JP2987877B2 - Character recognition method - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a character recognition method for recognizing a character from a read character pattern.

2. Description of the Related Art In recent years, there has been an increasing demand for using a character recognition device as an input device for a computer or the like, and a character recognition device capable of efficiently obtaining a stable recognition result is indispensable for improving the performance of a system such as a computer. It has become.

As a conventional character recognition method, the outermost point in a predetermined direction is determined from among the outlines of a binary image forming a character, and the outline is converted into a convex line segment, a concave line segment, and a hole line segment from this information. In some cases, a character amount is determined for each line segment, and a character pattern is determined by matching the determined characteristic amount with a dictionary prepared in advance.

Problems to be Solved by the Invention As described above, in the conventional character recognition method, in order to divide a contour line into a concave line segment, a convex line segment, and a hole line segment, the most specific direction of the contour line is determined. Although the outer point is obtained, this operation does not obtain the cluster of black pixels constituting the character pattern, that is, the position information of the island.
When a separated character having two or more islands such as "j" and "j" is to be recognized, the position information of the island has to be separately obtained.

Means for Solving the Problems In order to solve the above problems, the present invention determines a plurality of points constituting a convex hull as information for dividing a contour line of an original image into a concave line segment and a convex line segment. When the character pattern is viewed from the reference point, the contact point between the tangent line containing the character pattern and the character pattern and the data indicating the angle of the nodal line are obtained for each reference point, and the convex closure obtained by connecting the obtained contact points is obtained. Whether the character pattern protrudes with respect to each line segment, and the characteristics of the stored character pattern are obtained from data of the angle between the reference line and the tangent passing through the reference point.

Further, since the points constituting the convex hull are obtained by a table reference operation, no arithmetic operation is required.

Operation In the present invention, the position information of the island is obtained at the same time by the operation for obtaining the points constituting the convex hull.

EXAMPLES The present invention will be described with reference to additional drawings showing an example.

In FIG. 1, reference numeral 1 denotes an image input unit for inputting a recognition target character as a binarized original image and storing it in an image memory;
Is an image memory for storing the input original image, 3 is an outline detection unit for extracting the input original image from the image memory, and obtaining an outline which is a pixel row constituting an outline of the original image.
Are obtained from the azimuths of a plurality of reference points that define the convex hull of the original image in the contour line, and based on the azimuth information, a concavo-convex information detecting unit that divides the contour line into concave sections and convex sections. Is a feature extraction unit that obtains a graphic feature based on the concavo-convex information obtained in 4, and 6 compares the feature extracted in 5 with a dictionary prepared in advance, and determines a character having the closest feature amount as a candidate character. A recognition processing unit 7 for outputting a dictionary in which standard characteristics of characters to be recognized are obtained and stored in advance;
Is an internal bus connecting the parts 1 to 7. Reference numeral 9 denotes a character cutout unit that cuts out one character from the character pattern stored in the image memory 2.

FIG. 2 shows a flow chart of the overall processing of the character recognition apparatus of the present embodiment configured as described above, and its operation will be described below, taking as an example the case where the character "?" Is to be recognized.

Enter the character "?" As a binary image in the image input unit,
This is stored in the image memory as an original image. The input original image is shown in FIG.

The original image is taken out by the contour detection unit, and the contour detection processing is performed. Hereinafter, the processing will be described.

First, black pixels connected by eight connections in the original image are grouped together, and a different name (hereinafter referred to as a label) is assigned to each connected component.

Fig. 3 shows the result of label assignment, "a",
“B” is the label name of each connected component. After assigning labels, an outline is obtained for each label. An outline is a white pixel with a labeled black pixel on either the top, bottom, left, or right, and the starting point is the white pixel on the left of the first black pixel found by television scanning from the upper left corner Observe black pixels counterclockwise while looking at the left hand.

FIG. 4 shows the result of detecting the contour. In FIG. 4, a1
-A18 is the outline of the label a and b1-b6 are the outlines of the label b, and the numbers indicate the order determined in one outline.

Next, a convex hull of the original image is obtained based on the contour obtained by the above operation. The convex hull is a minimum convex set including S for an arbitrary set of pixels S. By finding the convex hull, an area not included in S, ie, a dent of S, is detected in the convex hull. be able to.

A method for obtaining a convex hull will be described. The outline forming the convex hull is obtained by drawing all the tangents of the target graphic, and a point (contact point) at which an arbitrary tangent comes into contact with the target graphic is a pixel constituting the convex hull.

In the present invention, a plurality of reference points are provided outside the convex hull of the target graphic, and a tangent passing through the reference point is obtained for the contour of the target graphic. The position and number of the reference points are determined based on the size and shape of the pattern that can be recognized. A tangent line passing through the reference point is obtained by finding a tangent point on the contour line of the target graphic, and a straight line passing through the reference point and a point on the contour line is represented by A, and a horizontal straight line passing through the reference point is represented by a reference line as shown in FIG. , Two points at which the angle θ between A and the reference line becomes the maximum and minimum become the contact points. In FIG. 5, points B and C are contact points.

In the present embodiment, the size of the character pattern to be recognized and the position of the reference point are taken as examples, and the convex hull is obtained as follows.

The size of the character pattern is within 9 pixels vertically and 9 pixels horizontally, and the positions of the reference points are four points of the upper left, upper right, lower left and lower right edges of the image area of 10 pixels vertically and 10 pixels horizontally.

As described above, the convex hull is obtained by examining the magnitude relationship between the angle formed by the straight line passing through the reference point and the contour and the reference line.

However, when trying to mathematically determine the angle between the straight lines, an operation such as division is required, which causes an increase in processing time.

Therefore, in this embodiment, the simplification of the processing is performed by obtaining the azimuth from the reference point in advance for the entire image area, and preparing an azimuth order matrix in which the order is set by paying attention only to the magnitude relation. Figure 6 shows 10 vertical pixels and 10 horizontal pixels.
It is an azimuth order matrix at the time of a pixel, where H is a reference point, HI
Are numbered in order of increasing angle with respect to.

The contour line shown in FIG. 4 is applied to this azimuth order matrix, and the value corresponding to the position of the contour point is determined by the value of each contour point (hereinafter, referred to as the value of each contour point)
A pixel having the maximum value and a pixel having the minimum value in one contour line are determined as contact points. Since the reference points are four points A, B, C, and D in FIG. 4, the contact point for A is shown in FIG.
HIJK is calculated in correspondence with B, C and D in Fig. 4 AB
FIG. 6 shows IJKH, JKHI, and KHIJ corresponding to the CD.

In FIG. 4, "O" marks are pixels constituting the convex hull.

After detecting the pixels constituting the convex hull (hereinafter referred to as convex hull points), it is determined whether the pixel sequence of the convex hull point curve is concave or convex. Judgment can be made based on the matrix value.If the matrix value at each reference point of the pixel between the convex hull points is smaller than the matrix value at the convex hull point which is the start point and end point, it is regarded as a convex section, and if there is a larger one, Consider a concave section.

That is, between a2 and a9, a11 and a14 are concave sections, and the other sections are convex sections.

With the above operation, the contour line of the original image can be divided into the concave section and the convex section.

In this embodiment, the following are obtained as graphic features.

(1) The opening direction of the concave section The direction with reference to the reference line when the end point is viewed from the start point of the concave section (2) The area ratio between the circumscribed rectangle and the concave section (3) The positional relationship between the islands (1) The starting point of the concave section The direction in which the end point is viewed from the reference line can be easily obtained from the coordinates of the start point and the end point.

(2) is the ratio of the area of the closed region formed when the start point and the end point of the concave section are connected to the area of the rectangle circumscribing the original image, that is, (area of the concave part) / (area of the circumscribed rectangle).

(3) indicates the positional relationship between the islands when the number of clusters of black pixels (hereinafter referred to as islands) constituting the character pattern, that is, when the number of labels obtained by the above-described processing is two or more, The average of the maximum value and the minimum value of the matrix values in FIG. 6 at each reference point of the contour of the island is held as the position information of the island.

Based on the graphic features obtained as described above, the recognizing unit performs comparison with a dictionary prepared in advance, and outputs candidate characters.

Effect of the Invention As described above, according to the present invention, a stable recognition result can be obtained by extracting graphic features of a character pattern on the basis of unevenness information of an outline of an original image, and an arithmetic operation can be performed. Since the unevenness information can be obtained without any change and the position information of the island can be obtained in the process of obtaining the unevenness information, it is not necessary to separately obtain the position information of the island, so that high-speed processing can be performed.

[Brief description of the drawings]

FIG. 1 is a block diagram of an apparatus using a character recognition method according to an embodiment of the present invention, FIG. 2 is an overall flowchart of processing of the character recognition method, FIG. 3 is an image image of a character "?" FIG. 4 is a diagram showing the detection result of the contour and the convex hull, FIG. 5 is a conceptual diagram of the convex hull detection method, and FIG. 6 is a diagram showing an azimuth order matrix for obtaining the convex hull. 1 ... image input unit, 2 ... image memory, 3 ... outline detection unit, 4 ... unevenness information detection unit, 5 ... characteristic extraction unit, 6 ...
... Recognition processing unit, 7 ... Dictionary, 9 ... Character cutout unit

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int.Cl. ⁶ , DB name) G06K 9/46 G06K 9/62

Claims (1)

(57) [Claims] 1. A recognition target character pattern is stored, and a contact point with a tangent character pattern including the character pattern when the stored character pattern is viewed from the reference point and an angle of the tangent are set for each of a plurality of reference points. It is determined whether the character pattern protrudes with respect to each line segment of the convex hull composed of a plurality of determined contact points, and the character pattern of the character pattern stored from the data of the angle between the tangent and the reference line passing through the reference point. A character recognition method characterized by extracting a feature and identifying a character pattern to be recognized from the extracted feature.