JPS58223884A - Sorting processor of character pattern - Google Patents

Abstract

PURPOSE:To enable recognition of a character pattern using deformed characters, by observing the character pattern from plural coordinate axis directions and counting the directional contribution degree of a character line in the direction orthogonal to the coordinte axis as well as the resemblance between two directional contribution degrees at each position on the coordinate axis. CONSTITUTION:A binary coded handwritten character pattern is stored 1, and the position and size of the pattern are normalized by a normalizing processor 2 with the centroid and a secondary moment. A device 3 frames characters in a forward region of W mesh of a character framing frame around on the centroid in order to cope with the extreme extrusion of a character line at the place near a contour of a handwritten character. Thus the outside of a frame is removed. A smoothing processor 4 fills up or removes the recessing/projecting parts of one black point mesh at the contour part of the character line. A feature extractor 5 performs a scan from the coordinate axis of direction K and then from direction M when it crosses a character to extract features. A sorter 6 obtains matching with a dictionary table.

Description

DETAILED DESCRIPTION OF THE INVENTION (1) Description of the field to which the invention pertains The present invention relates to a classification processing device for single character patterns, and in particular, a system for classifying character patterns obtained by photoelectric conversion into binary character patterns, such as one handwritten kanji. In order to efficiently classify character objects with multiple character types and various handwriting deformations, features related to nine character line structures are extracted from near the outline of the character pattern.
The present invention relates to a character pattern classification device that classifies input character patterns.

(2) Description of conventional technology Conventionally, as a recognition device or a classification device for character patterns including kanji characters, a core line processing is performed on a single handwritten kanji character pattern to obtain geometrical shapes of characters such as edge intersections and bending points. A method is known in which character patterns are recognized by extracting the information, using this to represent the positional relationship, connection relationship, etc. of each character line, and comparing this information with a dictionary table for each character that has already been stored. However, with this method, beards and gays caused by the core wire processing, 5. Not only does it become difficult to extract the geometric shape of characters due to the generation of noise components such as dots, but also the positional relationship and connection relationship of each character line changes significantly due to the deformation of handwritten characters. However, this method requires the preparation of a huge amount of data or a huge amount of recognition logic, and has the disadvantage that an efficient recognition system cannot be realized.

In addition, character patterns of 9 handwritten kanji characters were binarized, normalized in position and size, and smoothed, and observed from two coordinate axes, the horizontal and vertical axes.

perpendicular to the coordinate axis at each position on the coordinate axis -
Count the number of character lines that cross the character part in the direction, create a feature vector pattern from this information, and match it with the feature dictionary table for each character that has already been stored.
A method for classifying character patterns, or binarization 1 Divide the character pattern that has been normalized and smoothed in position and size into rectangular areas of coarse mesh, and divide the character parts existing in each rectangular area into On the other hand, it is observed from two coordinate axes directions, a horizontal axis and a vertical axis, and the number of character lines crossing the character part in the direction perpendicular to the coordinate axes at each position on the coordinate axes is counted.

A known method is to classify a single character pattern by creating a feature vector pattern from the information and matching it with a feature dictionary table for each character that has already been stored. However, with these character pattern classification methods, although the number of character lines crossed during counting can be distinguished from the rough complexity of the character line structure depending on the type of character, the direction component of one character line and the distribution by direction component, etc. Because there is no detailed information on the differences in character line structure, there are many similar characters.

Moreover, it has the disadvantage that it cannot efficiently recognize character objects that have many handwritten deformations such as misalignment of character lines.

C Wa・ Therefore, each of the above methods can be applied to many types of characters such as handwritten kanji,
It is difficult to expect sufficient recognition for character sets with large deformations.

(3) Purpose of the Invention The present invention provides the direction, connection relationship, and parallelism of one character line for a character pattern that has been binarized, nine positions and sizes normalized, one character frame removal processing, and smoothed using conventional techniques. The character line structure information such as
The purpose of this invention is to provide a device that counts this information by scanning characters from multiple directions, and to efficiently classify handwritten character patterns including handwritten kanji with multiple types and variations.

(4) Explanation of the structure and operation of the invention A detailed explanation will be given below using the drawings.

FIG. 1 is a block diagram of an embodiment of a character pattern classification device according to the present invention. In FIG. 1, a circuit 1 is a storage circuit that first stores a binary handwritten character pattern. The device 2 inputs this binary character pattern.

For example, the present invention is a normalization processing device for the position and size of a character pattern that normalizes the position and size using the conventionally known center of gravity and second-order moment. The device 3 inputs the normalized character pattern of 9NXN mesh obtained by the device 2, and creates a character frame W centered on the center of gravity in order to deal with the extreme protrusion of character lines that occurs near the outer edge of 9 handwritten characters. This is a device that frames characters in the square area of the mesh and removes the characters that protrude outside the frame. The device 4 is a smoothing processing device which inputs the character pattern obtained by the device 3 after character frame removal and fills in or removes concavities and convexities of one mesh of black dots in the outline of one character line.

Device 5 is a feature extraction device which is the main part of the present invention, and inputs a character pattern that has been subjected to smoothing processing.

From the predetermined nine direction coordinate axes (for example, in the case of four directions, the four coordinate axes are the horizontal axis, the +45' direction axis, the vertical axis, and the -45° direction axis).

(1) When a character is scanned in a direction perpendicular to the coordinate axis and intersects one character part, the black pixels of the intersecting character part are predetermined in the 9M direction. For example, in the case of 8 directions, O;
5τ90': 135', 180', 225°, 270
°.

The tentacles are extended in 8 directions of 315 degrees), and the number of sunspots connected in each direction is counted, and the directional contribution of the sunspots (Patent application 1986-
46659) and (2) when the character part is crossed M (M≧2) times by scanning in the direction orthogonal to the # coordinate axis,
fi% (2≦m≦M)-th crossover direction contribution and (m
-1) A feature extraction device comprising: a process of determining the degree of similarity with the degree of directional contribution at the time of the second crossover.

The device 6 creates a feature table for classifying the nine character patterns based on the directional contribution value assigned by the device 5 and its similarity, and also uses the feature table created by #means. In particular, this is a classification device that performs matching with the feature dictionary table of each character that has already been stored to classify zero character patterns.

As a specific example of the device 5, four directional coordinate axes (horizontal axis.

4 coordinate axes: +45° direction axis, vertical axis, -45° direction axis (
) to give numbers 1.2.3.4 respectively).

When a character is scanned in a direction perpendicular to each coordinate axis and intersects with the character part, 8 directions (0', 45°', 90', 1as
',xso'.

The tentacles extend in eight directions of 225-270° and 315° (numbering 1.2.3.4.5.6.7.8 to each direction), the directional contribution, and the two directions. An example will be explained in which a single character pattern is classified based on the similarity of contribution.

The first method is as follows. The character pattern of the WxW mesh obtained by the device 4 was observed from four directional coordinate axes with the horizontal axis as a reference, and the A-direction coordinate axis (m
=1.2.3.4) position on (m=1.3, /=1
．． 2. ..., W, m=2.4, /=1.2. ...
lW'. Here, when scanning in the direction perpendicular to the coordinate axis (W' = 2 W) and intersecting the 1st character part for the first time, the black point that changed from a white point to a black point at the time of the intersection (at the start of scanning, the previous pixel is Directional contribution of (assumed to be a white point) 6.
is G, = (pull, target, ..., a8) information □(1) becomes 8
Represented by a dimensional vector. Here, al.

α! , ..., α8 are directional contribution components in eight directions, and the sunspot connection length 7 (C4-1, 2, ..., 8) obtained in each direction by extending tentacles in eight directions from the sunspot. As an example, it is expressed by the following formula. It is possible to apply a distance other than the Euclidean distance shown here to this a(.

In addition, if the character part and the eyes intersect with each other during the scanning,
The directional contribution of the sunspot that changed from a sunspot to a white point at the time of the intersection ~' is @m' = (61', 8fu', ..., 68') Information □
(3) is expressed as an eight-dimensional vector. Here a4'
(,=1.2...,8) is the same as equation (2), the sunspot connection length l(((=1121...,
8), it is expressed by the following equation as an example.

Furthermore, when the scanning intersects with the character part M (M≧2) times, the direction contribution at the time of intersection (2≦≦M) times ~
and 6I and the direction contribution at the (m-t)th intersection -1
- similarity between plexus and 'm-1--1+m and'
m-1*** is expressed by the following formula as an example.

'fi, + '1% + ” obtained in this way
m-weight 1. .

Be 1. For example, fc.

'm+%-1+fi for m=1 to m=
mo (1≦gn6≦M).

``fN+ to 'l+, then from m=M to m=M-
m.

By choosing a value in the +1 range.

The feature pattern 14/ obtained by scanning the position on the four-directional coordinate axes is! Aノ == (g l , -1,"'j#%01
'1! l'! 81...1%o-1+yFLOl*M
+ -M-1+...IIIM-always 6 + 1 + '
M-g +M +... 4/
It is expressed as (7).

Therefore, the feature vector of one character pattern G#iG=(ft
t+#u+medium+#tw+#*t+#st+-+#tw+
txt +fH+”・+ fvw + Pat r t
4m+"'+J'4W')-(8).

A feature table is created by combining multiple elements of the feature vector G of the character pattern expressed in this way, using the averaged value as the feature of the character pattern, and a well-known discriminant function D (G) is calculated for one character pattern. to classify.

Next, the second method is as follows. The above-mentioned W×W mesh character pattern is observed from four directional coordinate axes with the horizontal axis direction as a reference, and the four directional coordinate axes ('=1.213.
4) Upper position (at '=t+a)=1.2. ...,
When W and A=2.4, J=1.2. ...becomes IW'.

Here, when scanning in the direction perpendicular to the coordinate axis with W' = 2 is the white point) is 1'1 11, WL=(bl, bl, ba, bi)g
(9) It is expressed as a four-dimensional vector. Here, bl.

blw bm+ba are directional contribution components in four directions.

Using the sunspot connection length 1 ((t=1, 2..., 8) obtained in each direction by extending the tentacle in 8 directions from the sunspot.

As an example, it is expressed by the following formula. It is possible to apply a distance other than the Euclidean distance shown here to this b(.

In addition, when the character part intersects with the coaster during the scanning, the directional contribution bm of the black dot changed from a black dot to a white dot at the time of the intersection
is 1m = (bI', b*'+ bs'+ b4') f
nC”), where b
,z,b! '+b:+ba' is the sunspot connection length 14 ((=1, 2
, . . . , 8), it can be expressed as an example.

Furthermore, when a palindrome intersects with the character part M' (M≧2) during the scanning, the direction contribution factor at the m (2≦poison≦M) intersection, and bm and (m-1) The degree of directional contribution at the time of the second crossover -1 and the degree of similarity between bm'-t and ro.

m and r,;-,,- is one example,! : Represented by the formula.

1'"%" fFL-j H obtained in this way
stomach.

Among 'm-x, m, for example, bgyl, 'm-1, i, poison = 1 to compassion = inertia.

(1≦Inertia.≦M).

For bm + 'm-t *fi, we recommend choosing a value in the range of m = M m6 + 1 from m = M.

The characteristic pattern /ha) obtained by scanning the position on the four-directional coordinate axis is J'4/ = (b 1 + 11 river bmo +'
ll, r＠＠+”'+rfflO-1.t+tl
+bM', l1ls-s,"',radius-"l"
';-'l'l”' l ”'-fi. .

M-mo+1)47H.

Therefore, the feature vector ■ of a single character pattern is 1 (= (
Act, /hu, , -, /Atw, /Aff
it, /lhn, *-, Axw'.

/As+, /hst, ・-・, /Asw, &*
t, /A41, ..., l4w') -H.

A feature table is created using a value obtained by averaging multiple elements of the feature vector H of the character pattern expressed in this way as a feature of the character pattern, and a known discriminant function D (goods) is obtained to classify the character pattern. do.

Figure 2 shows 0 circuit 1. Device 2. Device 3. An example of the operation of device #4 is shown below. FIG. 2 (4) is an example of a binarized literary pattern that is first seen in circuit 1. Figure 2 CB
) is an example of a character pattern obtained by normalizing the position and size of the character pattern shown in FIG. 2 by device 2Vc. This is an example of a character pattern obtained by performing one-character frame removal processing on the character pattern in FIG. 2 (0 is 03 in FIG. 2) by the device 3. Figure 2 0 is Figure 2 (for the letter pattern of O,
This is an example of a character pattern that has been smoothed by the device 4.

FIG. 3 is an explanatory diagram for explaining the operation of the device 5, and is an explanatory diagram illustrating the operation of the device 5, which is observed in order to determine the directional contribution and the similarity between the two directional contributions.
The direction coordinate axes and the operating range of position J within each axis are shown. In FIG. 3, the axis 8-1 is the horizontal coordinate axis, the axis 8-2 is the +45° coordinate axis, the axis 8-3 is the vertical coordinate axis, and the axis 8-4 is the -45° coordinate axis.
Indicates the ° direction coordinate axis.

FIG. 4 is an explanatory diagram illustrating the operation of the device 5- similarly to FIG. 3.

The prisoner in Figure 4 is at position 7 on the horizontal coordinate axis. in.

This figure shows a case where scanning is performed in a direction perpendicular to the coordinate axes and the nine character parts are crossed twice. The parts of black dot 9-1 and black dot 9-3 in Fig.
It shows the black dots that changed from white dots to black dots when they crossed each other the second time, and the black dots 9-2 and 9-4 are as follows:
The graph shows the black dot that changes from the focus to a white dot when it intersects with the character portion for the first and second time during the scan. )4.1 sentence */<I'-'/(D black point 0 black point 0 in Figure 4)
Each part of black point 9-1, black point 9-2, black point 9-3, and black point 9-4 in Figure 4, which shows the direction in which the tentacles of the 5/'- extending tentacles are shown by arrows. In the first method, the directional contribution '1 + 'l + '1' + a2' obtained in the range of character intersections of 1 to m = 2 times and the similarity of the two directional contributions '12 , 1 of 'I!'
Examples are shown in Tables 1, 2, and 3, and the directional contribution b1* b* + i'+ b obtained by the second method in the range of character intersections from m = 1 to m = 2 times is shown in Table 1, Table 2, and Table 3. ! ' and the similarity of the two directional contributions rIR, rl! ' are shown in Tables 4, 5, and 6.

Table 1 Table 2 Table 4 Table 5 Table 6 Therefore, in the first method, it10= (0,902,0, O, Oro, 0.26
5.0.212 + 0.265.

0.653.0.0.0.0, 0.490.0.40B
, 0.40B10.913.

0.0.960, 0.185.0.148.0.148
．． 0,0,0゜0.0.727.0.364,0.45
5,0.364.0,0.0゜0.886) Also, in the second method, ship H = (0,902, 0,265, 0,212, 0,2
65.

0.653, 0.49G, 0.408, 0.4
08.

0.913゜0.148.0.960.0.184.0.148゜0
．． 364, 0.727.0.364, 0.455°0.
886).

The value of A- of these 1hJ6 and 1htc is the direction, connection relationship, relative positional relationship, and 2
It represents the degree of parallelism of two character lines.

Take A1c obtained from the second method as an example.

In Ih1ha, the value of bl is -1, and in km', the value of b mushroom is
Because the value of is large, in the scan from above the nuclear position in Figure 4.

First, there is a character line in the horizontal direction, and when scanning from below, it can be seen that there is a character line in the +45° direction first. The first and second methods have the advantage of being able to obtain a stable feature pattern even with character deformations that tend to occur in handwritten kanji, etc., since they are affected by the positional deviation of these character lines in the scanning direction. .

Furthermore, if the first method is used, it has the advantage that detailed geometric features such as end points, intersection points, and bending points can also be extracted. By obtaining the degree of similarity between the directional contributions of two character lines, even when the inclination of a handwritten character line as shown in Fig. 5 (4) @ (C') is changed, each change in inclination is equal to the character of the turn. Since the character is maintained, characters can be classified without being affected by these transformations.

Accordingly, the feature extraction shown in the device 5 becomes an effective means for character recognition such as handwritten Chinese characters that focuses on geometric features.

1 (5) Description of Effects As explained above, according to the present invention, nine character patterns are observed from a plurality of coordinate axis directions, and at each position on the coordinate axes, the directional contribution of the character line in the direction perpendicular to the coordinate axes and the In order to count the degree of similarity between two directional contributions, it is possible to extract character line structure information such as the direction, connection relationship, parallelism, etc. of each character line in a 9-character pattern using a simple method, as well as to calculate the change in the slope of a single character line9.
It is resistant to handwritten deformations due to misalignment, etc., and has the advantage of being able to efficiently classify character objects including handwritten Chinese characters with many character types and many handwritten character deformations.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram of an embodiment of a character pattern classification processing device according to the present invention, and FIG. 2 is a circuit diagram of the circuit 1 of FIG. Device 2.
An explanatory diagram illustrating aspects of processing by the device 3 and the device 4,
3 and 4 are explanatory diagrams for explaining the feature extracting means of the device 5 which is the main part of the present invention, and FIG. 5 is an explanatory diagram for explaining handwritten transformation of handwritten characters.
□ 1 is a storage circuit, 2 is a character pattern normalization processing device, 3 is a character pattern frame processing device, 4 is a smoothing processing device, and 5 is a character pattern feature extraction device. 6 is a character pattern classification device, and 7 is a character frame. 8-1.8-2.8-3.8-4 are coordinate axes for observing the directional contribution and the similarity between the two directional contributions. 9-1.9-2.9-3.9-4 indicates a black dot that changes from a black dot to a white dot or from a white dot to a black dot when scanning a character0 Patent applicant Mori, patent attorney representing Nippon Telegraph and Telephone Public Corporation 1) Hiroshi (A) i No. 3 (7) (8) 4 Fig. 3) No. 5 (O) Fig.

Claims (1)

[Claims] means for normalizing the position and size of the character portion of the binarized character pattern; means for framing the character pattern obtained by the means with a character frame; and the frame. a means for smoothing the character pattern obtained by the smoothing means; and a means for smoothing the character pattern obtained by the smoothing means with respect to coordinate axes in a plurality of predetermined directions, perpendicular to the coordinate axes; When a character is scanned in a direction and intersects nine character parts, tentacles are extended in multiple predetermined directions for the black pixels of the crossed character part, and the black pixel connected length is determined for each direction, and from the black pixel connected length. The first step is to obtain the 4 degree deviation in direction of the crossed black pixel.
means; when a character part is crossed multiple times during scanning in a direction perpendicular to the coordinate axis, the directional contribution of the character part at a certain time of intersection among the multiple crossovers and the previously crossed character parts; and a means for classifying character patterns using at least the output information from the first means and the second means. Character pattern classification processing device.