JPH1091724A - Pattern recognizer - Google Patents

Abstract

PROBLEM TO BE SOLVED: To recognize the characters by a logical method by detecting the phase geographic features out of a segment block of a two-dimensional image, detecting the secondary curve of a skeleton line of the segment block, sorting the characters based on the phase geographic features and collating the skeleton line with the secondary curve of a structural feature dictionary part. SOLUTION: A curvature detection part 5 detects the curvature of a character contour, an angle detection part 6 detects a point where the curvature is maximized as an angle, and a segment block detection part 7 detects a segment. A phase geometric feature detection part 8 detects some of phase geometric features via the segment. A skeleton line detection part 9 decides an intermediate point of the shortest distance between a marked contour and its corresponding contour. A secondary curve detection part 10 selects a secondary curve where the error is minimized between the skeleton line and the secondary curve. A sorting part 11 sorts the characters based on the phase geometric features, and a sorting part 12 confirms the structural features of characters based on the structural features of the secondary curve.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a pattern recognition apparatus for characters, drawings, figures, fingerprints, remote sensing, and the like.

[0002]

2. Description of the Related Art A method for extracting a stroke of a character input as an image with a camera or the like has not been completed at present. Also, when there is a picture such as an advertisement and the character string is mixed with vertical writing, horizontal writing, and oblique writing, the character cannot be recognized by the current character recognition method. In the current recognition method, for example, the recognition rate is expressed as 97% and the erroneous recognition rate is 3%, and it is assumed that erroneous determination occurs.

The first problem arises from the fact that a thinning method is used as a current mainstream stroke extraction method for characters. As for the thinning method, no logical thinning method for solving the distortion of the corner mustache, the intersection and the branch point has been found. The second problem is that since a character is represented by a two-dimensional vector, only a straight line can be represented and a curve cannot be represented.
Therefore, current character recognition methods do not recognize characters in a logical manner, but only distribute them to certain characters based on probability theory.

[0004]

An image input unit for inputting a two-dimensional image, an edge detection unit for detecting an edge of the two-dimensional image, a curvature detection unit for calculating a curvature of an edge of the two-dimensional image, the curvature A corner detector 6 for detecting the corners of the edge of the two-dimensional image, a line segment block detector 7 for dividing the area into the areas surrounded by the corners of the edge of the two-dimensional image, Topological feature detecting section 8 for detecting, skeleton line detecting section 9 for detecting a skeleton line of the line segment block, quadratic curve detecting section 1 for detecting a quadratic curve of the skeleton line
0, a classification unit 11 for classifying characters by geometric features, a classification unit 12 for structural characteristics of quadratic curves for classifying characters by quadratic curves, a skeleton line of the input image and a structural feature dictionary Collating unit 1 for collating the quadratic curve of the unit 14
3. A determination unit 15 for determining the result of the comparison, an output unit 15 for outputting the determination result of the determination unit, a memory unit 2 for storing image data and quadratic curve data, and a control unit 1 for controlling each block. I do.

First, a contour of a character input as an image is detected by an edge detector 4, a curvature on the contour of the character is determined by a curvature detector 5, and a point at which the curvature is maximized is determined by a corner detector 6 as a corner. The section segmented by the corner by the line segment block detection unit 7 is defined as a line segment. The topological feature detection unit 8 uses this line segment to calculate a part of the topological feature (end point, T
Branch point, Y branch point, intersection). Next, the skeleton line detection unit 9 obtains an intermediate point of the shortest distance between the contour line of interest and the corresponding contour line, and sets this intermediate point as a point on the skeleton line.
In the case of an intersection or an inclined line, a skeleton line is obtained using an auxiliary line. The quadratic curve detector 10 selects a quadratic curve that minimizes the error between the skeleton line and the quadratic curve by the least square method. Let the selected quadratic curve be the stroke of the character. The character 2
The number of straight lines and the number of curves are obtained from the next curve.

In the character recognition method, first, the classification unit 11 based on the topological features uses the topological features (the number of loops, the number of end points, the number of T-branches, the number of Y-branches, the number of intersections, the number of straight lines, and the number of curves). Characters are classified by number, number of end points, number of quasi-end points) By this classification, characters are reduced to several types or less. Furthermore,
The structural characteristic of the character is confirmed by the classification unit 12 based on the structural characteristic of the quadratic curve. If the text is skewed, specify a center point and a straight line. Affine transformation is performed by the specified angle of inclination of the straight line to return to the normal state, and the structure characteristic dictionary unit 14 based on the quadratic curve is described (the structural characteristic of the character is described, and the center point and the straight line for the affine transformation are described). And recognize and recognize the structural features of the character.

Accordingly, a character rotated at an arbitrary angle can be recognized. Also, by confirming the structural characteristics of the character (described by a quadratic curve), the recognized character is 100% correct.
Unrecognizable characters are expressed as "don't know" and no misrecognition occurs.

[0008]

[Function] In order to represent a two-dimensional image by a quadratic curve,
General images containing not only straight lines but also curves can be processed. In addition, by transforming a character into a quadratic curve representing a skeleton line segment, the transformed character can be recognized. Also, expressing a two-dimensional image by a quadratic curve is as follows:
The image is invariant to the affine transformation, and the rotation and displacement of the image do not affect the collation.

[0009]

BRIEF DESCRIPTION OF THE DRAWINGS FIG. FIG. 1 shows a method for obtaining the curvature of the edge of an image. First, the outline of the character in the input image is detected. Next, the curvature θ and the angle θk on the outline of the character are obtained. When the traveling direction is the counterclockwise direction in the XY coordinates, the point on the line segment of the outline is represented by K
And -a points from the point K are (Ka), and + a from the point K
These points are (K + a). The angle between the straight line _{L 1} and the X-axis passing through the point K and the point (K-a) and theta _1. Further, the curvature when the angle between the straight line _{L 2} and X-axis and theta ₂ which passes through the points K and (K + a) θ is determined as _{θ = (θ 1 -θ 2)} . The relationship between the curvature θ and the angle θk is as follows. θk = π−θ When the contour is convex, the angle θk is positive, and when the contour is concave, the angle θk is negative.

For the point K on the contour, the absolute value of the angle θk is the minimum value of the angle from the point (K−a) to the point (K + a), and the absolute value of the angle θk is 110
If it is less than degrees, this K point is defined as a corner. Further, the absolute value of the angle is the minimum value, and the absolute value is 110 degrees or more and 155.
If the angle is equal to or less than the degree, the K point is set as a quasi-angle. In addition, candidates for refraction points are obtained as follows. The absolute value of the angle in the vicinity of ± 2a dots from the ± a dot of the point of interest is 160 degrees or more, the angle of the angle of the point of interest is θk, and the number of dots whose angle is 160 degrees or less in the vicinity of the point of interest is S. Letting the maximum number of vertical or horizontal dots of a character be L, Q = 100 * S / {(160-.theta.k) * L} Q <1, That is, if the angle is 1 dot or less per degree of angle, refraction occurs. Point candidates.

At the same point of interest, the angle θK4 of a = 4 dots, the angle θK6 of a = 6 dots, and the angle θK8 of a = 8 dots are simultaneously obtained, and θK4 <θK6 <θK.
In the case of 8, it is regarded as rattling of a line segment, and detection of an incorrect corner is prevented. After finding the end point, the midpoint of the line segment where the curvature sign is reversed is found as the inflection point.

The topological feature is obtained by the following method. The number of character loops is determined as the number of contour lines minus one. When the skeleton line is obtained, when the number of dots on the contour line from the point of interest to the intersection of the corresponding line segment is 20 dots or less and the line segment is within 20 dots, the line segment is regarded as the end point. Also, 20
When there is no line segment in the dot, if there is a positive angle of 75 degrees or less, it is determined as an end point having no line segment. When the line of interest is a straight line, the line of interest has two corresponding line segments, and the two corresponding line segment numbers are separated by 3 or more, or two corresponding line segments are different. Is a T-branch.

The starting point of the line segment of interest is a negative angle,
If there is one or more line segments between the corresponding line segment of the target line segment and the corresponding line segment before the target line segment, or if the two corresponding line segments belong to different loops, Y Branch. Y branch is 2
The point with the number is defined as the intersection. When there is an end point between two corresponding line segments of the Y branch (for example, an intermediate end point of the number “3”), this end point is set as a quasi-end point. If one of the refraction point candidates is a straight line, the refraction point is determined. When the Y branch overlaps the T branch, the T branch is prioritized and the Y branch is deleted.

FIG. 2 shows a method of detecting a skeleton segment based on a contour line. A line segment for which a skeleton segment is to be obtained is defined as a target line segment, and a line segment for which a skeleton segment is not determined is defined as a corresponding line segment. First, an outline is separated into a line of interest and a corresponding line. Therefore, the inflection point = 20
Weights of 0, quasi-end point = 200, and refraction point = 100 are assigned.
The number of dots of a connected line segment from the end point excluding the quasi-end point is counted. If there is an inflection point, a quasi-end point, or a refraction point on the way, a weight is added. A connected line segment having a large number of counts is defined as a line segment of interest. In the case of the line segment number of interest i, the skeleton line designation array for obtaining the skeleton line segment is set to K (i) = 210. In the case of the corresponding line segment number t, Let K (t) = 1 be the skeleton line designation array. The skeleton line segment is obtained from the line segment of the skeleton line designation array K () = 210.

The skeleton line segment is separated by the angle, the inflection point, the refraction point, the end point, and the quasi-end point where the absolute value of the angle of the line segment of interest is 110 degrees or less. In addition, the priority order of the line segment for obtaining the skeleton line is such that the auxiliary line is prioritized, the T-branch line is prioritized, the straight line at both ends is inflected, and then the inner loop line Priority.

[0016] Incidentally, in FIG. 1, .theta.k the angle of the point K on the focused line segment, and, when the angle of the straight line L ₁ and the X axis is theta _1, than the angle θs is the following equation of a perpendicular point K Desired. θs = θk / 2 + θ _{1 In} FIG. 2, the point K on the line segment L _{1 of the} contour line of interest is shown.
Are (Xa, Ya). A perpendicular line is drawn at the point K, and the intersection with the line segment L2 of the corresponding contour is defined as KT ₁ (Xb ₁ , Y
b ₁ ). The distance between the points K and KT ₁ and P. Moving the points KT ₁ on the line of the corresponding contour line, the distance P points is minimized and KT _2. The coordinates of the midpoint of the points K and KT ₂ and (X, Y). The i-th skeleton line has coordinates (x _i ,
y _i ). (I + 1) -th skeleton line does not make use of perpendicular point was one advancing on the interest segments, determine the shortest distance from a point near the point KT ₂ on the corresponding segment, as the midpoint of the shortest distance Find the skeleton line.

When there is an inner contour like the numeral "O", the curvature and angle of the inner contour are negative.
In the case of a T-branch, the distance between the contour line and the skeleton line immediately before the T-branch point is stored, and the skeleton line is obtained at the stored distance during the T-branch. In the case of an intersection, the angles of the four corners are negative, and an auxiliary line is written from the negative angle of the line of interest to the negative angle of the corresponding line. From the line segment before the auxiliary line to the line segment after the auxiliary line, a skeleton line at the intersection is obtained.

Similarly, when there is an inclined line like the character "N" and the line width is thick, an auxiliary line is drawn on an extension of the inclined line from a negative angle. The skeleton line of the inclined line is obtained using the auxiliary line. Skeleton Line Inspection Method (1) When the line number of interest of a character is i (i is 10 to n) and the corresponding line number is t, the skeleton line designation array of the following equation must be established. K (i) = 210, K (t) = <1, In the following case, the skeleton line segment is abnormal. K (i) = 210, K (t) = 210, or K (i) = <1, K (t) = <1,

Inspection method of skeleton line segment (2) The following Euler characteristic in the topology must be established. However, the number of endpoints includes the number of quasi-endpoints. Number of end points-number of branch points + 2 (number of loops) = 2

FIG. 3 shows detection of a quadratic curve from a skeleton line segment.
If there are five arbitrary points on the skeleton line, a quadratic curve equation is obtained. (1) Elliptic curve (including circle) ｛(y ₁ · sin (α) + x ₁ · cos (α)) / (R
₁ )｝ ² + ｛(y ₁ · cos (α) −x ₁ · sin
(Α)) / (R ₂ )｝ ² = 1 (2) Straight line (including two straight lines) (3) Hyperbolic or parabolic h · ｛(y ₁ · sin (α) + x ₁ · cos (α)) /
(R ₁ )｝ ² -h ｛(y ₁ · cos (α) -x ₁ · s
in (α)) / (R ₂ )｝ ² = 1

Where h = 1 or h = −1, x ₁ = x−
c. x, y ₁ = y−c. y c. x is the x coordinate of the center point c. y is the y coordinate of the center point, α is the angle of inclination of the axis, and R ₁ and R ₂ are _one of three types of curves: a long radius and a short radius.

Further, a quadratic curve which minimizes the error between the skeleton segment and the quadratic curve is selected from a large number of quadratic curves by the least square method. One detected quadratic curve corresponds to one stroke of the character.

FIG. 4 shows an electric block diagram of the apparatus of the present invention. An image input unit 3 for inputting a two-dimensional image; an edge detection unit 4 for detecting an edge of a two-dimensional image; a curvature detection unit 5 for calculating a curvature of an edge of the two-dimensional image;
A corner detector 6 for detecting a corner of the edge of the two-dimensional image, a line segment block detector 7 for dividing into a region surrounded by the corner of the edge of the two-dimensional image, and detecting a topological feature from the line block Topological feature detection unit 8, skeleton line detection unit 9 for detecting the skeleton line of the line segment block, 2 of the skeleton line
A quadratic curve detecting unit 10 for detecting a quadratic curve, a classifying unit 11, 2 for classifying characters by geometrical features
Classification unit 12 for structural characteristics of a quadratic curve for classifying characters according to a quadratic curve, skeleton line of input image and structural feature dictionary unit 1
4, a collating unit 13 for collating the quadratic curve, a judging unit 15 for judging the collated result, an output unit 15 for outputting a judgment result of the judging unit, a memory unit 2 for storing image data and quadratic curve data, The control unit 1 controls the blocks.

Next, characters are classified according to topological features. (1) Number of intersection points (2) Number of branch points Number of T-type branch points Number of Y-type branch points (3) Number of loops (4) Number of straight lines (5) Number of curves (6) Number of end points (7) Number of quasi-end points (Y-branch line segment) By this classification, characters are reduced to several types or less.

For example, the numbers “2” and “5” and the number “3” in FIG. 6C are the same topological features (the number of intersection points = 0, the number of T-type branch points = 0, the Y-type branch). In some cases, the number of points = 1, the number of loops = 0, the number of straight lines = 2, the number of curves = 1, the number of end points = 2, and the number of quasi-end points = 1).

A structural feature dictionary unit 14 is created by a quadratic curve in which the following structural features are described for each character. (1) In the case of a curve, there are eight open directions depending on the combination of up, down, left, and right. (2) In the case of a straight line, the inclination angle and length. (3) Positions of connection points and end points of straight lines and curves. (4) Position of straight line and curve. (5) Center point and straight line designated for affine transformation.

When the character is inclined, the character is returned to the normal state by performing affine transformation by the inclination angle of the designated straight line. It is checked whether the quadratic curve of the character in the normal state satisfies the structural characteristics of the character described in the structural characteristic dictionary unit 14 based on the quadratic curve. If the structural characteristics of the character are satisfied, the character has been recognized. In this method, since the structural features of characters are confirmed, character structure recognition is possible.

Therefore, it is possible to cut out a designated character (when not connected to another character) in which no erroneous recognition occurs and a rotated character can be recognized. Further, even if the character is deformed, if the character satisfies the structural characteristics of the character, the character can be recognized, so that the character is resistant to character deformation.
The present recognition method can be used for numerals, uppercase alphabets, lowercase alphabets, katakana, hiragana, kanji type characters, handwritten characters, and the like.

FIG. 5 shows the progress from the numerical image to the quadratic curve. The image is shown in FIG. FIG. 2B shows the edges of the image. FIG. (C) shows the angle, which is the maximum value of the curvature, and the inflection point with a triangle. FIG. 4D shows a line segment block. FIG. 5E shows a state where the line segment block is separated. Figure (f)
Shows the skeleton line of the line segment block. FIG. 7G shows all the quadratic curves of the skeleton line of the line segment block. FIG. 7H shows the start point to the end point of the quadratic curve of the skeleton line of the image. Fig. 5-
2 shows the progress from the image of the number “2” to the quadratic curve. FIG. 5-3 shows the progress from the image of the number “3” to the quadratic curve. FIG. 5-4 shows the progress from the image of the number "4" to the quadratic curve. Fig. 5-5 shows the progress from the image of the number "5" to the quadratic curve.

(1) The character size of the input image is such that the number of vertical dots is 40 to 100 dots. (2) The error of the least square method is 10 dots or less. (3) Horizontal is within 0 ° ± 30 °, ± 180 ° ± 30
Within degrees. Vertical means within ± 90 ° ± 30 ° (4) Extract an intended single number from an image in which several to about ten numbers and characters are mixed.

Inflection point = 200N, quasi-end point = 200, refraction point = 100 are weighted. In the case of the number “2”, there are an inflection point and a refraction point, and the line segment inside the curve is set as the line segment of interest.
In the case of the number “3”, there is a quasi-end point or a refraction point, and the line segment inside the curve is set as the line segment of interest. In the case of the number “4”, an auxiliary line passing through the intersection where the auxiliary line and the inflection point are present, a line segment on the extension of the auxiliary line, and a line segment having the inflection point are taken as the line segment of interest. Numbers "
In the case of 5 ", there are a quasi-end point and a refraction point, and the line segment inside the curve is taken as the line of interest. A skeleton line is obtained from the line segment of interest. Further, a quadratic curve that minimizes the error between the skeleton line segment and the quadratic curve is selected from a large number of quadratic curves by the least squares method. Extract as a curve.

FIG. 5A shows an original drawing of the numeral "3". FIG. 5H shows the stroke of the numeral “3”. Topological features of the number "3" (number of intersection points = 0, number of T-type branch points = 0, number of Y-type branch points = 1, number of loops = 0, number of straight lines = 0, number of curves = 2, number of end points = 2, Classification is based on the number of quasi-end points = 1). Furthermore, the structure of the number "3" (the direction in which the upper curve is open is left, lower left, or lower, and the direction in which the lower curve is open is left, upper left, upper one, or upper The lower endpoint of the curve and the upper endpoint of the lower curve are connected). If the above structure is not satisfied, the end point is set as the center point of the affine transformation, and the affine transformation is performed so that the inclination angle of the straight line connecting the both end points becomes 90 degrees.
If the above structure is satisfied, it is recognized as the numeral "3".
If the above structure is not satisfied, then affine transformation is performed so as to be -90 degrees. If the above structure is satisfied, it is recognized as the numeral "3". If the above structure is not satisfied, the result is "unknown".

FIG. 5-4 shows the numeral "4". FIG. 5A shows the original drawing of the numeral “4”. FIG. 5H shows the stroke of the numeral “4”. Topological features of the number “4” (number of intersection points = 1, number of T-type branch points = 0, number of Y-type branch points = 2, number of loops = 1, number of straight lines = 3, number of curves = 0, number of end points = 2, Classification is based on the number of quasi-end points = 0). In addition, the numbers "
Check the 4 "structure (the horizontal straight line has a point of intersection with the vertical straight line, the third straight line is connected at the left end of the horizontal straight line, and the end point of the third straight line is above the horizontal straight line) If the above structure is not satisfied, the intersection is set as the center point of the affine transformation, and the inclination angle of a straight line passing through the intersection is 0 degree, 90 degrees, and 180 degrees.
Affine transformation is performed so that the angle is -90 degrees. If the above structure is satisfied, it is recognized as the numeral "4". If the above structure is not satisfied, the result is "unknown".

FIG. 5-5 shows the numeral "5". FIG. 5A shows an original drawing of the numeral “5”. FIG. 5H shows the stroke of the numeral “5”. Topological features of the number “5” (number of intersection points = 0, number of T-type branch points = 0, number of Y-type branch points = 1, number of loops = 0, number of straight lines = 2, number of curves = 1, number of end points = 2, Classification is based on the number of quasi-end points = 1). In addition, the numbers "
Confirm the 5 "structure (the horizontal straight line is above the curve, the vertical straight line is connected at the left end of the horizontal straight line, and the open direction of the curve is either the left, upper left, or upper direction). If the above structure is not satisfied, the end point of the straight line is set as the center point of the affine transformation, and the inclination angle of the straight line having the end point is 0 degree,
Affine transformation is performed so that it becomes 80 degrees. If the above structure is satisfied, it is recognized as the numeral "5". If the above structure is not satisfied, the result is "unknown".

In the line segment for obtaining the skeleton line, the line segment of the loop is prioritized. That is, the number "8" is a structure of two loops. The numbers "6" and "9" are the structure of one loop and the outer curve.

Note that several types of shapes of the same character are described in the structural feature dictionary unit 14 based on the quadratic curve. Figure 6 shows numbers
3 (a). FIG. 6 (a) shows a case composed of two curves, FIG. 6 (b) shows a case composed of four straight lines, and FIG. FIG. 6 (d) showing a case where it is composed of two straight lines and one curve, and shows a case where it is composed of one loop and two curves.

In the case of the number "3", the above four types of "3"
Possess different features in shape. Therefore, the number "3"
In the case of the above, the above four kinds of “3” are described in the structural feature dictionary unit 14 using a quadratic curve.

[0038]

Since a two-dimensional image is represented by a quadratic curve, not only a straight line but also a general image including a curve can be processed. In addition, it is possible to recognize a deformed character by regarding the deformation of the character as a deformation of a quadratic curve indicating a skeleton line segment. In addition, since the shape other than the inclination angle of the axis of the quadratic curve is invariant to the affine transformation, the rotation of the image and the recognition of the character shifted in position are possible.

Therefore, it is possible to recognize the structure of a character instead of the character guessing technique. No misrecognition occurs, but unrecognizable characters are expressed as "don't know." 360 degrees such as vertical writing, horizontal writing, diagonal writing in advertising etc.
Recognition of rotated characters is possible. Extraction of specified characters (when not connected to other characters) is possible. The transformed character can be recognized if it is within the description range of the quadratic curve expression dictionary. Therefore, it is possible to recognize printed characters, handwritten characters, symbols, drawings, fingerprints, remote sensing images, three-dimensional objects, and the like.

[0040]

[Brief description of the drawings]

FIG. 1 is a diagram showing a method of calculating a curvature of an edge of an image.

FIG. 2 is a diagram showing a method for obtaining a skeleton segment.

FIG. 3 is a diagram showing detection of a quadratic curve from a skeleton line segment.

FIG. 4 is an electrical block diagram of the device of the present invention.

FIG. 5 is a diagram showing the progress from the image to the detection of a quadratic curve.

FIG. 6 shows the type of numeral “3”.

[0041]

[Explanation of symbols]

 X, Y ... Two-dimensional coordinates 1 Control unit 2 Memory unit 3 Memory unit 3 ... Image input unit 4... Edge detection unit 5... Curvature detection unit 6... Corner detection unit 7. ………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………… 9 ... Quadratic curve detection unit 11 Classification unit based on geometric features 12 Classification unit based on structural features 13 …………………………………………………………………………… ...... Matching unit 14 ...... Structural feature dictionary unit 15 ...... Judgment unit, output unit

────────────────────────────────────────────────── ───

[Procedure amendment]

[Submission date] December 2, 1996

[Procedure amendment 1]

[Document name to be amended] Drawing

[Correction target item name] All figures

[Correction method] Change

[Correction contents]

FIG.

FIG. 2

FIG. 3

FIG. 5

FIG. 9

FIG. 4

FIG. 6

FIG. 7

FIG. 8

[Procedure amendment 2]

[Document name to be amended] Statement

[Correction target item name] Brief description of drawings

[Correction method] Change

[Correction contents]

[Brief description of the drawings]

FIG. 1 is a diagram showing a method of calculating a curvature of an edge of an image.

FIG. 2 is a diagram showing a method for obtaining a skeleton segment.

FIG. 3 is a diagram showing detection of a quadratic curve from a skeleton line segment.

FIG. 4 is an electrical block diagram of the device of the present invention.

FIG. 5 is a diagram showing a process from an image of a number “2” to detection of a quadratic curve;

FIG. 6 is a diagram showing the progress from the image of the number “3” to the detection of a quadratic curve.

FIG. 7 is a diagram showing the progress from the image of the number “4” to the detection of a quadratic curve.

FIG. 8 is a diagram showing the progress from the image of the number “5” to the detection of a quadratic curve.

FIG. 9 is a diagram showing the type of the numeral “3”;

[0041]

[Explanation of Signs] X, y: Two-dimensional coordinates 1: Control unit 2: Control unit: Memory unit 3: Image Input unit 4 Edge detection unit 5 Curvature detection unit 6 Angle detection unit 7 Angle detection unit 7 ………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………………… 9 ... Quadratic curve detector 11 Classifier based on geometric features 12 Classifier 13 based on structural features 13 Matching unit 14 Structural feature dictionary unit 15 Judgment unit, output unit

Claims (1)

[Claims] An image input unit for inputting a two-dimensional image, an edge detection unit for detecting an edge of the two-dimensional image, a curvature detection unit for calculating a curvature of an edge of the two-dimensional image, and an edge of the two-dimensional image based on the curvature. A corner detecting unit for detecting a corner, a line segment block detecting unit for dividing into a region surrounded by the corner of the edge of the two-dimensional image, and a topological feature detecting unit for detecting a topological feature from the line block A skeleton line detection unit for detecting a skeleton line of the line segment block, a quadratic curve detection unit for detecting a quadratic curve of the skeleton line, and a classification unit based on topological features for classifying characters by topological features. A classification unit based on the structural characteristics of the quadratic curve, a verification unit that verifies the quadratic curve of the skeleton line of the input image with the quadratic curve of the structural characteristic dictionary unit, a determination unit that determines the result of the verification, and a determination unit that determines Output unit for outputting the result, image data And a memory unit for storing quadratic curve data, and a control unit for controlling each block.
In particular, a pattern recognition apparatus is provided with a classification unit based on topological features and a classification unit based on the structural characteristics of a quadratic curve, and recognizing characters by checking the structural characteristics of the quadratic curve.