JP2658137B2 - Character recognition method - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a character recognition method in a character reading device.

(Prior Art) Various methods have been proposed as character recognition methods in conventional character reading devices. Stroke analysis and pattern matching methods are widely used for character recognition for alphanumeric characters and katakana.In addition, methods to analyze the line structure of patterns and background structures to absorb character deformation are used. There is a method of analysis.

However, in the above-described method, for example, in the case of the stroke analysis method, since the characteristics of each character are described in a dictionary including the deformation, the dictionary becomes enormous, and thus the processing time becomes long. There is a disadvantage that the size of the wear becomes large.

In a method of analyzing a background structure, a character is identified by focusing on a background portion of the character and extracting a feature such as a loop of a character line or a concave or convex shape. There is a disadvantage that the processing method is complicated.

Conventionally, as a technique in such a field for removing such a defect, there is a technique described in Japanese Patent Application Laid-Open No. Sho 59-161784 (online character recognition large classification method). Hereinafter, the configuration will be described with reference to FIG.

A tablet 11 that generates data indicating the coordinates of a writing instrument when writing characters, and an online character recognition device that recognizes written characters based on data generated from the tablet, converts analog data from the tablet into digital data. Means 12 for temporarily storing digital data including the start point and the end point of the stroke and
(3) and (14), a means 16 for storing the result of performing the calculation 14 of the feature amount including the length of the line segment, the relative position, and the like from the data after the preprocessing, and data stored in advance or stored in advance. Means 18 for comparing the magnitude relationship between the data 17 which is the modified data and the data of the predetermined calculation result, and identifies the distribution position of the stroke line segment of the written character by writing the written character. Was aware of.

Then, the line length of the entire stroke and the position where each segment exists are used as the feature amounts.

(Problems to be Solved by the Invention) However, in the conventional configuration shown in FIG.
In character recognition, the starting point and the ending point of a stroke must be recorded, and there is a problem in shortening the processing time for reducing the hardware scale. For this reason, using the contour extraction method, the starting point may be any point on the contour, that is, it is not necessary to consider the stroke. Since the method using the contour extraction method is not the object of the present invention, the description is omitted. In order to further improve the character recognition ability, a feature amount is extracted for each line segment as a feature amount.

However, when extracting and storing the feature amount for each line segment inclination, if the length of the line segment is short, the amount of information increases with the increase in the number of line segments to be stored, so that the scale of hardware and the processing time can be reduced. Causes problems.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a character recognition method in a character reading device that extracts a bending point so as to reduce the recording of a feature amount for each line segment inclination while keeping the character recognition power constant.

(Means for Solving the Problems) The present invention provides a character having a means for reading a character on a form and storing binarized character pattern data and a means for extracting a contour of the character pattern from the data. Means for extracting a bending point at which adjacent outline segments bend from each other from data obtained by extracting the outline of the character pattern, and determining a slope of the line segment between the coordinates at which the bending point is extracted. Means for calculating a characteristic amount indicating a distribution position that is a relative position from a predetermined reference point based on the x, y contour coordinates of each line segment for each of the determined inclinations, and a characteristic amount stored in advance And a means for comparing the feature amount of the calculation result with the calculation result, and by recognizing the character by identifying the distribution position of the line segment of the input character pattern according to the inclination of the line segment between the x and y outline coordinates. Features.

(Operation) According to the present invention, the character recognition method is configured as follows, and the means for extracting the inflection point uses x and y coordinates at both ends of the contour line from the data from which the contour of the character pattern is extracted. , y-direction difference is calculated as a sine (sign of positive, negative, 0), the change point of the sine state is extracted as a bending point, and the means for judging the inclination of the line segment is the data obtained by extracting the contour of the character pattern The slope of the segment, which is a line segment between the coordinates connecting the inflection points, rises to the upper right, descends to the right, determines whether it is horizontal or vertical, and the means for calculating the feature amount is based on the slope of the segment. X,
Based on the y-contour coordinates, a feature amount indicating a distribution position that is a relative position from a predetermined reference point is calculated.

The means for comparing the previously stored feature amount with the feature amount of the predetermined calculation result includes, for all the recognition target characters, the difference in matching between the previously stored feature amount and the calculation result feature amount. Then, sorting by the matching distance is performed, and the character having the minimum matching distance is selected as the recognition result.

Therefore, it is possible to provide a character recognition method in a character reading device which can eliminate the above-mentioned problems and can reduce the scale of hardware and the processing time.

(Embodiment) FIG. 1 is a block diagram showing an embodiment of the present invention. In the figure, 1 is a pattern register for storing a binarized character pattern, 2 is an outline extraction unit for extracting the outline of the character pattern based on the contents of the pattern register 1,
Reference numeral 3 denotes a bending point extraction unit for extracting a line segment bending point from the x, y contour coordinates extracted by the contour extraction unit 2, and 4 denotes a gradient of a line segment between bending points extracted by the bending point extraction unit 3, that is, right. An inclination determining unit for determining whether the image is upward, downward, rightward, horizontal or vertical; a characteristic amount calculating unit for calculating a characteristic amount based on a relative position of a line segment of a character pattern for each of the inclinations determined by the inclination determining unit;
Is an identification unit for identifying a character pattern.

 Hereinafter, the operation of the present embodiment will be described.

The pattern register 1 reads the characters on the form,
The binarized character pattern is input by a character pattern input unit (not shown). Here, it is assumed that the character pattern shown in FIG. 3 is stored.

The contour extracting unit 2 determines the fourth
As shown in the figure, contour tracing is performed with the coordinates (9, 32) and (11, 24) as starting points, and contour features are extracted. Since the contour extraction method is not the object of the present invention, the description is omitted, but the contour can be easily extracted. The contour tracking start point may be any point.

Next, the inflection point extraction unit 3 signs each difference in the x and y directions at both ends of the line segment from the x and y outline coordinate data of both ends of the line segment extracted by the outline extraction unit 2 into a sine (positive, negative, 0 Is calculated as
The change point of the sign state is extracted as a bending point.

X between data, y direction sign Xsign, Ysign look at _{Xsign = X i -X i-1} Ysign = Y i -Y i-1, +, 0, - expressed in. The number of times the states of Xsign and Ysign between the data obtained in this way are continued is counted, and when the same state continues for n or more times, the state changes to a different state (x _i , y _i ) as an inflection point. register. In this description, n = 2.

FIG. 4 shows a bending point obtained by adding a start point and an end point in addition to the points obtained in this manner. The line segment connecting the bending points is hereinafter referred to as a segment.

The feature point information obtained as described above is used as the
Is output to Based on the feature point information obtained by the feature point extraction unit 3, the slope determination unit 4 determines the slope of the segment, that is, whether the segment is up-right, down-right, horizontal or vertical.

For right up the x _i> x _{i + 1} with y _i> y _{i + 1} or x _i <x _{i + 1} with y _i <y _{i + 1} for downhill x _i> x _{i + 1} with y _i < y _{i + 1} or x _i <x _{i + 1} and y _i > y _{i + 1 For} horizontal, x _i = x _{i + 1} and for y _i ≠ y _{i + 1 For} vertical y _i = y _{i + 1} and x By determining each condition of _i ≠ x _{i + 1} , the inclination of the segment is determined.

FIG. 5 shows the coordinates of the x and y bending points and the inclination of the character pattern. In the tilt term in FIG. 5, “/” indicates upper right, “Δ” indicates lower right, “−” indicates horizontal, and “|” indicates vertical.

The feature amount calculating unit 5 separates each of the tilts determined by the tilt determining unit 4
The feature quantity consisting of the relative position of the segment is calculated from the coordinates of the x and y bending points.

 The feature amount calculation will be described in detail below.

The position where each segment exists is used as the feature value. The location where a certain segment exists in the x and y directions is quantified for each of the x and y components, and the average value is calculated.

 First, consider the upper right segment.

For the position of the x component on the x axis, Similarly, for the position of the y component on the y axis, Is represented by

Here, H _x, H _y indicates the character pattern character width.

For the position of the x component on the y axis, And the position of the y component on the x axis is Is represented by

Similarly, for the segment descending to the right, a relational expression equivalent to Q ₅ = Q ₁ for the position of the x component on the x axis, a relational expression equivalent to Q ₆ = Q ₂ for the position of the y component on the y axis, The position on the y-axis of the x component is represented by a relational expression equivalent to Q ₇ = Q _3, and the position of the y component on the x-axis is represented by a relational expression equivalent to Q ₈ = Q ₄ .

For a horizontal segment, a relational expression equivalent to Q ₉ = Q ₁ for the position of the x component on the x axis, a relational expression equivalent to Q ₁₀ = Q ₃ for the position of the x component on the y axis, and more vertical For the segment, the position on the y-axis of the y component is expressed by a relational expression equivalent to Q ₁₁ = Q _2, and the position of the y component on the x-axis is expressed by a relational expression equivalent to Q ₁₂ = Q ₄ .

The Q ₁ to Q ₁₂ are rectangular bottom left (FIG. 5 (1,1)) circumscribing a character pattern performs calculation by the coordinate system with the origin. Therefore, the position where the x, y components of each segment are present decreases as the position on the y-axis goes down and the position on the x-axis goes to the left, and the feature amount decreases. Since all the weights related to the positions of all segments are added, the change in the segment near the origin appears as a small difference, and the change in the segment in the upper right part on the rectangle circumscribing the character pattern causes Will be done. Therefore, the calculations relating to Q _{1 to} Q ₁₂ in the coordinate system having the origin at the upper right of the rectangle circumscribing the character pattern (FIG. 5 (22, 32)) are performed, and the results are expressed as Q ₁₃ to Q _13
24 .

Using the formula Q ₁ to Q ₂₄ described above, x shown in FIG. 6, performs a calculation by y inflection point coordinates and inclination data. Operations are performed on the outer edge contour shown in FIG. 6A and the inner edge contour shown in FIG. 6B, and as a result, Q _{1 to} Q ₂₄ are obtained as the sum of the outer edge and the inner edge contour. Thus, the above operation is
For a character having a plurality of inner edges, the operation is performed on all the segments.

The identification unit 6 stores the feature amounts of Q _{1 to} Q ₂₄ for each recognition target character. The identification unit 6 matches the characteristic amounts Q _{1 to} Q _{24 of} the input character pattern calculated by the characteristic amount calculation unit 5 with the characteristic amounts Q _{1 to} Q _{24 of the} character pattern stored in the identification unit 6. The difference in these matchings is defined as a matching distance, and this calculation is performed for all the recognition target characters stored in the identification unit 6, and sorting based on the matching distance is performed.

According to the sorting result, the character having the minimum matching distance is selected as the recognition result.

(Effects of the Invention) As described in detail above, according to the present invention, the inflection point is extracted, and the feature amount for each inclination is extracted in consideration of the inclination of the line segment between the coordinates. It can be expected that the size of the hardware in the character reading device and the processing time can be reduced while maintaining the character recognition.

[Brief description of the drawings]

FIG. 1 is a block diagram of one embodiment according to the present invention, FIG. 2 is a block diagram of a conventional online character recognition large classification method, FIG. 3 is a character pattern diagram, FIG. The drawing is a bent point extraction diagram, and FIG. 6 is a bent point coordinate diagram. 1: Pattern register, 2: Contour extraction unit 3: Bend point extraction unit, 4: Skew determination unit 5: Feature calculation unit, 6: Recognition unit

Continuation of the front page (56) References JP-A-62-290985 (JP, A) JP-A-61-70674 (JP, A) JP-A-58-114181 (JP, A) IEICE Technical Report PRL82- 29 IEICE Technical Report PRL82-79

Claims (1)

(57) [Claims] 1. An apparatus for recognizing a character, comprising: means for reading a character on a form and storing data of a binarized character pattern; and means for extracting a contour of the character pattern from the data. Means for extracting a bending point at which adjacent contour lines bend from each other from data obtained by extracting the contours of the pattern; means for determining the inclination of the line segment between x, y contour coordinates connecting the bending points; Means for calculating a feature amount indicating a distribution position that is a relative position from a predetermined reference point based on the x, y contour coordinates of each line segment for each of the determined inclinations; and a feature amount stored in advance and the calculation Means for comparing the result with the characteristic amount, and the character is recognized by identifying the distribution position of the line segment of the input character pattern for each line segment inclination between the x and y outline coordinates. Character recognition method.