JPH04205287A - Device and method for extracting feature of character - Google Patents

Abstract

PURPOSE:To improve the recognition ratio of characters by measuring the average thickness of respective characters and deciding reference width in stroke end point decision concerning the respective characters. CONSTITUTION:An average character thickness measurement part 5 measures the average line width of the character to extract the feature, and a reference value 30 is decided for extracting the stroke end point based on the measured value. Therefore, the stroke end point can be decided by using the reference width corresponding to the character for each character to be recognized. Thus, the stroke end point can be extracted without fail while suppressing an influence from the thickness of lines constituting the character to the detection of the stroke end point and therefore, the recognition ratio of the characters can be improved.

Description

[Detailed Description of the Invention] [Summary] A scanning unit that scans a character to be recognized in one direction to assume an outline figure; The present invention relates to a character feature extraction device for a character recognition device, which has a stroke end point that is a writing start point and a stroke end point determination unit that determines the stroke end point, and performs character recognition in the character recognition unit based on the distribution feature of the stroke end points of the extracted recognition target character. , In order to be able to reliably extract stroke end points when extracting character features, an average character thickness measuring unit is provided to measure the average thickness of each character in the character feature extraction device as described above. At the same time, the reference width of the stroke end point determining section for each character is determined based on the measured line width.

[Industrial application field]

The present invention relates to a character feature extraction device, and more particularly, a scanning unit that scans a character to be recognized in one direction to assume an outline figure, and an outline end point of an outline character that intersects the scanning direction and has a reference width or less. It has a stroke end point that is the writing start point of a character line and a stroke end point determination unit that determines the stroke end point, and based on the distribution characteristics of the stroke end points of the extracted character to be recognized,
The present invention relates to a character feature extraction device for a character recognition device that performs character recognition in a character recognition unit.

[Conventional technology]

As shown in FIG. 5, the above-mentioned character feature extraction device includes a scanning unit 12 that scans the recognition target character 11 in one direction and assumes the outline figure of the character, and a scanning unit 12 that assumes the outline figure of the character by scanning the recognition target character 11 in one direction, and There is also a stroke end point determination unit 13 that determines the outline end point of an outline character having a width less than or equal to the reference width 13a as the stroke end point, which is the starting point of a character line. The character recognition section 14 performs character recognition using the distribution of the positions of the determined stroke end points as features extracted from the recognized characters.

These processes are first performed in the scanning section 12, for example, at the sixth
As shown in the figure, scanning is performed in a certain direction, for example from left to right in FIG. 6, and the first point that changes from white to black is captured. As a result, the characters shown in Figure 6 (1) (Figure 6 (
1) shows the katakana ``su''. ) is extracted on the left side as shown in FIG. 6(2). Then, as shown in FIG. 7, the bending points of the contour line are extracted. Here, there are two types of bending points, and the first bending point is one bent outward from the inside of the contour, and is indicated by an open circle in FIG.

Further, the second bending point is bent inward with respect to the inside of the contour, and is indicated by a black mark in FIG. 7.

As shown in FIG. 8(1), the bending points on both sides of the contour intersecting the scanning direction are the first bending points, and as shown in FIG. 8(2), the bending points are the first bending points. A point with a dimension smaller than the specified value is determined as a stroke end point, which is the starting point of a character line (stroke). That is, in the example of FIG. 8 (2), the reference value is m (dot), and the measured dimension is n.
(dot), when m≧n, that point is considered to be a stroke end point. Here, the reference value m is determined statistically from past data.

[Problem to be solved by the invention]

By the way, when recognizing handwritten characters, it is important to note that even the same character may have various shapes and line thicknesses due to differences in writing instruments and differences in the writing pressure of the writer, as shown in Figure 9. It has become.

For this reason, the conventional character feature extraction device described above may not be able to extract some of the stroke end points of handwritten characters that should be extracted, and there is a problem that the character recognition rate cannot be improved. . This occurs when the width of the stroke end point of a thick character exceeds a reference value and is not recognized as a stroke end point, as shown in FIG. 9(2).

Therefore, an object of the present invention is to provide a character feature extraction device that can reliably extract stroke end points in character feature extraction.

[Means to solve the problem]

In the present invention, means for solving the above-mentioned problems relate to a character feature extraction device, and as shown in FIG.
and a stroke end point determining unit 3 that determines the outline end point of the outline character that intersects the scanning direction and has a reference width 3a or less as the stroke end point that is the starting point of the character line.
In the character feature extraction device of the character recognition device that performs character recognition in the character recognition unit 4, the average thickness of each character is measured based on the distribution characteristics of the stroke end points of the extracted characters to be recognized. A measuring section 5 is provided, and a reference width 3a for each character is determined in the stroke end point determining section 3 based on the measured line width.

In addition, in the present invention, a second method for solving the above problems is provided.
This means relates to a character feature extraction method, in which a character to be recognized is scanned in one direction to assume a contour map, and the contour end points of the contour character that intersects the scanning direction and have a reference width or less are drawn as character lines. Determine the starting point as the stroke end point,
In a character feature extraction method in a character recognition device that performs character recognition, the average thickness of each character is measured to determine a reference width for determining stroke end points for each character based on the distribution characteristics of stroke end points of extracted characters to be recognized. It is.

(Operation) According to the present invention, the average character thickness measurement unit measures the average line width of the character for which feature extraction is to be performed, and the reference value for extracting stroke end points is determined based on the measured value. By setting a reference value according to the characteristics of each character, it is possible to reliably extract stroke end points, and as a result, the recognition rate of characters can be increased.

〔Example〕

Embodiments of the character feature extraction device according to the present invention will be described below with reference to the drawings.

2 to 4 show a character recognition device including an embodiment of the character feature extraction device according to the present invention. This character recognition device includes an image scanner 21 serving as a scanning unit that reads a document as binary image data, a character extraction processing unit 22 that extracts characters to be recognized from the read document, and an image scanner 22 that stores the read binary image data. storage memory 2
3, a character feature extraction processing unit 24 to which the present invention is applied,
The recognition processing unit 25 performs character recognition based on the extracted character features. Further, each of these parts is connected by a bus 26 and controlled by a microprocessor 27.

The character feature extraction unit 24 extracts various types of features, such as contour features including stroke end position features, multidimensional compression features, and major classification features. In this embodiment, in order to extract stroke end points as character features, the character feature extraction unit 24 includes a stroke end point determination unit 28 that extracts stroke end points, and an average thickness measurement unit that measures the average thickness of characters to be recognized. 29. Further, the stroke end point determination section 28 has a reference width storage means 30 that stores a reference width value that is a reference for determining the stroke end point, and the stroke end point determination section 28 has a reference width storage means 30 that stores a reference width value that is a reference for determining the stroke end point. The stroke end points are determined by storing the average thickness as its reference width.

Here, the average thickness measuring section determines the average thickness of the characters by the method shown in FIGS. 3 and 4.

That is, first, in order to measure the average width of the vertical strokes constituting the character, the character is scanned in the horizontal direction (X direction) to create a histogram (histogram (1)) of the average width of one stroke. At this time, if there are two vertical strokes 31a and 31b while scanning the position shown by line A in FIG. 3, the average width of the two vertical strokes is recorded in a histogram.

Further, when the horizontal stroke 32 substantially parallel to the scanning line is crossed while scanning the position shown by line B, the width thereof is captured to be large, and the value of the histogram is also large.

Next, regarding this created histogram, we will examine the vertical direction (y
direction) to create a histogram (histogram (2)) similar to histogram (1), and in the histogram, the value at the point where the downward trend is the sharpest is set as the average thickness of the vertical stroke W□.

Also, for horizontal strokes, in the same way as measuring the average value of vertical strokes, as shown in Fig. A histogram (3)) is created, a histogram (histogram (4)) is created for this histogram, and the portion of the histogram where the downward trend is greatest is determined as the average width W2 of the horizontal stroke.

Then, the average value of the obtained vertical stroke average thickness W□ and horizontal stroke average thickness W2 is stored in the reference width storage means 30 as the average character thickness W, and the stroke end point is determined by this value. I take it as a thing.

Therefore, according to this embodiment, since the stroke end point can be determined for each character to be recognized using the reference width suitable for that character, the influence of the stroke end point detection by the thickness of the lines that make up the character By suppressing the stroke end points, it is possible to reliably extract the stroke end points, thereby increasing the character recognition rate.

〔Effect of the invention〕

As explained above, according to the present invention, when extracting character features in a character recognition device, the average thickness of each character is measured to determine the reference width for determining the stroke end point of each character, so that recognition is possible. Since the stroke end points can be determined for each character using the reference width that matches that character, stroke end points can be reliably extracted without affecting stroke end point detection due to the thickness of the lines that make up the character. This has the effect of increasing the character recognition rate.

[Brief explanation of the drawing]

FIG. 1 is a principle diagram of the present invention, FIG. 2 is a block diagram showing a character recognition device to which a character feature extraction device according to an embodiment of the present invention is applied, and FIGS. 3 and 4 are character features according to an embodiment. A diagram showing the operation of the average thickness measuring section of the extraction device, FIG. 5 is a block diagram showing a conventional character feature extraction device, and FIG. 6 (1) (2)
) is a diagram showing contour detection in the character feature extraction device,
Figure 7 shows two types of bending points, Figure 8 (1) (2)
is a diagram showing the conditions for determining the stroke end point, Figure 9 (1) (
2) is a diagram showing differences in the thickness of characters to be recognized. 1...Character to be recognized 2...Scanning section 3...Stroke end point determining section 3a...Reference width 4...Character recognition section 5...Average character thickness measuring section

Claims (1)

[Claims] 1) A scanning unit (2) that scans the character to be recognized (1) in one direction to assume a contour figure, and a contour that intersects the scanning direction and has a reference width (3a) or less; It has a stroke end point determination unit (3) that determines the outline end point of the character as the stroke end point that is the starting point of the character line, and the character recognition unit (4) uses the distribution characteristics of the stroke end points of the extracted character to be recognized. In a character feature extraction device of a character recognition device that performs character recognition, an average character thickness measurement unit (5) that measures the average thickness of each character.
In addition, based on the measured line width, the stroke end point determination unit (3) determines the reference width (3) for each character.
A character feature extraction device characterized by determining (a). 2) Scan the character to be recognized in one direction and imagine a contour map,
Regarding the contour, the contour end point of the contour character that intersects the scanning direction and is less than the reference width is determined as the stroke end point, which is the starting point of the character line, and character recognition is performed based on the distribution characteristics of the stroke end points of the extracted character to be recognized. A character feature extraction method in a character recognition device, characterized in that the average thickness of each character is measured to determine a reference width for determining a stroke end point for each character.