). If not, try heading up again to split.

When the character rectangle cannot be divided by the clockwise tracking, the coordinate of P is returned to Pm (S412), and the division is attempted by the counterclockwise tracking (S413). In this case as well, the process proceeds in the same manner as clockwise, and if the condition for stopping the tracking is met, it is determined that the character rectangle cannot be divided, and the tracking is stopped (S419).

If the upward division is successful, then the downward division is tried. The process is shown in the flowchart of FIG. This is also similar to the upward division, and the tracking point P
If the character reaches the bottom of the character rectangle, the division is successful (S51).
9). Finally, as shown in FIG. 3D, the character rectangle 302 is divided into three parts 309, 310, and 311 and the character cutting process is completed.

Therefore, according to the present embodiment, by using the character segmentation using the projection and the character segmentation by the boundary line tracking, even when the character images overlap each other,
Characters can be cut out, and it is possible to cut out characters even in italicized documents. Further, since the boundary line is traced locally and not in a place other than necessary, there is an effect that the processing can be performed at high speed.

In the above description, a line is drawn at half the height of the character rectangle in order to search for a dividable position in S203.
This position may be changed, or a histogram of black pixels in the horizontal direction may be taken for each rectangle and a line may be drawn at the maximum portion. In that case, there is an effect that the dividable position is reduced and the number of times the second character cutout process is performed is reduced. Further, in the above-described embodiment, the dividable position is the midpoint of the line segment of the white pixel, but it may be the point where the black pixel on the center line changes to the white pixel or the point where the white pixel changes to the black pixel.

[Embodiment 2] This embodiment shows a method of cutting out characters when adjacent characters are in contact as shown in FIG.

The configuration of the image processing apparatus in this embodiment is as follows.
This is similar to the first embodiment and is shown in FIG.

The character cutout process shown in this embodiment is shown in the flowchart of FIG. 8 and will be described in detail.

First, in step S801, the image scanner 107
The image is input from and the character image is stored in the RAM 105. In step S802, character extraction is performed by the method described in the conventional example or another known method, for example, a method such as projection. Then, from the cut-out frame, the standard character width is calculated for horizontal writing and the standard character height is calculated for vertical writing (S80).
3). The process of obtaining the standard character width is shown in the flowchart of FIG. 12 and will be described later in detail.

In step S804, it is determined whether or not one block cut out in step S802 needs to be divided into a plurality of blocks using the standard character width (high) obtained in step S803. In the determination in S804, for example, in the case of horizontal writing, the width of each character block cut out in S802 is S8.
If it exceeds 1.5 times the standard character width obtained in 03, it can be determined that division is necessary. If the division is recognized in S804, division blocks are created (S805).

The process of further division in S805 will be described by taking a horizontally written character image of Japanese as shown in FIG. S804
If the character block selected in step is almost an integral multiple of the standard character width W, it is divided by the standard character width (90
1, 902) and when not divided (903), three character blocks are created (S805). FIG. 10 shows FIG.
This is an example of a complicated case where half-width characters are included in. Considering half-width and double-width characters, half of the standard character width is used as a unit of division, and when the width of the character block exceeds 1.5 times the standard character width, creation of the divided block is started.
In the case of this example, half-width blocks (1001 to 1006),
Full-width blocks (1007 to 1011) and double-width blocks (1012 to 1014) are created.

The character block created as described above is subjected to feature extraction (S805), similarity calculation (S807), and S80.
In S809, it is determined which block is to be adopted for the divided block created in 5.

Regarding the determination method of S809, FIG. 9 and FIG.
This will be described using an example of division of 0. In FIG. 9, block 90
The similarity between 1 and 902 is smaller, and the similarity between 903 is compared. If the similarity between 903 is large, 901 and 902 are not adopted, 903 is taken as the character cutting result, and vice versa.
01 and 902 are adopted and 903 is not used. In this example, it is expected that the degree of similarity of 903 will be small, so that the character cutting frames 901 and 902 can be obtained even for a contact character. Next, in the case of FIG. 10, as in the above example, 1
All combinations are considered using 001 to 1014, and among the divided blocks having the minimum similarity in each combination, the combination having the highest similarity is adopted as the character cutting result. Alternatively, the block to be adopted as the character cutting result is determined from the left side, that is, the block having the maximum similarity among the blocks (1001, 1007, 1012) including 1001 is set as the character cutting result. Go to the next block. In this case, 1007 is expected to be the maximum similarity and that is adopted. Next is 1003, 1009, 10
The same judgment is made at 14. As a result, both of the above two examples are 10
It is expected that 07, 1003, 1010, 1006 will be adopted, and a correct character cutting result can be obtained.

The recognition result obtained as described above is displayed on the display 109 (S810).

According to the present invention, contact characters included in a document whose circumscribed rectangle of a character image is almost square and whose major part is large, such as a Japanese document, are compared only with a standard character size and similarity. Allows accurate recognition, and correct recognition is possible even when characters having different character sizes, such as half-width characters and double-width characters, are mixed in the contact character.

Now, another example of the method of creating divided blocks in S805 will be described.

FIG. 11 shows a character image in which characters come into contact with each other to form one character block. In this case, the width of the character block is only slightly larger than the standard character width, and if the standard character width or half of the standard character width is used as the division position as in the previous embodiment, division is performed so that a correct recognition result cannot be obtained. (1101). Therefore, a pattern in which the division pattern is changed little by little is created before the similarity calculation. After the similarity calculation, the combination with the highest similarity is adopted.

According to the present invention, even if a character different from the standard character size comes into contact, a highly accurate recognition result can be obtained.
Also, the same effect can be obtained for documents such as English whose character width varies depending on the characters.

In the above embodiments, the character cutout processing is performed by the CPU 101, but the character cutout processing unit may be independent, and the recognition calculation may be performed by the CPU 101.

The image is input by the image scanner 10.
However, if the external storage device is configured, the image data obtained by another means may be temporarily stored and taken in from there.

A method of obtaining the standard character width (height) in S803 described above will be described with reference to the flowchart of FIG.

First, the character block cut out in S802 is found to have the maximum height, and its height is hm.
Ax (S120). The character block of interest is set as the first block of the line (S121), and the following processing is performed until all blocks in the line are checked.

The width W of the target block is compared with hmax (S123), and if the width of the target character block is sufficiently narrow or not wide enough than hmax, this block is set as a standard character block (S124), and the standard character width is set. Used to ask for.

If the block width is sufficiently narrower than hmax, it is combined with the next character block (S126) and the width of the combined character block is compared with hmax (S12).
7). If the width of the combined block satisfies the same condition as the judgment condition of S127, the block is set as a standard character block, and if it is still narrow, it is further combined and the same judgment is performed. On the other hand, if it is sufficiently large, it means that the composition is too much and the composition is canceled. When it is determined whether the block is a standard character block, the block of interest is moved to the next (S125). The average width of the standard character blocks thus obtained is calculated, and this is set as the standard character width W (S129).

Also in the case of vertical writing, if the horizontal direction is replaced by the height direction and the vertical direction is replaced by the width direction, the standard character height can be obtained by the same method.

[0045]

As described above, regarding a document in which the character pitch and the character width are not constant, a separable position is searched for even if the first character cutout alone cannot provide sufficient certainty. By performing the second character cutout, there is an effect that the accuracy of the character cutout is improved and erroneous recognition is reduced. As a result, the subsequent correction work is reduced, and the document input time can be shortened.

As described above, some character cutting patterns are prepared for suspicious character images even if there are contact characters etc. that cannot be accurately performed by the conventional character cutting methods, and after the identification calculation. By selecting the most probable pattern as the recognition result, false recognition is reduced,
This has the effect of reducing correction work.

Further, there is an effect that the input operation of the existing document can be easily performed in a short time.

[Brief description of drawings]

FIG. 1 is a block diagram of a character recognition device according to an embodiment.

FIG. 2 is a flowchart of character recognition processing according to the first embodiment.

FIG. 3 is a character image for explaining the first embodiment.

FIG. 4 is a second flowchart of character cutting processing according to the first embodiment.

FIG. 5 is a second flowchart of character cutting processing according to the first embodiment.

FIG. 6 is an exemplary diagram of character cutout.

FIG. 7 is an explanatory diagram of boundary line tracking.

FIG. 8 is a flowchart of character recognition processing according to the second embodiment.

FIG. 9 is a first exemplary diagram of a division pattern.

FIG. 10 is a second exemplary diagram of a division pattern.

FIG. 11 is a third exemplary diagram of a division pattern.

FIG. 12 is a flowchart of processing for obtaining a standard character width (height).

FIG. 13 is a flowchart of conventional character recognition processing.

FIG. 14 is an exemplary diagram of character segmentation by projection.

FIG. 15 is an exemplary diagram of character segmentation.

FIG. 16 is a first exemplary diagram of a character cutting target.

FIG. 17 is a second exemplary diagram of a character cutting target.

Claims (14)

[Claims] 1. The image information including a plurality of characters is obtained with a dividable position, white pixels are traced up and down from the dividable position, and the image information is divided by dividing lines obtained by the tracing. Image processing method. 2. The image processing method according to claim 1, wherein the image information including the plurality of characters is one image information frame obtained by performing character cutting processing from document image information. 3. The image processing method according to claim 1, wherein the dividable position is on a white pixel continuous area on a line segment set in the image information including the plurality of characters. 4. The image processing method according to claim 3, wherein the line segment set in the image information is at a position half the height of the image information. 5. In image information including a plurality of characters, a division candidate position is derived, image information divided at the derived division candidate position is recognized for each divided area, and a similarity is calculated. An image processing method, wherein the division position is determined according to the degree of similarity. 6. The image processing method according to claim 5, wherein the image information including the plurality of characters is one image information frame obtained by performing character cutting processing from document image information. 7. In image information including a plurality of characters, a dividable position deriving means for obtaining a dividable position, a tracing means for tracing white pixels up and down from the dividable position, and a tracing means for tracing the white pixels. An image processing apparatus comprising: a dividing unit that divides the image information by a dividing line. 8. The image processing apparatus according to claim 7, wherein the image information including the plurality of characters is one image information frame obtained by performing character cutting processing from the document image information. 9. The image processing apparatus according to claim 7, wherein the dividable position is on a white pixel continuous area on a line segment set in the image information including the plurality of characters. 10. The image processing apparatus according to claim 9, wherein the line segment set in the image information is at a position half the height of the image information. 11. In image information including a plurality of characters, a division candidate position is derived, image information divided at the derived division candidate position is recognized for each divided region, a similarity is calculated, and the calculated similarity is calculated. An image processing method, characterized in that the division position is determined according to the degree. 12. The image processing method according to claim 11, wherein the image information including the plurality of characters is one image information frame obtained by performing character cutting processing from document image information. 13. In image information including a plurality of characters, division candidate position deriving means for deriving a division candidate position, and image information divided at the division candidate position derived by the division candidate position deriving means is recognized for each divided area. An image processing apparatus comprising: a similarity calculation means for calculating a similarity and a division position determination means for determining the division position according to the similarity calculated by the similarity calculation means. 14. The image processing apparatus according to claim 13, wherein the image information including the plurality of characters is one image information frame obtained by performing character cutting processing from document image information.