JPH10143610A - Character recognizing device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To properly recognize continuously handwritten characters. SOLUTION: The coordinate columns of handwritten input characters are successively fetched (P1) a re-sampling processing (P2) and a normalizing processing (P3) are operated, a character pattern is approximated with a cuneiform character pattern constituted of only vertical lines directed from an upper part to the other directions and horizontal lines directed from the left direction to the right direction(P4), a matching processing is operated between the cuneiform character pattern and a dictionary pattern obtained by approximating the cuneiform character pattern with a standard character pattern (P5), and a character corresponding to the input pattern is recognized.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition device for recognizing handwritten input characters.

[0002]

2. Description of the Related Art As a character recognition device, there is known a portable terminal device called a PDA which accepts input of characters by handwriting input.

In such an apparatus, in order to recognize handwritten characters, a dictionary pattern prepared for each character is checked for the degree of matching with the handwritten characters, and a character corresponding to the dictionary pattern having a large degree is matched with a handwritten character. A matching method for recognizing a character as a matching character is widely used.

[0004] On the other hand, it is desired that such a character recognizing device recognize characters that are continuously written and characters that are handwritten in an incorrect stroke order more correctly.

Therefore, the following techniques (1) to (3) have conventionally been used as techniques for recognizing continuously written characters or characters handwritten in an incorrect stroke order.

(1) For each character, a plurality of dictionary patterns including continuous writing, different stroke order, and deformation are registered in a dictionary, and matching between each dictionary pattern and handwritten characters is performed.

(2) From one dictionary pattern for each character, a dictionary pattern of continuous writing and stroke order is generated according to a predetermined cool, and matching between each dictionary pattern and handwritten characters is performed.

(3) After associating each dictionary pattern with an image of the input pattern, if there is a surplus image in the dictionary pattern, it is considered that the input pattern is continuous writing, and the surplus image of the dictionary pattern is replaced with another image. Then, a continuous writing dictionary pattern is generated, and matching with handwritten characters is performed.

Here, FIG. 17 shows an outline of the technique y of (2).

In this technique, as shown in the figure, a dictionary pattern that is approximated by a plurality of points for equally dividing an image for each image is prepared in a dictionary in advance, and in the process of recognizing handwritten characters, A dictionary pattern for continuation characters is generated, and the input character pattern is matched with an approximate pattern approximated by a plurality of points for each image (continuous trajectory).

[0011]

SUMMARY OF THE INVENTION Conventional (1) and (2)
According to the technique of (1), the number of times of matching between the dictionary pattern and the input pattern increases, so that the processing amount increases. In addition, in order to reduce the effect of handwritten character pattern deformation due to continuous writing, when using a large amount of processing such as DP matching to match the dictionary pattern and handwritten character pattern, the number of times of such matching is reduced. The increase in the processing amount due to the increase becomes enormous. Furthermore, in the technique (1), since it is necessary to hold a large number of dictionary patterns in advance, a memory having a large storage capacity is required to hold the dictionary patterns.

For this reason, when using these techniques, in practice, restrictions are often provided so as not to use a dictionary pattern corresponding to excessive continuous writing or incorrect stroke order.
In this case, handwritten characters that can be correctly recognized are limited.

Further, in the technique (3), it is difficult to associate a dictionary pattern with an image of an input pattern, and correct recognition cannot be performed for handwritten characters that are continuously written so that the number of strokes is reduced to half or less.

Therefore, the present invention provides a character recognition device that can more correctly recognize characters that are continuously written or characters that are input in an incorrect stroke order and that include deformation, with a processing amount that is not so large. The purpose is to provide.

[0015]

To achieve the above object,
The present invention is a character recognition device for recognizing a handwritten character, comprising a means for inputting a handwritten character pattern of a character to be recognized, and a pattern including a vertical line and a horizontal line for each character. A storage unit storing a dictionary in which a dictionary pattern is registered, a character pattern is compared with each dictionary pattern registered in the dictionary, and a character corresponding to a dictionary pattern similar to the character pattern is corresponded to the character pattern. And a recognizing means for recognizing the character as a character to be recognized.

That is, the present invention has been made based on a new finding that the characteristics of characters are formed on the basis of vertical lines and horizontal lines, and the effect of continuous writing often appears as oblique lines. In this way, the effect of continuous writing is unlikely to appear, and each character has its own unique vertical line,
According to the present invention in which a horizontal line pattern is used as a dictionary pattern, even a continuously written character or a character that is input in an incorrect stroke order and includes a deformation can be more correctly recognized by processing with a modest amount of processing. can do.

[0017]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the character recognition device according to the present invention will be described below.

FIG. 1 shows a configuration of a character recognition device according to the present embodiment.

In the figure, the liquid crystal tablet 1 takes in a pattern handwritten by a user thereon as a time series of coordinate points and passes it to a pen manager 2. The pen manager 2 receives the time-series coordinate point sequence from the liquid crystal tablet 1, and if the coordinate point sequence recognizes the character according to the attribute given to the area on the liquid crystal tablet 1 defined by the application 3. If it is in the area to be processed, it is passed to the preprocessing unit 4.
Note that the application 3 divides the area on the liquid crystal tablet 1 into several parts and determines area attributes. For example, a part of the screen is defined as a character input area.

The pre-processing unit 4 receives a time-series coordinate point sequence from the pen manager 2, and
The late part resamples the dense coordinate point sequence to make the density of the coordinate point sequence constant, and further normalizes the position and size of the character pattern represented by the coordinate point sequence. For example, the normalization is performed by normalizing the position by moving the pattern in parallel so that the center of gravity of the pattern is superimposed on the origin, so that the average value of the distance from the center of gravity to each coordinate point becomes a constant value. This is done by enlarging or reducing. Then, the coordinate point sequence after this processing is passed to the wedge-shaped character approximation unit 5.

Here, the wedge-shaped character approximating unit 5 includes a vertical / horizontal element decomposing unit 5-1, a characteristic element extracting unit 5-2, and an element rearranging unit 5-3. The wedge-shaped character approximation unit 5 pays attention to the stroke direction of the character pattern, and extracts those having the stroke direction of downward and right as non-continuously written portions. Character pattern represented by the coordinate point sequence passed from 4
To a non-continuous character pattern. The name of the wedge-shaped character approximating unit 5 is because the pattern generated by such approximation matches the characteristic of an ancient wedge-shaped character composed of only a downward or rightward stroke. It has been adopted.

Now, the vertical and horizontal element decomposition section 5-1 of the wedge-shaped approximation section 5
Uses the part that can be approximated to the straight line included in the character pattern as an element, classifies this element according to the four stroke directions (up, down, left, and right), and extracts a sequence of classified elements as feature element extraction Hand over to Part 5-2.

The characteristic element extraction unit 5-2 includes a vertical and horizontal element decomposition unit 5-1.
, And identifies whether each element corresponds to the part where the continuous writing character pattern is continuously written or the element corresponding to the part where the continuous writing is not performed. Element. In this case, a portion where the stroke direction of the character pattern portion corresponding to each element is not continuously written down or right is a feature element, and a portion where the upper or left one is continuously written is a feature portion. Then, the element sequence and the information on the identified characteristic element sequence are passed to the element rearranging unit 5-3 as characteristic element information.

The element rearranging section 5-3 can determine whether each element of the dictionary pattern and each element of the input pattern are appropriate for the element of the character pattern written in an incorrect stroke order in accordance with the position of each element. In order to be able to attach the feature elements, the feature element strings received in advance from the feature element extraction unit 5-2 are arranged in order, with the feature elements whose stroke direction is The feature elements arranged in the rightward direction are arranged in order from the leftmost feature element, and the order is used as feature element order information. Then, the element sequence, the characteristic element information, and the characteristic element order information are passed to the matching management section 7-1 of the matching processing section 7 as input patterns to be subjected to the matching processing.

Next, the dictionary 6 stores, for each character, the characteristic element information and the characteristic element order obtained by processing the coordinate point sequence of the standard character pattern by the preprocessing unit 4 and the wedge-shaped approximation unit 5. The information is associated with the corresponding character codes and the dictionary pattern
And keep it as At this time, if you write without continuing the character pattern, most of the element sequence will be a characteristic element,
The dictionary capacity can be reduced. However, the feature element order may be obtained by the element rearranging unit 5-3 each time the recognition processing is performed. The input / dictionary matching unit 7 includes a matching management unit 7-1, a distance calculation unit 7-2, a dictionary-only feature element search unit 7-3, an input-only feature element search unit 7-4, and a recognition result output unit 7. Consists of -5.

The matching management section 7-1 reads out the dictionary patterns one by one from the dictionary 6. The distance calculation unit 7-2, the dictionary-only feature element search unit 7-3, the input-only feature element search unit 7-4,
Calculate the distance value between the feature element sequence of the called dictionary pattern and the feature element sequence of the input pattern, and determine the ID of the dictionary pattern or the character code corresponding to the dictionary pattern. Dictionary puta
The calculated distance value is passed to the recognition result output unit 7-5.

More specifically, the distance calculation unit 7-2 outputs
Feature element row in which the bottom is downward (direction from top to bottom),
The characteristic element sequence of the dictionary pattern and the characteristic element sequence of the input pattern are compared for the characteristic element sequence in which the stroke is rightward (from left to right), and the position (element rearranging section) is compared. (The order after the sorting of 7-1) is associated with each other, and the distance value between the associated feature elements is determined by the matching management unit.
Only the dictionary is passed to the feature element search unit 7-3 via 7-1. However, for the feature element of the dictionary pattern and the feature element of the input pattern for which there is no corresponding feature element, a value corresponding to the length of the feature element is given as a distance value. result,
Distance values are given to all of the feature elements of the input pattern and to the feature elements of the dictionary pattern that are not associated with the feature elements of the input pattern.

When the sum of the distance values (the distance value between the input pattern and the dictionary pattern) becomes larger than a predetermined value, the processing for this dictionary pattern is terminated, and the fact is checked by the matching management unit 7-. Notification may be made to 1 so that processing for the next dictionary pattern is performed.

Next, the dictionary-only feature element search unit 7-3
As a result of the processing of the distance calculation unit 7-2, for a feature element of the dictionary pattern that does not have a corresponding feature element of the input pattern,
(1) First, in the element sequence classified by the vertical / horizontal element decomposition unit 5-1, whether or not there is an element other than the characteristic element of the input pattern corresponding to the characteristic element of the dictionary pattern Is searched, and if it exists, it is associated with the characteristic element of the dictionary pattern, the distance value between the element associated with the characteristic element of the dictionary pattern is obtained, and the characteristic element of the dictionary pattern is determined. Is changed to the calculated distance value. This is because when the degree of continuous writing of the input character pattern is high, the part of the input character pattern corresponding to the characteristic element of the dictionary pattern is not extracted as the characteristic element of the input character pattern. It takes into account that there is. On the other hand, if there is no corresponding element (2), the distance value is used as it is according to the length of the characteristic element in the dictionary pattern. Here, when giving the distance value according to the length of the characteristic element, when writing a character, a large element is hard to be omitted, but a small element is easy to be omitted, so that there is no corresponding element. This is because, when there is a large feature element, it is highly likely that this dictionary pattern is not related to the character corresponding to the input pattern. By the way, when the above processing is completed, the dictionary-only feature element search unit 7-3 compares the correspondence calculated by the distance calculation unit 7-2 and the dictionary-only feature element search unit 7-3 with the input pattern.
The distance values of the dictionary pattern feature elements that were not associated with all of the feature elements of the input pattern and the feature elements of the input pattern were input through the matching management unit 7-1.
Pass to.

The input-only feature element searching unit 7-4 searches the input pattern feature elements which are not associated with the dictionary pattern feature elements for the feature elements that are considered to correspond to the part of continuous writing. Gives a small distance value. This is the input pattern
When the feature elements of the application are viewed in the order indicated by the feature element order information,
The feature elements before and after the feature element that could not be matched with the feature element of the dictionary pattern can be matched with the feature element of the dictionary pattern, and at the position where the feature element of the dictionary pattern is connected, If there is a characteristic element of the input pattern that is not associated with the characteristic element of the dictionary pattern, it is considered that the characteristic element corresponds to the part of continuous writing, and the distance value of the characteristic element is set. The distance is changed to a small value, and in other cases, the distance is set to a value corresponding to the length of the feature element. And when the above process is completed,
The sum of the distance values of all of the feature elements of the input pattern and the feature elements of the dictionary pattern not associated with the feature elements of the input pattern is passed to the matching management unit 7-1.

The matching management unit 7-1 transfers the received sum of the distance values and the ID or character code of the dictionary pattern for which the distance values have been obtained to the recognition result output unit 7-5.

The recognition result output unit 7-5 includes a matching management unit 7
When the sum of the distance value and the ID or character code of the dictionary pattern is received from -1, the one having the smaller sum of the distance values is selected and arranged in the order of the sum to obtain the recognition result. Once the recognition result is passed to the detailed identification unit 8 and the returned recognition result is obtained again, the ID or the character code of the dictionary pattern is input to the liquid crystal tablet 1 via the pen manager 2. The character corresponding to the input character pattern is displayed, the user selects the character corresponding to the input character pattern, and the selected character code is passed to the application 3.

Here, the detailed identification section 8 is composed of a vertical and horizontal element decomposition section 5.
The order of recognition results is switched based on the characteristics of the input pattern that are lost in the element sequence divided by -1. The detailed identification unit 8 outputs a dictionary pattern from the recognition result output unit 7-5.
After obtaining the ID or character code of the input character pattern, other characteristics of the input character pattern are matched with the dictionary of the other characteristics, and if necessary, the recognition result order is changed, and the result is replaced. Is passed to the recognition result output unit 7-5. For example,
If the recognition result includes a plurality of characters that are approximated to have the same shape due to the loss of curves and corners by the processing of the wedge-shaped character approximation unit 5, the detailed identification Using a dictionary stored for each character with features such as characters and corners, the degree of matching between each character and the input character pattern is examined. Change the recognition result order of multiple characters.

Hereinafter, the operation of the character recognition apparatus according to the present embodiment will be described in detail.

First, FIG. 2 summarizes the processing flow of the character recognition device according to the present embodiment described above.

As shown in the figure, when the character recognition apparatus receives a character pattern handwritten by the user with the liquid crystal tablet 1 as a time-series coordinate point sequence in step P1, as described above, the preprocessing is performed. The spatial coordinate point sequence density forming the stroke is made constant by the part 4 (P2), and the position and size are normalized (P3). Next, a cuneiform character approximation unit 5 performs a cuneiform character approximation process focusing on the stroke direction of the input character pattern (P4).
The matching process is performed by focusing on the stroke direction and the arrangement of the feature elements between the pattern and the dictionary pattern as features (P5), and the detail recognition unit 8 performs the detail recognition process on other features.
(P6) The result is output from the recognition result output unit 7-5 (P7).

Hereinafter, the details of each processing shown in FIG. 2 will be described.

First, the wedge-shaped character approximation process performed by the wedge-shaped character approximation unit 5 in step P4 of FIG. 2 will be described.

FIG. 3 shows a processing procedure of this wedge-shaped character approximation processing.

As shown in the figure, in this process, first, in step P4-1, the vertical / horizontal element decomposition section 5-1 decomposes the input character pattern into vertical and horizontal elements. That is, the input character pattern is approximated by vertical and horizontal lines.

Specifically, this processing is performed by the processing shown in FIG.

That is, first, at step p4-1-1, local M in the x-axis and y-axis directions of the input character pattern
Extract points that take IN / MAX values. Local MIN / M in x direction
As points for taking the AX value, a change point in the x-axis direction in the stroke direction, and a start point and an end point of the stroke are extracted. In addition, as points that take local MIN / MAX values in the y-axis direction, a change point in the y-axis direction in the stroke direction, and a start point and an end point of the stroke are extracted.
In steps p4-1-2 and p4-1-3, all MINs in the x-axis direction
For a point at / MAX value, create a vertical line through that point. In p4-1-4 and p4-1-5, MIN / MAX in all y-axis directions
For a value point, create a vertical line through that point.

Next, in steps p4-1-6 and p4-1-7,
For all points, examine the relationship with the vertical and horizontal lines of the next point on the stroke in the order according to the input order,
If they intersect, that is, if one is a vertical line and the other is a horizontal line, an intersection is determined in step p4-1-8.

If the horizontal lines are adjacent, the step
Performs S-shape vertical and horizontal element decomposition (horizontal) of p4-1-9, and if vertical lines, the S-shape vertical and horizontal element decomposition of step p4-1-10 (vertical)
And find the intersection. However, the start point and end point of the stroke are local MIN / MA in both y-axis and x-axis directions.
Since it is a point that takes the X value, it becomes the target of step p4-1-8 and p4-1-9 or p4-1-10. Therefore, for the starting point and ending point of the stroke, see step p4-1-8
And only one of p4-1-9 or p4-1-10 is performed.

Now, in the S-shaped vertical and horizontal element decomposition (horizontal) in step p4-1-9, as shown in FIG. 5A, in step P4-1-9-1, the adjacent points Pi-1, Pi are determined. Generate a horizontal line L passing through the average y value.
Then, in step P4-1-9-2, an intersection Q between the horizontal line L and the original stroke between the points Pi-1 and Pi is obtained. And step P4-1
A vertical line passing through the intersection Q is obtained at -9-3, and this vertical line and the point Pi-1
At the intersection with the horizontal line passing through the point Pi.

S-shaped vertical and horizontal element decomposition in step p4-1-10 (vertical)
5 is realized by a process in which the vertical and horizontal relationships of the process of FIG. 5A are reversed.

The S-shaped vertical and horizontal element decomposition (horizontal) in step p4-1-9 may be performed as shown in FIG. 5 (b). That is, a vertical line passing through the average x value of the adjacent points Pi-1 and Pi is obtained, and the intersection of this vertical line and the horizontal line passing through the point Pi-1 is obtained in the order of the intersection with the horizontal line passing through the point Pi. You may do so. Also, the S-shaped vertical and horizontal element decomposition (vertical) in step p4-1-10 may be realized by processing in which the vertical and horizontal relationships of the processing in FIG. 5B are reversed.

Further, as shown in FIG.
Obtain the vertical line of the center of gravity (passing the average x value) of the stroke between Pi-1 and Pi, and make the intersection of this vertical line and the horizontal line passing through point Pi-1 with the horizontal line passing through point Pi. May be obtained in this order. Also,
The S-shaped vertical / horizontal element decomposition (vertical) in step p4-1-10 may also be realized by processing in which the vertical / horizontal relationship of the processing in FIG. 5 (c) is reversed.

In each of the processes shown in FIGS. 5A, 5B and 5C, a new vertical line is generated, but its position is slightly different. In the example shown in the figure, in the case of FIG.
The shape becomes close to

Now, returning to FIG. 4, the last step p4-1-11
Then, a line segment sequence is obtained in which the intersections obtained in the processing of steps p4-1-7 to p4-1-10 are connected in the order in which the intersections were obtained. This line segment sequence is an element sequence in which the input character pattern is approximated by vertical and horizontal lines. The element sequence obtained in this way is as follows.For the vertical line, the vertical line table I.TB shown in FIG. 9A, and for the horizontal line, the horizontal line table I.YB shown in FIG. , The coordinates of the start point and end point of each line are
-Register in the direction of the arrow (the order in which the intersections were found).
However, instead of the start point and the end point, the length of the line and the coordinates of a representative point (for example, the start point) on the line may be registered.

The vertical and horizontal element decomposition processing in FIG. 4 may be performed as shown in FIG.

In this process, in step P4-1'-1, the flash
After decomposing the stroke direction into four directions (up, down, left and right) and extracting the point where the direction fluctuates and the start and end points,
The point that has the average x value of the two points at both ends of the part whose vertical direction is the vertical direction, the point that has the y value of one of the two points at both ends, and the average x value of the two points at both ends
Obtains a vertical line segment connecting the value and the point with the other y value of the two points at both ends. Also, in Step P4-1'-5, the point having the average y value of the two points at both ends of the portion where the stroke direction is the left and right direction, the x value of one of the two points at the both ends, and the two points at the both ends. Obtain a vertical line segment connecting the average y value and the point with the other x value of the two ends.

The vertical and horizontal element decomposition processing at step p-4-1 in FIG. 3 has been described above.

In the feature element extraction process of step p-4-2 in FIG. 3 to be executed next, in the feature element extraction unit 5-2, of the elements whose input character patterns are approximated by vertical and horizontal lines, The direction in which the direction from the start point to the end point is downward and the direction in which the direction is rightward are extracted.

More specifically, as shown in FIG. 7, first, in step P4-2-1, the original stroke direction of the element row consisting of the vertical line and the horizontal line is the left direction and the downward direction. Is extracted as a feature element. Then, in step P4-2-2, the lines in the same direction (vertical lines or horizontal lines) connected with a short orthogonal line interposed between the extracted feature elements as one line Thus, instead of the lines in the same direction that are connected by being moved in parallel and connected with the short line interposed therebetween, the connected lines are featured. In addition, instead of the same line connecting the short lines registered in the vertical line table I.TB and the horizontal line table I.YB in the same direction, a connected line is registered. Then, in Step P4-2-3, a short line is used as a feature element and a vertical line
-Remove from I.TB and I.YB. Then, the obtained characteristic elements are shown in the vertical characteristic element table i.kt in FIG. 9B for the vertical line characteristic elements, and in FIG. 9 for the horizontal line characteristic elements.
Register in the horizontal characteristic element table I.KY in (e). The registration is performed by registering the numbers given in the vertical line table I.TB and the horizontal line table I.YB in the elements extracted as the characteristic elements.

Although the processing of steps P4-2-2 and P4-2-3 can reduce the processing time for matching between the input and the dictionary, which will be described in detail later, this processing need not be performed.

Incidentally, the characteristic element extraction processing in step p-4-2 may be performed as shown in FIG.

That is, first, in step P4-2'-1, portions of the input character pattern where the stroke direction is rightward and downward are extracted. For example, in the example in FIG. 8, portions l1, l3, l4,... Indicated by thick lines are extracted.

The following operation is performed for all the original strokes extracted in step P4-2'-2. First, if it is rightward and downward in P4-2'-3, the horizontal and vertical lines obtained for the input character pattern portion including this stroke are extracted as characteristic elements. In the example in the figure,
-The horizontal line Y2 and the vertical line T3 are extracted as feature elements for the mark l3.

Next, in step P4-2'-4, the horizontal line obtained for the portion of the input character pattern including the rightward stroke is extracted as a characteristic element. In step 5, the horizontal line and the horizontal line obtained for the portion of the input character pattern including the downward stroke are extracted as characteristic elements. In the example,
Vertical lines T1 and T3 are extracted as feature elements for the strokes l1 and l4. And finally, steps P4-2'-6, P4-2 '
At -7, the same processing as the processing at P4-2-2 and P4-2-3 in FIG. 7 is performed.

Returning to FIG. 3, in the next step P4-3, the element rearranging section 5-3 performs element rearranging processing. That is, the extracted vertical line feature elements are arranged in order from the top one in accordance with the position, and the horizontal line feature elements are arranged in order from the left side in accordance with the position. Then, the information of the arrangement order is shown in the vertical feature element order table I.IKT in FIG. 9C for the vertical line feature element, and the horizontal feature element order in FIG. 9F for the horizontal line feature element. Register for Table I.IKY. Registration is based on the elements extracted as feature elements,
This is performed by registering the numbers given in the vertical line table I.TB and the horizontal line table I.YB in the order of arrangement. In this way, by rearranging according to the position, the character pattern input in an incorrect stroke order can
Matching makes it possible to properly associate feature elements.

The wedge-shaped character approximation processing in step p4 in FIG. 2 has been described above.

Next, the matching process in step P5 in FIG. 2 will be described.

In this process, the dictionary pattern shown in FIG. 10 is replaced with the vertical line table I.TB, horizontal line table I.TY,
The input pattern formed from the vertical characteristic element table I.KT, the horizontal characteristic element table I.KY, the vertical characteristic element order table I.IKT, and the horizontal characteristic element order table I.IKY Used for matching.

The dictionary patterns are determined in advance by using the vertical / horizontal line table, vertical / horizontal feature element table, and vertical / horizontal feature element table shown in FIG. Te
Bull. Here, a dictionary pattern was obtained using a character pattern that is not written continuously so that it can be easily decomposed into feature elements. The vertical line table D.TB / horizontal table D.TY in the figure is
Corresponds to the vertical line table I.TB / horizontal table I.TY shown in (a) and (d), and the vertical characteristic element table D.KT / horizontal characteristic element table DK
Y corresponds to the vertical characteristic element table I.KT / horizontal characteristic element table I.KY shown in FIGS. 9 (b) and 9 (e), and the vertical characteristic element order table D.IKT /
The horizontal characteristic element order table D.IKY corresponds to the vertical characteristic element order table I.IKT / horizontal element order table I.IKY shown in FIGS. 9 (c) and 9 (f).

In the matching process, as shown in FIG. 11, first, in step P5-2, the distance calculation unit 7-2 uses the distance calculation unit 7-2 to determine the distance between the characteristic element that minimizes the distance value between the input pattern and the dictionary pattern. And a distance value are calculated.

As shown in FIG. 12, this correspondence and the calculation of the distance value are performed for each of the vertical line characteristic element (step p5-21) and the horizontal line characteristic element (p5-2-2).

The association of the characteristic elements with respect to the lines in each direction and the calculation of the distance value between the input pattern and the dictionary pattern are performed using DP.
(Direct Patern) Performed by processing using matching.

FIG. 13 shows a process for a vertical characteristic element.

This processing is performed by setting I as the total number of vertical line feature elements of the input pattern and J as the total number of vertical line feature elements of the dictionary pattern. The coordinates of the start point of the i-th feature element of the vertical line of the input pattern are (Pis.x, Pis.y), and the coordinates of the end point are
(Pie.x, Pie.y), the coordinates of the start point of the jth feature element of the dictionary pattern are (Pjs.x, Pjs.y), and the coordinates of the end point are (Pje.x,
Pje.y), d (i, j) = | Pis.x-Pjs.x | + | Pis.y-Pjs.y | + | P
ie.x-Pje.x | + | Pie.y-Pje.y | d (*, j) = | Pjs.x-Pje.x | + | Pjs.y-Pje.y | d (i, *) = | Pis.x-Pie.x | + | Pis.y-Pie.y | Here, d (i, j) is a distance value between the feature element i and the feature element j, and represents a displacement between the start point and the end point of the two. The smaller the value, the i-th of the input pattern And the j-th feature element of the dictionary pattern are similar. d (*, j) represents the size of the j-th feature element of the dictionary pattern, and d (i, *) represents the size of the i-th feature element of the input pattern. Here, in the present embodiment, d (*, j) represents the degree of the likelihood that there is no feature element corresponding to the j-th feature element in the dictionary pattern, that is, the distance value between nothing and the feature element. And d (i, *) is used to represent the degree of the likelihood that there is no feature element corresponding to the i-th feature element in the input pattern, that is, the distance value between nothing and the feature element. Therefore, as d (i, *) and d (*, j), an appropriate function other than the function representing the size of the feature element may be used, or may be an appropriate constant value. Also, d (i, j) is also | Pis.x-Pjs.x | + | Pis.y-Pjs.y
| + | Pie.x-Pje.x | + | Any function other than | Pie.y-Pje.y | that represents the similarity between feature element i and feature element j may be used.

In the processing shown in FIG. 13, first, g (0,
0) = 0 (step 1301), and for each i from i = 1 to i = I, Δ (i 0) = (-1,0), g (i, 0) = g (i-1, 0) + d (i, *)
(Steps 1302 to 1305), for each j between j = 1 to j = J, Δ (0, j) = (0, -1), g (0, j) = g (0, j-1) + d (*, j) is obtained (steps 1306 to 1309).

Next, while changing i from i = 1 to i = I, while changing j for each i from j = 1 to j = j, j is changed from i = 1 to i = I. i and all j from j = 1 to j = J
For the combination of (i, j), Δ (i, j) and g (i, j) are obtained (steps 1310 to 1322).

In this processing, for the set (i, j),
First, g1 = d (*, j) + g (i-1, j) and g2 = d (i, j) + g (i-1, j-1)
And g3 = d (i, *) + g (i, j-1) (from steps 1312 to 131
4) Let g (i, j) take the smallest value among g1, g2, and g3 (steps 1315 to 1318). When g1 is selected as g (i, j), Δ (i, j) = (-1,0), and when g2 is selected as g (i, j), Δ (i, j) = (-1, -1), g3 is g (i, j)
Is set as Δ (i, j) = (0,1) (steps 1315 to 1318).

When Δ (i, j) and g (i, j) are obtained for all combinations of (i, j), the finally obtained g (I, J) is input. Distance value D between pattern and dictionary pattern
(i, j) (step 1323).

Finally, the characteristic element of the input pattern is associated with the characteristic element of the dictionary pattern (step 1324). This association is performed as follows.

First, P (0) = (I, J), and P (1) = ΔP (n) + ΔP
(0). Hereinafter, P (k) until P (*) = (0, 0)
= P (k-1) + ΔP (k-1). Here, P (n) = (nx, n
y), ΔP (n) = (Δnx, Δny), P (n) + ΔP (n) = (nx + Δ
nx, ny + Δny).

When P (n) is obtained from P (0) in this way, ΔP (k) is checked for each k between k = 0 and n.
If ΔP (k) is (-1, -1), the characteristic element ik of the input pattern is associated with the characteristic element jk of the dictionary pattern as P (k) = (ik, jk). If ΔP (k) is (-1,0), P
Assuming that (k) = (ik, jk), there is no dictionary pattern feature element corresponding to the input pattern feature element ik, and ΔP (k)
Is (0, -1), it is assumed that P (k) = (ik, jk) and there is no input pattern feature element corresponding to the dictionary pattern feature element jk.

Here, the distance value D (i, j) between the input pattern and the dictionary pattern obtained earlier is associated with the feature elements obtained for P (0) to P (n) or associated with each other. It is the sum of the distance values of the missing feature elements for nothing. For example, P (k) = (i
k, jk), and if the feature element ik is associated with the feature element jk, the distance value between the feature element ik and the feature element jk is d
(ik, jk). Also, if P (k) = (ik, jk) and the characteristic element ik of the input pattern is not associated with the characteristic element of the dictionary pattern, the distance value of the characteristic element ik is d (i
k, *), P (k) = (ik, jk), and if the feature element of the input pattern and the feature element jk of the dictionary pattern were not associated, the feature element ik Is d (*, jk).

The details of the processing of the vertical line characteristic element in step 5-2-1 in FIG. 12 have been described above.
The processing for the characteristic element of the horizontal line is performed in the same manner.

However, the processing of the vertical line characteristic element in step 5-2-1 in FIG. 12 and the processing of the horizontal line characteristic element in step 5-2-2 are the processing using the DP matching described above. Instead, for example, the distance may be obtained for all combinations of the characteristic element of the input pattern and the characteristic element of the dictionary pattern, and the correspondence that minimizes the distance value may be obtained. Also, at this time, in order to reduce the processing amount, instead of obtaining the distances for all the combinations, the combinations are limited to some extent in advance by the correspondence between the length and the position of the characteristic element, and then the processing is performed. It may be performed.

Now, the diagrams shown in FIGS.
As shown in FIG. 14, the correspondence between the characteristic elements of the input pattern and the characteristic elements of the dictionary pattern is obtained by the distance calculation processing in step 5-2 of FIG.

FIG. 14 shows that the distance between the input character pattern shown and the character pattern represented by the dictionary pattern is calculated by the distance calculation process. The correspondence between the table I.IKT () and the vertical characteristic element order table D.IKT () of the dictionary pattern is performed as shown by the thick line, and the characteristic element of the horizontal line is For example,
I.IKY () and dictionary pattern
This indicates that association is performed between the horizontal feature element order and the position order D.IKY () as indicated by a line.

When the correspondence between the characteristic elements and the distance value between the input pattern and the dictionary pattern are obtained in this way,
Next, only the dictionary in step P5-3 in FIG. 11 is searched for characteristic elements.

This processing will be described with reference to FIG.

FIG. 15 shows that the mouth portion of the character “old” is
As shown in (a), a case is shown in which an input character pattern input by handwriting in a circular shape is targeted. As shown in FIG. 15 (b), the input character pattern is elementized into vertical and horizontal lines by a wedge-shaped character approximation unit, and characteristic elements are extracted from the elements. In the figure, the elements indicated by thick lines are the elements extracted as characteristic elements. In this case, when the distance calculation process of FIG. 13 is performed between the feature elements of the dictionary pattern of FIG. 15C, the vertical 2 and the horizontal 4 of the feature elements of the dictionary pattern correspond to the corresponding input patterns. This is a feature element without the -n feature element.

In this case, in step P5-3, only the dictionary-specific feature element search processing is performed. First, (1) a dictionary that is not associated with an input pattern feature element to an element other than the input pattern feature element. A search is made to see if there is one corresponding to the characteristic element of the pattern, and if there is, a distance value between this element and the characteristic element of the dictionary pattern is obtained, and the distance value of the characteristic element of the dictionary pattern is calculated. , Change to this. The determination as to whether or not they correspond can be made, for example, by setting two feature elements as i and j, and d (i, j) = | Pis.x-Pjs.x | + | Pis.y-Pjs.y | + | P
ie.x-Pje.x | + | Pie.y-Pje.y | d (*, j) = | Pjs.x-Pje.x | + | Pjs.y-Pje.y | d (i, *) = | Pis.x-Pie.x | + | Pis.y-Pie.y | is calculated, and it is assumed that d (i, j) is the minimum value.

However, as understood from FIG. 15, the characteristic elements of the dictionary pattern and the elements of the input pattern are not
-The direction of the input pattern may be reversed, so for the input pattern elements, d (i, j)
And the smaller one of d (i, j), which is not reversed and which is not, is adopted.

As the distance value, d (i, j) may be used, but d (i, j) multiplied by N (> 1) or d (i, j)
It is preferable to use a value obtained by adding a constant value to j).

In the case shown in FIG. 15, the characteristic element of the dictionary pattern, width 4 can be associated with the width 2 or width 4 of the input pattern element. When a plurality of elements are associated with an element, one of the plurality of elements having a smaller distance value is preferably adopted as the distance value of the characteristic element. Also, a horizontal line connecting the element, horizontal 2 and horizontal 4 is generated, and if the distance value with this is smaller than the distance value between the element, horizontal 2 and horizontal 4, the distance value is adopted. Is also good.

Next, in the input-only feature element search processing in step P5-4 in FIG. 11, the following processing is performed.

FIG. 16A shows the characteristic elements of the input pattern.
Assuming that FIG. 16 (b) is a characteristic element of the dictionary pattern, FIG.
In (a), the horizontal 1 of the characteristic element of the input pattern indicated by the thickest line remains without being associated with the characteristic element of the dictionary pattern. When there is a characteristic element of the input pattern that is not associated with the characteristic element of the dictionary pattern, there is a possibility that this characteristic element is generated due to continuous writing.

Therefore, when the characteristic element sequence of the input pattern is viewed in the order of input, it can be associated with the characteristic element of the dictionary pattern before and after the characteristic element that could not be associated with the dictionary pattern. If there is a feature element or element, and there is a feature element that could not be matched with the dictionary pattern at the position where the preceding or following feature element or element is linked, this feature element is assumed to have been generated for continuous writing. Considering this, the distance value given to this feature element is changed to a smaller value.

If not, the distance value is changed to a larger value. This was not caused by continuations,
This is because there is an input pattern for a feature element that does not match the dictionary pattern.

More specifically, for example, in FIG. 16 (a), the end point of the front and rear 1 and the start point of the horizontal 2 of the characteristic element horizontal 1 of the input pattern that is not associated with the dictionary pattern are set. If this feature element horizontal 1 is included in a rectangle that is a vertex on a diagonal line or a rectangle obtained by expanding this rectangle by a predetermined amount, the horizontal 1 of the feature element of this input pattern is determined to be a continuous writing part I do.

The character recognition device according to the present embodiment has been described.

In the above description, the case has been described where the input character pattern whose coordinates are input in time series is the target. For example, in the case where character recognition processing is performed offline, the target input character pattern is input. -If there is no information indicating the input order of the coordinates of the
When a character pattern is given, the character pattern is approximated as a polygonal line, and in the feature element extraction, a horizontal line corresponding to the direction of left-right direction or up-down direction or upper left-lower direction,
If the vertical line portion is extracted as a feature element, character recognition can be similarly performed by applying the above embodiment.

Further, of course, the present invention can be applied to recognition of not only Chinese characters but also Hangul characters and other characters having vertical and horizontal lines as main basic elements of the characters.

[0098]

As described above, according to the present invention, even a character which is continuously written or a character which is input in an incorrect stroke order and includes a deformation can be processed with a processing amount which is not so large. A character recognition device capable of correctly recognizing can be provided.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of a character recognition device.

FIG. 2 is a flowchart illustrating a procedure of a process performed by the character recognition device.

FIG. 3 is a flowchart illustrating a processing procedure of a wedge-shaped character approximation processing;

FIG. 4 is a flowchart showing a processing procedure of an element decomposition processing.

FIG. 5 is a flowchart illustrating a processing procedure of an S-shaped element decomposition process.

FIG. 6 is a flowchart showing a processing procedure of an element decomposition processing.

FIG. 7 is a flowchart showing a processing procedure of a feature element extraction process.

FIG. 8 is a flowchart showing a processing procedure of a feature element extraction process.

FIG. 9 is a diagram showing various tables constituting an input pattern of a matching process.

FIG. 10 is a diagram showing various tables constituting a dictionary pattern of a matching process.

FIG. 11 is a flowchart illustrating a processing procedure of a matching process.

FIG. 12 is a flowchart illustrating a processing procedure of distance value calculation processing.

FIG. 13 is a flowchart illustrating a processing procedure of a DP matching process.

FIG. 14 is a diagram illustrating correspondence between characteristic elements of an input pattern and characteristic elements of a dictionary pattern.

FIG. 15 is a diagram showing an outline of a search process for a dictionary-only feature element.

FIG. 16 is a diagram illustrating an outline of a search process for an input-only feature element.

FIG. 17 is a diagram showing an outline of a conventional character recognition process.

[Explanation of symbols]

 1 LCD tablet 2 Pen manager 3 Application 4 Preprocessing unit 5 Wedge-shaped character approximation unit 5-1 Vertical and horizontal element decomposition unit 5-2 Feature element extraction unit 5-3 Element rearrangement unit 6 Dictionary 7 Matching processing unit 7- 1 Matching management unit 7-2 Distance calculation unit 7-3 / Dictionary-only feature element search unit 7-4 Input-only feature element search unit 7-5 Recognition result output unit 8 Detailed identification unit

 ──────────────────────────────────────────────────続 き Continued on the front page (72) Inventor Masaki Miura 7-1-1, Omikacho, Hitachi City, Ibaraki Prefecture Inside Hitachi, Ltd. Hitachi Research Laboratory, Ltd. (72) Inventor Akihiro Katsura 7-1 Omikacho, Hitachi City, Ibaraki Prefecture No. 1 Inside the Hitachi Research Laboratory, Hitachi, Ltd.

Claims (13)

[Claims] 1. A character recognition device for recognizing a handwritten character, comprising: means for inputting a character pattern of a handwritten character to be recognized; and a pattern comprising a vertical line and a horizontal line for each character. A storage unit storing a dictionary in which a certain dictionary pattern is registered, a character pattern is compared with each dictionary pattern registered in the dictionary, and a character corresponding to the dictionary pattern similar to the character pattern is converted into a character pattern. A character recognition device comprising: a recognition unit that recognizes a corresponding character. 2. A character recognition apparatus for recognizing a handwritten character, comprising: means for inputting a character pattern of a handwritten character to be recognized; Input pattern generation means for generating an approximated input pattern; storage means for storing a dictionary in which a dictionary pattern, which is a pattern composed of vertical lines and horizontal lines, is registered for each character; and an input pattern, which is registered in the dictionary. A character recognition device comprising: a recognition unit that compares each of the dictionary patterns with a corresponding one of the dictionary patterns and recognizes a character corresponding to the dictionary pattern similar to the input pattern as a character corresponding to the input character pattern. 3. The character recognition device according to claim 2, wherein said input pattern generation means includes: an element decomposition means for approximating each stroke forming the input character pattern with a vertical line and a horizontal line; Among the vertical lines approximating the stroke, the direction of the stroke corresponding to the vertical line is from the top to the bottom, and the horizontal line approximating the stroke is the direction of the stroke corresponding to the horizontal line from the left to the right. Extract the horizontal line toward
A character recognition apparatus comprising: a feature element extracting unit that forms an input pattern that approximates the character pattern using the extracted vertical and horizontal lines. 4. The character recognition device according to claim 1, wherein each of the dictionary patterns registered in the dictionary includes a stroke forming a character pattern selected for each character as a vertical line. , A stroke drawn from a vertical line approximating a horizontal line and approximating a stroke, a handwritten direction of a portion of the stroke corresponding to the vertical line is directed from top to bottom, and a horizontal line approximating the stroke A character recognition apparatus characterized in that a handwritten direction of a portion of a stroke corresponding to the horizontal line extracted from the above is formed by a horizontal line running from left to right. 5. A character recognition apparatus according to claim 3, wherein said element decomposing means obtains a vertical change point and a horizontal change point in the direction of each stroke forming a character pattern. Means for generating a horizontal line passing through the change point with respect to the change point in the direction, and a vertical line passing through the change point with respect to the change point in the horizontal direction; for each change point, the change point; When a vertical line and a horizontal line are generated with respect to two change points located adjacent to each other on the stroke, an intersection between the generated vertical line and the horizontal line is determined for the two change points. ,
When two vertical lines are generated, a horizontal line passing through a point in a rectangle having the two change points at both ends of a diagonal line is generated, and a horizontal line passing through a point in the rectangle and the two vertical lines are generated. An intersection is obtained for the two change points, and when two horizontal lines are generated, a vertical line passing through a point in a rectangle having the two change points at both ends of a diagonal line is generated, and a vertical line in the rectangle is generated. An intersection calculating means for obtaining an intersection of a vertical line passing through a point and the two horizontal lines with respect to the two change points; and calculating each obtained intersection according to the stroke order of the change point corresponding to the intersection. A character recognition device comprising means for generating a vertical line and a horizontal line obtained by connecting in order as a vertical line and a horizontal line approximating each stroke forming the input character pattern. 6. The character recognition device according to claim 5, wherein the intersection calculating means determines an intermediate point between the two vertical lines when two vertical lines are generated for the two change points. Generating a vertical line passing therethrough, obtaining an intersection between the generated vertical line and a stroke between the two change points, generating a horizontal line passing through the obtained intersection as a horizontal line passing through a point in the rectangle, 2 for one change point
When two horizontal lines are generated, a horizontal line passing through the middle of the two horizontal lines is generated, an intersection between the generated horizontal line and a stroke between the two change points is obtained, and a vertical line passing through the obtained intersection is obtained. A character recognition device for generating a vertical line passing through a point in the rectangle. 7. The character recognizing device according to claim 3, wherein the element disassembling means includes four directions in up, down, left, and right directions according to an angle between a direction of a stroke forming a character pattern and a predetermined reference direction. Categorized into directions, and for the four directions,
A change point at which the direction of the stroke changes is obtained, and the stroke between the change points is approximated to a vertical line when the direction of the stroke between the change points is upward or downward, and the direction of the stroke between the change points A character recognition device that approximates a horizontal line when is leftward or rightward. 8. The character recognition device according to claim 2, wherein said recognition means comprises: a correspondence between a vertical line of the input pattern and a vertical line of the dictionary pattern;
The correspondence between the horizontal line of the input pattern and the horizontal line of the dictionary pattern,
Along with the sum of the similarity related to the position and length of the associated vertical line and the vertical line, and the similarity related to the position and length of the associated qualifying and horizontal line, together with the sum of the similarity, The character recognition device, wherein the recognition unit determines the dictionary pattern similar to the input pattern according to the sum of the similarities obtained for each dictionary pattern. 9. The character recognition device according to claim 2, wherein the input pattern generation means includes: an element decomposition means for approximating each stroke forming the input character pattern with a vertical line and a horizontal line; A polygonal line approximating means for approximating each stroke forming the input character pattern as a polygonal line, and the polygonal line directions correspond to a left-right direction part, an upper-lower part, and an upper left-lower right part. A character recognition apparatus comprising: a vertical line and a horizontal line; and a characteristic element extracting unit configured to form an input pattern that approximates the character pattern using the extracted vertical and horizontal lines. 10. The character recognition device according to claim 3, wherein said recognition means comprises: a correspondence between a vertical line of the input pattern and a vertical line of the dictionary pattern;
The correspondence between the horizontal line of the input pattern and the horizontal line of the dictionary pattern,
Along with the sum of the similarity related to the position and length of the associated vertical line and the vertical line, and the similarity related to the position and length of the associated qualifying and horizontal line, together with the sum of the similarity, The character recognition device, wherein the recognition unit determines the dictionary pattern similar to the input pattern according to the sum of the similarities obtained for each dictionary pattern. 11. A character recognition apparatus according to claim 10, wherein said recognition means comprises: an input pattern which is an array of vertical lines arranged in position order; and a dictionary pattern which is an array of vertical lines arranged in position order. Direct pattern matching between the input pattern, which is an array of horizontal lines arranged in position order, and a dictionary pattern, which is an array of horizontal lines arranged in position order, and performs the vertical line of the input pattern and the vertical direction of the dictionary pattern. Correspondence with the line, correspondence between the horizontal line of the input pattern and the horizontal line of the dictionary pattern, and the similarity of the associated vertical line and the position and length of the vertical line, and the position and length of the corresponding qualifying and horizontal line A character recognition device for calculating a sum total of similarities with respect to each other. 12. The character recognition device according to claim 10, wherein the recognition unit determines a similarity between a horizontal line of the dictionary pattern and a horizontal line of the dictionary pattern that is not associated with the horizontal line of the input pattern. The element decomposition unit searches for a horizontal line that is larger than a predetermined degree from horizontal lines that are approximate strokes and is not included in the input pattern, and determines the similarity between the horizontal line of the dictionary pattern and the searched horizontal line. In addition to the sum of the vertical lines of the input pattern, the vertical lines of the dictionary pattern that are not associated with the vertical lines of the input pattern are approximated by strokes that are similar to the vertical lines of the dictionary pattern by a predetermined degree or more. Search from among vertical lines that are not included in the input pattern, and calculate the similarity between the vertical line of the dictionary pattern and the searched vertical line to the sum of the similarities. Character recognition apparatus characterized by comprising a obtaining means. 13. A character recognition apparatus according to claim 10, 11 or 11, wherein said recognition means determines a horizontal line / vertical line of an input pattern which is not associated with a horizontal line of a dictionary pattern. Whether or not corresponds to the portion caused by continuous writing of the character pattern, the correspondence between the horizontal line or vertical line adjacent to the horizontal line / vertical line and the horizontal line or vertical line of the dictionary pattern, and the horizontal line, Judgment is made according to the horizontal line or the vertical line adjacent to the vertical line, and if it corresponds to the portion generated by continuous writing, the total sum of similarities is increased, or it is generated by continuous writing. A character recognition device that reduces the sum of similarities when the portion does not correspond to the part that has been set.