JP2844618B2 - Character segmentation device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial Application Field] The present invention relates to a character cutout device which is one of the main components of a character recognition system.

[Prior Art] Conventionally, a connected pixel set is obtained by tracing the connected state of character pixels one pixel at a time from character image data obtained by reading a character string, and a character circumscribing frame circumscribing the pixel set is obtained. By reading the coordinates of, characters were cut out one by one from the character string. For example, "The Story of Image Recognition" (Yuji Kiuchi, published by Nikkan Kogyo Shimbun, 90 pages)
Among them, character extraction is described as a pre-process of character recognition.

[Problems to be Solved by the Invention] In the conventional character segmentation device, it takes a long time to track the pixels, and one character may be divided into many character circumscribed frames. There is a problem that the process becomes complicated when there is a lot of variation. In particular, when character recognition is performed in real time, this is a factor that significantly increases the recognition time, and the emergence of a device that can perform a cutout process in a short time has been desired.

An object of the present invention is to provide a character extracting device capable of executing character extracting from a document having a high degree of freedom at high speed, instead of directly obtaining a character circumscribing frame coordinate in order to solve such a problem. I have.

Means for Solving the Problems In order to solve the above-mentioned problems, the invention according to claim 1 is a character image storing an image of a read character string as character image data including character pixels and background pixels. A character area for forming a right-angled polygon including a set of the character pixels connected to each other from the character image data and creating character-area image data representing a character-area image of the right-angled polygon; Image data creating means; character area image data storage means for storing the character area image data; feature point reading means for extracting feature points at vertices of the character area image and reading feature coordinate values; Feature point storage means for storing coordinate values; character circumscribed frame forming means for obtaining a character circumscribed frame from the feature point coordinate values; And a character circumscribed frame storage means for storing coordinate values of the character circumscribed frame. The character area image data creating means comprises: a coordinate value (X, Y) in the character area image and a pixel value in the character area image. For G (X, Y), there is provided a means for generating the character area image so as to satisfy the following expressions (a) and (b).

For any X where Xs ≦ X ≦ Xe, Ymin (X) = Ys (a) For any X where Xs ≦ X ≦ Xe, G (X, k) ≠ G0,
Ymin (X) ≦ k ≦ Ymax (X) (b) where Xs and Xe are domains in the X coordinate of the character area image, Ys and Ye are domains in the Y coordinate of the character area image, and Ymin (X), Ymax (X) is the minimum and maximum values of the Y coordinate value of the character pixel group at the coordinate X, and G0 is the pixel value of the background pixel.

Here, the right-angled polygon is a polygon whose angle at each vertex is a right angle.

In the character cutout device according to claim 2, the feature point reading means has means for extracting a plurality of types of feature points by performing a mask process on the character area image data, Applying a combination of the feature points to a plurality of basic patterns to obtain a combination of feature points corresponding to any of the plurality of basic patterns, and obtaining coordinates of a character circumscribed frame from the obtained combination of the feature points. Having.

[Operation] In the character segmentation device of the first aspect, the above expressions (a),
Since the character area image is generated so as to satisfy (b), it is possible to generate a right-angle polygonal character area image according to the connection state of the character pixels. After extracting the characteristic points at the vertices of the character area image, the character frame is determined from the characteristic points, so that an appropriate character frame can be determined according to the connection state of the character pixels.

In the character extracting apparatus according to the second aspect, since the character frame is obtained by applying the combination of the feature points to the plurality of basic patterns, an appropriate character frame can be obtained from the combination of the feature points.

Embodiment FIG. 1 is a block diagram showing a configuration of a character cutout device as one embodiment of the present invention.

The character image storage unit 1 stores character image data of a binarized character string in a character portion and a background portion sent from the scanner unit.

The character area image data creating means 2 creates character area image data from the character image data and stores it in the character area image data storage means 3.

The characteristic point reading means 4 obtains characteristic points and their coordinate values from the character area image data and stores them in the characteristic point storage means 5.

The character circumscribed frame forming means 6 obtains a character circumscribed frame from the feature point data and stores it in the character circumscribed frame storage means 7.

With the above configuration, it is possible to create character area image data from character image data, extract feature points characterizing the connectivity of pixels, obtain the coordinate points, and theoretically obtain a character circumscribed frame.

The character extracting device will be described below. First, the features of the character area image data creating means will be described using character image data and an example of character area image data created by the means ("i").

FIG. 3 (a) shows input character image data, and FIG. 3 (b) shows data in the process of being processed (see FIG.
FIG. 3 (c) shows the character area image data at the stage when the character area image processing has been completed.

Each image in the character area image data created by the character area image data creating means 2 has the following properties.

For any x where X _s ≦ x ≦ X _e , for any x where Y _min (x) = Y _s X _s ≦ x ≦ X _e , G (x, k) ≠ 0, Y _min (x) in ≦ k ≦ Y _max (x) _{where, X s, X e, Y} s, the Y _e is a domain in the X-coordinate, Y-coordinate of the character region image as shown in FIG. 3 (c). G (X, Y) is the value of the pixel at the coordinates (X, Y), and takes the values of 0 (background pixel) and 1 (character pixel and a pixel replaced as a character pixel). Y _min (X) and Y _min (X) are the minimum and maximum values of the Y coordinate of the character pixel group at the X coordinate. The expression indicates that no chipping occurs in the upper side of the character area image, and the expression indicates that no hollow area or a horizontal (parallel to the X coordinate) dent occurs in the character area image.

In order to determine the character circumscribed frame from the character area image data, it is sufficient to know the coordinate values of at least three types of pixels, but there is also a method of using four or more types. These pixels are called feature points. FIG. 4 (a) shows a 3 × 3 mask for extracting a feature point, and FIG. 4 (b) shows an example of a condition of pixels surrounding the feature point when extracting each feature point with a 3 × 3 mask. ing. Specifically, the feature point P is a pixel at the upper left corner of the character area image, and the feature point Q is a pixel at the lower left corner. The feature point R is a pixel that is concave to the lower right and has a concave corner. Although other combinations of pixels serving as feature points are conceivable, the following description focuses on feature points obtained under the conditions shown in FIG. 4B.

Since the character region image has the property shown by the equation, there can be only one feature point P for one character region image. On the other hand, a plurality of feature points Q exist when there is a chip on the lower side of the character area image. If the lower left is missing, there is only one. When there are a plurality of feature points Q, a feature point R exists. The characteristic point R always has a characteristic point Q having the same value of the X coordinate as a pair due to the property shown by the equation. Further, since the X coordinate values are the same, it can be understood that there is no paired feature point R at the rightmost feature point Q (X coordinate value is the largest) (if the feature point R exists, There must be a feature point Q with a larger X coordinate value). Therefore, the number of feature points R is equal to the number of feature points Q minus one.

FIG. 5 shows, as an example, three types of characteristic points in the character area image data shown in FIG. 3 (c) (type P has one point, type Q has two points, and type R has one point. ing).

 The characteristics of the feature points are summarized as follows.

(A) Only one type P exists in one character area image.

(A) One or more types Q exist in one character area image.

(C) The number of type R is one less than the number of type Q, and there is always a pair of characteristic points of type Q, and their X coordinate values are the same.

As described below, by utilizing the properties of (a), (b), and (c), it is possible to theoretically obtain the character circumscribed frame coordinates from the feature point coordinate values using the character circumscribed frame forming means 6. Becomes

Therefore, the feature point reading means 4 only needs to locally extract each feature point and its coordinate value, and there is no need to extract any information as to which character rectangular image each feature point is. Conventionally, compared to the case where character-by-character circumscribed frame coordinates are obtained directly from image data,
Here is the rationale for significantly reducing the processing time.

The functions and grounds of the character circumscribing frame forming means 6 will be described below. FIG. 6 shows a basic pattern of a character area image, its characteristic points, and a character circumscribed frame. Since any complicated character area image can be reduced to the pattern shown in FIG. 6, the principle will be described using basic patterns.

When a feature point A of type P and a feature point B of type Q satisfy the following conditions, the two feature points belong to the same character area image.

(1) x-coordinate value of point b <x-coordinate value of point b (2) y-coordinate value of point b <y-coordinate value of point b (3) A rectangle having points a and b as diagonal points was considered. At this time, there are no feature points other than the paired types Q and R in the rectangle.

(4) There is no type P feature point having an x coordinate value larger than the x coordinate value of the point a and satisfying the conditions (1), (2), and (3).

(5) There is no type Q feature point having a y coordinate value smaller than the y coordinate value of the point b and not belonging to any character area image.

In the description of FIG. 6, it is assumed that the conditions (4) and (5) are satisfied. (Condition (5) is automatically satisfied by determining the character area to which the feature point belongs in ascending order of the y-coordinate value of the feature point of type Q. Condition (4) is satisfied with conditions (1) and (2). (It is satisfied by selecting a point having the maximum x-coordinate value among the feature points of type P that satisfy (3).) Also, it is clear that the conditions (1), (2), and (3) are satisfied. . Therefore, it can be seen that the feature points A and B shown in FIGS. 6A, 6B and 6C belong to the same character area image.

In the case of FIG. 6 (a), when a rectangle having the feature points A and B as diagonal points is considered, the other feature points are not included in the rectangle. Becomes
In the case of FIG. 6 (b), when the feature points A and B are character circumscribed frames, it can be seen that the character circumscribed frame is not a complete character circumscribed frame since the feature point D of type R is included inside. In the case of a feature point of type R, there is always a feature point c of type Q which is a pair, and belongs to the same character rectangular image. Therefore, the feature point c is searched, a feature point e is created as a new feature point of type Q from the coordinate values of the feature points b and c, and the feature points d, c, and b are deleted.
Then, it checks which basic pattern the combination of the feature points A and E matches. In the case of FIG. 6C, the feature point B itself is paired with the type R feature point d. Type R
It is clear from the definition of the characteristic point that the character area image also exists on the right side of the characteristic point d, and as a result, the characteristic point e must exist. Then, the feature points D and E are deleted, and the feature point E is searched to find out which basic pattern the combination of the feature points A and E matches.

By repeating the processing described above, a complete character area can be obtained. Therefore, the circumscribed frame for all characters is
Character circumscribed frame coordinates can be logically obtained from at least three types of feature point coordinates.

2 and 3 show one embodiment of the present invention. Second
The figure is a control system diagram, and pointers 12, 13, and 14 are pointers indicating pixels in the character image storage means 8 and the character rectangular image data storage means 9. FLAG15
Is a condition selection flag for selecting a condition for determining the value of the pixel B described later.

FIG. 7A is a flowchart showing an embodiment of the character area image data creating means, and FIG.
(A) Subroutine of step 103 in the figure. In step 101, the character area image data storage means 9 is cleared, and the pointer 12 points to the pixel A (x,
y), the pointer 13 is set so as to point to the pixel B (x, y) in the character area image data storage means, and the pointer 14 is set to point to the pixel C (x, y-1) immediately above the pixel B. x and y are variables indicating the positions of the pixels in the character image storage unit and the character area image data storage unit, and are set to x = 0 and y = 1. In step 102, the pointers 12, 13, and 14 are set to addresses indicating the respective pixels, and FLAG is cleared. In step 103, a subroutine for determining whether or not to set the pixel B shown in the flowchart of FIG. 7B is called.

Hereinafter, a flowchart of the subroutine will be described. In step 201, the FLAG is checked. If the FLAG is cleared, the state of the pixel A indicated by the pointer 12 is checked in step 202, and if the pixel A is 0, the process ends. If 1, step
After setting the pixel b pointed to by the pointer 13 to 1 in 203, step 2
In step 04, the pixel c indicated by the pointer 14 is checked. If the pixel c is 0, the process is terminated. If the pixel is 1, FLAG is set in step 205 and the process is terminated. If the FLAG is set in step 201, the pixel C is checked in step 206. If 1, the pixel B is set to 1 in step 207 and the processing is terminated. If the pixel C is 1, the FLAG is cleared in step 208. Then, control is transferred to step 201.

After returning from the above-described subroutine processing, it is checked in step 104 whether or not x is the end point (the left end of the image area). If it is not the end point, x is incremented by one in step 105, and the flow returns to step 102. . If it is the end point, the value is set in each pointer in step 106, and then FLAG is cleared. In step 107, the same subroutine as in step 102 is called to determine whether or not the pixel B is set. In step 108, it is checked whether or not x is 0 (the right end of the image area). If it is not 0, x is decreased by 1 in step 109, and the process returns to step 106. If x is 0, y is incremented by 1 in step 110, and it is checked in step 111 whether y is the end point (the lower right point of the image area). If it is not the end point, the pointer is changed and then step 102 Return to If it is the end point, in step 112 the address is reset so that the pointer 12 points to the pixel D (x, y + 1) immediately below the pixel B in the character area image data area, and the process proceeds to step 113. Steps 113 to 120 are the same as steps 102 to 109, respectively. After decreasing y by one in step 121, it is checked in step 122 whether or not y is 0. If y is not 0, the value of the pointer is calculated, and the process returns to step 113. If y is 0, the process ends.

Next, an embodiment of the feature point reading means 4 will be described. In the reading of the feature points, the pixels of the character area image data are sequentially scanned with the 3 × 3 mask shown in FIG. 4A, and the pixels matching the conditions of each type shown in FIG. 4B are extracted. The feature point is obtained by doing this. Since the coordinate values are stored in the tables P, Q, and R of the feature point coordinate storage means 10 for each type of feature point, each feature point is arranged in ascending order by the value of the y coordinate.

The character circumscribed frame forming means 6 calculates the character circumscribed frame coordinates from the coordinate values of the character area image and stores the coordinates in the character circumscribed frame storage means 11. FIG. 8 is a flowchart showing means for forming a character circumscribed frame. In step 301, check whether there is feature point data in the table Q.
At 302, the coordinate value is read into registers X1 and Y1 with the feature point data at the head of table Q as feature point b. Step 30
In step 3, a feature point in a rectangular area having (0,0) and (X1, Y1) as diagonal points is searched from the table P, and if there is, the feature point A having the largest X coordinate value among them is found. The coordinates are taken out and read into registers X2 and Y2 (step 305). If not, 0 is read into the registers X2 and Y2 (step 306).
In step 307, a feature point in a range fixed by a rectangle having (X1, Y1) and (X2, Y2) as diagonal points is searched from the table R, and if not, it is found (corresponding to the case of FIG. 6 (a)). In step 309, (X1, Y1) and (X2, Y2) are written into the character circumscribed frame storage device 11 as character frame coordinate values, and the characteristic point b is deleted from the table Q and the characteristic point a is deleted from the table P (step 309).
310) and return to step 301. Step 311 if any
Then, the feature point having the largest X coordinate value among the feature points searched from the table R is taken out and the coordinate values are read into the registers X3 and Y3 as the feature point d. X3 and X1 in step 312
Are checked for equality, and if they are equal, the routine proceeds to step 315 (corresponding to FIG. 6 (c)). If not, step 31
In step 3, the x coordinate is equal to X3 at the characteristic point of table Q, and y
The coordinate value having the minimum value is set as the feature point C, and the coordinate value is read into the registers X4 and Y4 (corresponding to FIG. 6B). In step 314, compare the registers X1 and X4, assign the larger value to the register X4, then compare the registers Y1 and Y4, assign the larger value to the register Y4, and change the values of the registers X4 and Y4. The value on the table Q is rewritten as the coordinate value of the feature point c. In step 315, feature point d is deleted from table R and feature point b is deleted from table Q, and the process returns to step 301. As described above, it was proved that the character segmenting device according to the present invention can be realized.

[Effects of the Invention] As described above, according to the first aspect of the invention, it is possible to generate a right-angled polygonal character area image according to the connection state of the character pixels, and to appropriately generate the character area image from the character area image. Character frame can be obtained.

According to the invention described in claim 2, an appropriate character frame can be obtained from the combination of the feature points.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a character segmenting apparatus according to the present invention. FIG. 2 is a control system diagram. FIGS. 3A to 3C are diagrams showing examples of input data (a), data (b) being processed, and output data (c) in the character area image data creating means 2. FIG. FIGS. 4 (a) and 4 (b) are diagrams showing a 3 × 3 mask for searching for a feature point in the feature point reading means 4 and conditions of peripheral pixels. FIG. 5 is a diagram showing an example of feature points in the character area image data shown in FIG. FIGS. 6 (a) to 6 (c) are diagrams showing three types of basic patterns aggregated as character area image patterns and their characteristic points. FIGS. 7A and 7B are flowcharts showing an embodiment of the character area image data creating means 2. FIG. FIG. 8 is a flowchart showing an embodiment of the character circumscribing frame forming means 6.

──────────────────────────────────────────────────続 き Continuation of the front page (72) Inventor Yuji Nakajima 3-3-5 Yamato, Suwa-shi, Nagano Seiko Epson Corporation (56) References JP-A-63-37489 (JP, A) (58) Field surveyed (Int.Cl. ⁶ , DB name) G06K 9/34

Claims (2)

(57) [Claims] 1. A character image storage means for storing an image of a read character string as character image data composed of character pixels and background pixels; and a set of the character pixels connected to each other from the character image data. To form a right-angled polygon,
Character area image data creating means for creating character area image data representing the character area image of the right-angled polygon; character area image data storage means for storing the character area image data; and feature points at vertices of the character area image A feature point reading means for extracting feature point coordinate values, a feature point storage means for storing the feature point coordinate values, a character circumscribed frame forming means for obtaining a character circumscribed frame from the feature point coordinate values, A character circumscribing frame storing means for storing coordinate values of the character circumscribing frame, wherein the character area image data creating means comprises: a coordinate value (X, Y) in the character area image and a pixel value in the character area image Means for generating the character area image so as to satisfy the following equations (a) and (b) for G (X, Y): Character extraction device. For any X where Xs ≦ X ≦ Xe, Ymin (X) = Ys ...
(A) For any X satisfying Xs ≦ X ≦ Xe, G (X, k) ≠ G0, Ym
in (X) ≦ k ≦ Ymax (X) (b) where Xs and Xe are domains in the X coordinate of the character area image, Ys and Ye are domains in the Y coordinate of the character area image, and Ymin (X), Ymax (X) is the minimum and maximum values of the Y coordinate value of the character pixel group at the coordinate X, and G0 is the pixel value of the background pixel. 2. The character segmenting apparatus according to claim 1, wherein said characteristic point reading means has means for extracting a plurality of types of characteristic points by performing a mask process on said character area image data. The circumscribing frame forming unit applies a combination of the feature points to a plurality of basic patterns, thereby obtaining a combination of feature points corresponding to any of the plurality of basic patterns. A character segmentation device having means for obtaining coordinates of a character.