JP2881080B2 - Feature extraction method - 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 reader (OCR) for optically reading a document or the like.
The present invention relates to a feature extraction method for extracting information of a straight line portion that crosses a character graphic pattern horizontally or traverses vertically, such as a strike-through line used when deleting a character graphic pattern, with high accuracy. .

[0002]

2. Description of the Related Art Conventionally, a flow chart shown in FIG. 6 is used to extract a straight line portion which traverses a character / graphic pattern horizontally or vertically, such as an erasing line used for deleting a character / graphic pattern. Was applied. The method of FIG. 6 extracts all stroke information of a character / graphic pattern, and can naturally also extract a straight line portion such as a strike-through line which is a type of stroke.

In step S60 shown in FIG.
First, a character / graphic pattern is converted into a binary image 60 by a photoelectric conversion device.
, And in the next step S61, feature points 61 of the binary image
Is extracted. The feature points 61 are obtained by performing a 3 × 3 mesh mask scan on the thinned binary image, and finding end points, branch points, intersections, isolated points, and the like obtained from the pattern, and character lines connecting those points. Is the refraction point obtained by the tracking. These feature points are regarded as approximate stroke points of the binary image 60.

In the next step S62, feature point matching is performed between a standard pattern (online character standard pattern 62) having stroke order information prepared in advance and the feature point 61 of the binary image obtained in step S61. In this feature point matching, all Euclidean distances between each stroke approximation point of the online character standard pattern 62 and the feature point 61 of the binary image are obtained.

[0005] In the next step S63, the step S
The feature point 61 of the binary image closest to the distance measured by the Euclidean distance obtained in 62 is associated. At this time, the feature points 63 of the binary image associated with each stroke approximation point of the online character standard pattern 62 are stored in the memory.

In the last step S64, based on the stroke order information held in the online character standard pattern 62, each stroke approximation point of the standard pattern 62 is connected to the corresponding feature point 63 of the binary image to obtain stroke information. 6
4 is extracted.

[0007]

However, if the above-described stroke information extraction method is applied to the extraction of a straight line portion which traverses a character / graphic pattern horizontally or vertically, it has the following problems. Occurs. (1) Since an online character information pattern having stroke order information must be prepared, labor and memory capacity increase. (2) The amount of calculation increases due to feature point matching. (3) In particular, in the case of an erasing line or the like, the position coordinates of the center line are not constant, and the number of lines is not always one. For straight lines with such uncertainties,
It is very difficult to apply the extraction method using the matching result with the standard pattern.

According to the present invention, the use of the above-described on-line character standard pattern having stroke order information and feature point matching, the problem of labor, an increase in memory and the amount of computation, and a straight line having an uncertain element such as an erasure line In order to eliminate problems such as difficulty in extracting parts, a method to extract information about linear parts more efficiently by finding universal features that are not related to the position or number of linear parts such as strike-through lines etc. The purpose is to provide.

[0009]

In order to solve the above-mentioned problems, the present invention provides a step of extracting information on a circumscribed rectangle circumscribing a character graphic pattern, and a step of extracting a peripheral distribution of the character graphic pattern. A step of extracting an end point of the character / graphic pattern; and a step of extracting a straight line portion of the character / graphic pattern based on information on the circumscribed rectangle, a peripheral distribution, and an end point.

[0010]

According to the present invention, a provisional straight line portion is extracted based on the circumscribed rectangle information and the peripheral distribution, and based on the principle that the shape of the rectangle determined by both ends of the true straight line portion is a flat shape. Since the value of a simple evaluation formula using the information on the circumscribed rectangle, the position information of the provisional straight line portion, and the position information of the end point is used to determine whether or not the line portion is a true straight line portion, the memory and the amount of calculation increase. Problems such as the difficulty of extracting straight lines with uncertain elements such as strike-throughs have been solved, and universal features independent of the position and number of straight-lines such as strike-throughs can be extracted. Information about a portion can be extracted more efficiently.

[0011]

FIG. 1 is a flowchart showing an embodiment of the present invention. Each processing step will be described below in order. First, in step S1, a character / graphic pattern written on a document or a form is converted into a binary image 10 by a photoelectric conversion device.

In the next step S2, characters are cut out from the binary image 10, and the X of the rectangular frame circumscribing the character portion is extracted.
The minimum and maximum values 11 of the coordinates and the Y coordinates are derived. FIG. 2 (A)
2A shows an example of a cut-out binary image. The character portion in FIG. 2A is an example in which one strike-through line is vertically written for the numeral 2, and is cut out by the circumscribed frame 20. Have been. Here, assuming that the vertex at the upper left corner of the circumscribed frame 20 is the origin, the minimum and maximum values of the X coordinate and the Y coordinate are as follows. Minimum value of X coordinate (MinX) = 0 (1) Maximum value of X coordinate (MaxX) = Xe (2) Minimum value of Y coordinate (MinY) = 0 (3) Maximum value of Y coordinate (MaxY) = Ye ( 4) The width (ΔX) and height (ΔY) of the circumscribed frame are as follows. ΔX = MaxX−MinX + 1 = Xe + 1 (5) ΔY = MaxY−MinY + 1 = Ye + 1 (6)

In step S3, a peripheral distribution 12, that is, a histogram representing the distribution of the number of black spots when the binary image is projected on the X axis or the Y axis, is extracted from the binary image 10. FIG.
2B shows a peripheral distribution of the binary image shown in FIG. 2A projected on the X axis, and FIG. 2C shows a peripheral distribution projected on the Y axis.

In step S4, the end points of the binary image are extracted. The end points are, for example, points 21, 22, 2 in FIG.
Points at the leading end of the character line, such as 3 and 24, are shown in FIG.

In step S40 of FIG.
The contour tracking of 0 is performed, the contour point coordinate series 40 is stored in the memory, and the line width is calculated in step S41.

In the next step S42, a point at which the manner of change of the X coordinate or the Y coordinate changes is determined with reference to the contour point coordinate series 40 in the forward direction (counterclockwise). This point is called an inflection point, and all the inflection points 42 of the binary image are stored in the memory. For example, in the example of FIG. 5A, points 50 and 51 are extracted as inflection points. However, the definition of the inflection point in this case is different from the usual mathematical definition.

Next, in step S43, starting from the inflection point 42 (for example, the inflection point 50 in FIG. 5A) by the forward search, the line width × α (α is a constant, α to 1.. 5)
And an inflection point showing a change similar to the inflection point that is the starting point is extracted. When the inflection point (inflection point 51 in the example of FIG. 5A) exists, in step S44, the middle point between the inflection point by the forward search and the inflection point by the reverse search is set to the temporary end point 44. And FIG.
In the example of (A), the point 52 is a temporary end point.

In step S45, it is determined whether or not the temporary endpoint is a true endpoint as follows. That is,
Two points separated by a line width × β (β is a constant, β−1.5) in the forward and reverse directions on the contour point coordinate series 40 from the temporary end point are obtained, and the distance d between the two points is determined by the line width × γ (γ is a constant, γ <
When it is 2β, γ to 2) or less, the temporary end point is regarded as an end point. In the example of FIG. 5B, points 53 and 54 which are separated from the temporary end point 52 by a line width × β in the forward and reverse directions are detected, and the distance d between these two points is equal to or less than the line width × γ. The temporary endpoint 52 is extracted as an endpoint. As described above, the end point 13 in FIG. 1 is extracted.

In step S5 of FIG. 1, a temporary straight line portion 14 is extracted from the width (ΔX) and height (ΔY) of the circumscribed frame and the peripheral distribution 12. Hereinafter, this extraction method will be described. First, using ΔX and ΔY, thresholds Thy and Thx applied to the peripheral distribution projected on the X-axis and the peripheral distribution projected on the Y-axis are obtained by, for example, the following equation. Thy = (ΔX + ΔY) / 2 (however, when Thy <2/3 (ΔY), Thy = 2/3 (ΔY)) (7) Thx = (ΔX + ΔY) / 2 (where Thx <2/3 (ΔX) Thx = 2/3 (ΔX)) (8) These thresholds are applied to the respective peripheral distributions, and a region where the number of black spots is equal to or larger than the threshold is defined as a temporary straight line portion. For example, in the marginal distribution projected on the X axis in FIG.
A region where the number of black points is equal to or larger than the threshold Thy is x ₁ ≦ x ≦ x ₂ (9), and this portion is extracted as the temporary straight line portion 14. In FIG. 2C, in the marginal distribution projected on the Y axis, the area where the number of black points is equal to or larger than the threshold Thx is represented by y ₁ ≦ y ≦ y ₂ (10). That is, in the binary image of FIG. 2A, there is a possibility that there is one straight line portion that crosses in the horizontal direction and one straight line portion that crosses in the vertical direction.

However, depending on the character, the marginal distribution 1
It is possible that the number of black points is less than or equal to the threshold value in the entire region of No. 2. In such a case, it is determined in step S8 that there is no linear portion, and the process returns to a state of waiting for input.

When the provisional straight line portion 14 exists, it is necessary to calculate a distance D indicating a deviation from an ideal straight line. As a preprocessing for the calculation, it is determined whether or not the distance D can be calculated. Is determined (step S9). In the calculation of the distance D, it is assumed that the provisional straight line portion 14 has one end point B at each of both ends. Therefore, assuming that the number of the end points B is n, the condition for calculating the distance D can be, for example, n ≧ 2 (11). In the example of FIG. 2A, since n = 4, it is determined that the distance D can be calculated.

When the equation (11) is satisfied, the distance D is calculated in step S6. 2 (A) as an example.
This calculation method will be described. First, the X coordinate xm of the intersection of the straight line passing through the center and the X-axis in the temporary straight line portion (see FIG. 2B) that vertically crosses is determined. The coordinate value x _m is defined as x _m = (x ₁ + x ₂ ) / 2 (12) as the X coordinate which gives the maximum number of black points or the middle point of the area shown by the above equation (9).

Next, a description will be given distance D _V representing the displacement between the straight line and the ideal vertical line of temporary traversing vertically. An ideal straight line is defined by the coordinates of both ends at an end point a (x _a ,
y _a), the end point b (x _b, when the y _b) (however, y _a ≦ y
and _b), it is possible say linearly as close to x _m can their X coordinate. Therefore, | x _a −x _m | and | x _b −
x _m | can be a component of the distance D _V. Further, since the vertical character portion, Y-coordinate value y _a of one of the end points a is equal to MinY, the Y-coordinate value y _b of the other end point b should be equal to MaxY. Therefore y _a -MinY and MaxY-y _b even if the distance D _V
Element.

[0024] From the foregoing, wherein at four elements can be regarded as a distance D _V all added amounts of, because it can not absorb the variation due to the size of the graphic character pattern that it anyway, of the graphic character pattern size ΔX + Δ
And Y, by performing normalization using the amounts, such as an expression representing the distance _{D V, D V = 100 ×} (| x a -x m | + | x b -x m | + y a -MinY + MaxY−y _b ) / (ΔX + ΔY) (13)

In the same manner, the distance D _H with respect to a tentative straight line crossing horizontally can be obtained. As shown in FIG. 2C, an intersection point y _m between the center line of the temporary temporary straight line portion in the horizontal direction and the Y axis is defined as the center of the region represented by the equation (10), similarly to the above x _m . y _m = (y ₁ + y ₂ ) / 2 (14) Also, the end points selected as end point candidates at both ends of the provisional straight line portion are c (x _c , y _c ) and d (x _d ,
Assuming that x _c ≦ x _d as y _d ), as an example of an expression representing the distance D _H , D _H = 100 × (| y _c −y _m | + | y _d −y _m | + x _c −Min X + Max X −x _d ) / (ΔX + ΔY) (15)

With respect to the binary image shown in FIG.
An example of calculating the distance D based on 3) and (15) is shown below. It is assumed that four end points 21, 22, 23, and 24 have been extracted from the binary image of FIG. 2A in the processing before the calculation of the distance D. First, the end point 21 having the closest X-coordinate in x _m as the end points candidates of the linear portion of the vertical provisional
(X ₂₁ , y ₂₁ ) and end point 23 (x ₂₃ , y ₂₃ ) are selected. From these coordinate values and the expressions (3), (4) and (13), the above _{DV is given} as _DV = 100 × (| x ₂₁ -x _m | + | x ₂₃ -x _m | + y ₂₁ + y _e -y ₂₃ ) / (ΔX + ΔY) (16) is obtained.

Also, as a candidate for both end points of the temporary temporary straight line portion in the horizontal direction, y is selected from those excluding the end points 21 and 23.
_The end point 22 (x ₂₂ , y ₂₂ ) and the end point 24 (x ₂₄ , y ₂₄ ) having the Y coordinate value closest to _m are selected. These coordinate values and the formula (1), (2) and as the D _H from equation _{(15), D H = 100} × (| y 22 -y m | + | y 24 -y m | + x 22 + x _e− x ₂₄ ) / (ΔX + ΔY) (17) is obtained. Thus, the process of step S6 in FIG. 1 ends.

In step S7, based on the distance D calculated in step S6, it is determined whether or not the provisional straight line portion is a true straight line portion. As a method of this determination, for example, there is a method of determining a true straight line portion when D <Th (Th is a constant Th-10) (18).

When applying the determination based on equation (18) to the above equations (16) and (17),
(A) and (B) clearly show that | x ₂₁ −x _m | << ΔX, | x ₂₃ −x _m | << ΔX, y ₂₁ = 0,
Since the _{y e -y 23 = 0, D} V it can be seen that equation becomes extremely small value (18) is satisfied. Therefore, the temporary linear portion in the vertical direction can be determined as a true linear portion. However, equation (1)
In FIG. 7), it is apparent from FIGS. 2A and 2C that | y ₂₂ −y _m |> ΔY / 2, | y ₂₄ −y _m | << ΔY, x _{22 to} Δ
Since X / 4 and x _e −x ₂₄ = 0, _DH takes a relatively large value, and the equation (18)
Does not hold. Therefore, it can be determined that the temporary linear portion in the horizontal direction in FIG. 2A is only a part of the stroke constituting the character and not a linear portion that crosses the character like a strike-through line. As described above, the distance D is equal to the threshold Th.
Only the smaller temporary straight line portion is extracted as a true straight line portion, and all the steps are completed.

In the binary image of the character graphic pattern as shown in FIG. 3A, the peripheral distribution projected on the X axis (FIG. 3B) and the peripheral distribution projected on the Y axis (FIG. 3C) ) Have two local maxima, and two tentative straight line portions are extracted in the horizontal and vertical directions, respectively. It goes without saying that even when there are two or more temporary straight line portions, the expressions (13) and (15) can be applied to each of them.

Equations (13) and (15) are merely examples, and include information on the position of the tentative straight line extracted from the peripheral distribution, the position coordinates of the end point selected by an appropriate method, and the circumscribed frame. If the distance indicating how far the provisional straight line portion 14 is away from the ideal straight line portion can be expressed by combining the minimum and maximum values of the X coordinate and the Y coordinate, the same result as in the present embodiment can be obtained. Can be

In the above-described embodiment, when there are two or more portions extracted as temporary straight lines, the end points used for calculating the distance D as both end point candidates are prohibited from being used repeatedly. However, it can be changed to allow duplication. At this time, end points 23 and 24 are selected as both end point candidates of the temporary horizontal straight line portion in FIG. 2A. In this case as well, the distance _DH is compared with the true straight line portion. It becomes larger and the same judgment result as in the above-described example can be obtained.

In the above-described embodiment, the linear distribution parallel to the X-axis and the Y-axis is extracted using the marginal distribution projected on the X-axis and the Y-axis. By using the marginal distribution projected on the Y 'axis, X'
This embodiment can be extended to extract a straight line portion parallel to the axis and the Y 'axis, that is, a straight line portion obliquely crossing the character / graphic pattern.

[0034]

As described above in detail, according to the present invention, a temporary straight line portion is first extracted by the peripheral distribution without assuming in advance the number and the position of the straight line portion such as an erasure line. In addition, the distance to the ideal straight line portion is expressed by the information on the position of the temporary straight line portion, the end point, and the circumscribed frame coordinates. Based on the distance, the temporary straight line portion is a true straight line portion. Or not.
Flexible feature extraction that does not depend on the position or number of the straight line portions can be performed. Further, since a large amount of calculation is not required and a standard pattern or the like is not used, processing time and memory can be saved. There is also an advantage that automatic extraction is possible.

[Brief description of the drawings]

FIG. 1 is a flowchart showing a processing procedure according to an embodiment of the present invention.

FIG. 2 is an explanatory diagram of a feature extraction method.

FIG. 3 is an explanatory diagram of a feature extraction method.

FIG. 4 is a flowchart showing a procedure for extracting an end point;

FIG. 5 is an explanatory diagram of an end point extraction method.

FIG. 6 is a flowchart showing a processing procedure of a conventional stroke extraction method.

Claims (1)

(57) [Claims] A step of extracting information about a circumscribed rectangle circumscribing the character / graphic pattern; a step of extracting a peripheral distribution of the character / graphic pattern; a step of extracting an end point of the character / graphic pattern; Extracting a straight line portion of the character / graphic pattern based on information, a peripheral distribution, and an end point.