JPH10334187A - Device and method for character recognition - Google Patents

Abstract

PROBLEM TO BE SOLVED: To recognize an object character fast with high precision even when the input character has variation in stroke order or the number of strokes as to handwritten character recognition for reading a handwritten character inputted on-line. SOLUTION: An input means 1 inputs the handwritten character (input pattern), a stroke extracting means 2 extracts features of the strokes constituting the inputted handwritten character, and a stroke classifying means 4 classifies the strokes by linear strokes, nonlinear strokes, and connecting character strokes according to the features; and a linear stroke correspondence means 5, a nonlinear stroke correspondence means 7, and a connecting character correspondence means 8 make the features of the classified strokes correspond to stroke features of the character to be recognized (standard pattern) stored in a stroke feature dictionary 3, and a character evaluating means 10 recognizes the handwritten character according to the result.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to handwritten character recognition for reading handwritten characters written on a tablet with a pen, and classifies the handwritten characters into strokes constituting the characters and recognizes the recognition based on the characteristics of the classified strokes. The present invention relates to a character recognition device and a character recognition method for recognizing characters.

[0002]

2. Description of the Related Art A conventional online handwritten character recognition device that absorbs both stroke order variation and stroke number variation describes a stroke of a handwritten character by a sequence of feature points, and based on the assumption that stroke combination occurs in the stroke order, excessively responding. One-to-one correspondence between strokes of a handwritten character pattern including stroke order and stroke count fluctuation and strokes of a standard character pattern in a correct typeface with correct stroke order and stroke count is achieved by two complementary optimal matching algorithms: resolution type and shortage resolution type. In this way, stroke order fluctuations are absorbed, and strokes that are not supported are absorbed by the strokes that correspond to the smallest distance when combined with the preceding and succeeding strokes. Recognition, NTT R & D Vol.45 NO.11 1996).

For example, FIG. 34 shows a stroke order shown in the conventional example.
FIG. 9 is a block diagram illustrating a configuration of a conventional device that absorbs a change in the number of strokes. In FIG. 34, reference numeral 201 denotes an input unit for inputting an input pattern which is a handwritten character written on a tablet with a pen; 202, a preprocessing unit for normalizing the input pattern such as sampling, position, and size; A feature point extracting means 204 for extracting, as feature points, points obtained by cutting each stroke constituting the pattern at equal intervals, a standard pattern 204 which is an average character pattern created by a number of square characters written in the correct stroke order and stroke number A standard pattern dictionary that stores
205 is an inter-stroke distance calculating means for calculating an inter-stroke distance between the stroke of the input pattern and the stroke of the standard pattern, and 206 is a one-to-one correspondence between the stroke of the input pattern and the stroke of the standard pattern by two kinds of complementary search algorithms. A one-to-one stroke associating means for absorbing a stroke order variation by associating, 207 is a selective stroke combining means for absorbing a stroke number variation by combining an uncorresponding stroke in the standard pattern with a preceding and succeeding stroke, and 208 is a matching result 209 is a control means for controlling each means.

The operation of the conventional example will be described with reference to the block diagram of FIG. First, the control unit 209 controls the input unit 201
To obtain the input pattern.

[0005] Next, the control means 209 is connected to the input means 201.
Is sent to the preprocessing means 202 to perform sampling and normalization of position and size.

The control means 209 sends the preprocessed input pattern to the feature point extracting means 203, and converts the input pattern into a sequence of feature points at equal intervals on each stroke of the input pattern.

The control means 209 sends the characteristic point sequence of the input pattern to the one-to-one stroke correspondence means 206. The one-to-one stroke association unit 206 associates the input pattern with the standard pattern of the standard pattern dictionary 204 based on the stroke having the smaller number of strokes so that each feature point is not excessively associated (excessive association). Dissolution type correspondence algorithm).

Next, one-to-one stroke association means 20
6 associates the input pattern with the standard pattern of the standard pattern dictionary 204 based on the stroke having the larger number of strokes so that there is no insufficient association between feature points (an association algorithm of a shortage correspondence elimination type). ).

[0009] Of the matching results obtained by the two matching algorithms, the result with the smaller distance is the result obtained by the final one-to-one stroke matching means 206. Here, the distance calculated by the inter-stroke distance calculating means 205 at the time of association includes the end point matching distance which is a value obtained by dividing the sum of the distance between the start point and the end point by 2, the point of the stroke having the smaller number of points. Are sequentially associated from the top point of the stroke having the larger number of points, the sum of the inter-point distances is calculated, and partial matching, which is a value obtained by multiplying by the point ratio, is used.

Next, selective stroke connecting means 207
Reorders the strokes associated by the one-to-one stroke association means 206 according to the stroke order of the stroke with the largest stroke. Strokes that are not associated by the one-to-one stroke association means 206 are combined based on the assumption that "stroke combination occurs in the stroke order".

More specifically, uncorresponding strokes written before the first stroke of the strokes associated by the one-to-one stroke correlating means 206 are connected to the first stroke in a stroke order. Similarly, uncorresponding strokes written after the final stroke are connected to the final stroke in the stroke order. Other non-corresponding strokes are assigned to the one-to-one stroke correlating means 20.
6 exists between any of the strokes associated with each other, the unsupported strokes are temporarily combined with the preceding and following strokes, and cut at a point where the distance between the two strokes is minimized. Break into strokes.

Here, the selective stroke connecting means 207
Then, the points of the stroke with the larger number of points are thinned out at equal intervals, aligned with the smaller number of points, and sequentially associated to calculate the sum of the point-to-point distances, and the overall matching distance, which is the value obtained by dividing by the smaller number of points, use. However, for high-speed processing, the end point matching distance is used in the large classification processing.

Finally, based on the associated results, a proper distance is normalized for the combined stroke, a final distance is calculated, and the character having the shortest distance is determined as a recognition result.

[0014]

As described above, in the conventional character recognition apparatus for absorbing stroke order / stroke number fluctuations, a feature point sequence in which each stroke of a handwritten character is carved at equal intervals is used. Since the distance is obtained, there is a problem that a noise component such as splash and hold, a deformation of a stroke, and a deviation of a coordinate value of a feature point due to a positional deviation affect the stroke distance and cause erroneous reading.

[0015] In addition, in the case of characters having a large number of strokes, such as kanji, individual strokes are often composed of simple straight strokes, so that the characteristic points in the strokes do not contain important information. Since the association is performed at the point level, there is a problem that the association takes a long time and is easily affected by a deviation of the coordinate value.

In addition, in order to reduce the amount of calculation at the time of absorbing the stroke number fluctuation, the assumption that "stroke combination occurs in the order of strokes" is used. There was a problem.

The present invention has been made in order to solve the above-mentioned problems, and it is possible to completely associate a stroke of a handwritten character whose stroke order and stroke number fluctuates with a stroke of a stroke feature dictionary. It is another object of the present invention to provide a character recognition device and a character recognition method capable of realizing high-precision recognition that is not easily affected by local deformation or displacement of a stroke, and performing high-speed character recognition.

[0018]

According to a first aspect of the present invention, there is provided input means for inputting handwritten characters online, and features of a linear stroke and a non-linear stroke constituting each of the recognition target characters are stored in advance for a plurality of recognition target characters. A stroke feature dictionary, a stroke feature extraction unit that extracts a feature of a stroke constituting the handwritten character from the handwritten character input by the input unit, and a feature extracted by the stroke feature extraction unit. Stroke classifying means for classifying each stroke of the handwritten character into one of the straight stroke, the non-linear stroke, and a continuous character stroke which cannot be associated with a stroke constituting the character to be recognized; Of linear stroke classified by means and said stroke characteristic Linear stroke associating means for associating with the features of the linear strokes constituting the recognition target characters stored in the character recognition target, and the features of the non-linear strokes classified by the stroke classifying means and the recognition stored in the stroke feature dictionary A non-linear stroke associating means for associating with a feature of the non-linear stroke constituting the target character; and a continuation character stroke classified by the stroke classifying means, divided into the linear stroke or the non-linear stroke. A continuation character stroke associating means for associating the characteristics of the divided linear strokes or non-linear strokes with the characteristics of the linear strokes or non-linear strokes constituting the recognition target characters stored in the stroke characteristic dictionary; Associating means, the non-linear stroke associating hand Or based on correlation results obtained by the cursive strokes correlating means, but with a character evaluation means for recognizing the handwritten characters.

In a second aspect, the present invention further comprises a stroke associable area determining means for determining an existing area of each stroke of the handwritten character on the basis of the feature extracted by the stroke feature extracting means. Means, the non-linear stroke associating means or the continuation character stroke associating means, for recognizing the recognition target character stored in the stroke feature dictionary corresponding to the existence area of each stroke determined by the associable area determining means. This is for associating with the features of the constituted linear stroke or non-linear stroke.

According to a third aspect of the present invention, the non-linear stroke associating means associates the recognition target characters associated by the linear stroke associating means.

According to a fourth aspect of the present invention, the continuous character stroke associating means associates the recognition target characters associated with each other with the linear stroke associating means or the non-linear stroke associating means.

According to a fifth aspect of the present invention, there is provided an input step of inputting a handwritten character online, a stroke characteristic extracting step of extracting a characteristic of a stroke constituting the handwritten character from the handwritten character input in the input step. Based on the feature extracted in the stroke feature extraction step, each stroke of the handwritten character is associated with the straight stroke, the non-linear stroke, or the continuation character stroke that cannot be associated with the stroke constituting the recognition target character. A stroke classification step of classifying the stroke into any one of the strokes, and a straight stroke correspondence for associating the features of the straight stroke classified by the stroke classification step with the features of the straight stroke constituting the recognition target character stored in the stroke feature dictionary. Attaching step and the straw A non-linear stroke associating step of associating the characteristics of the non-linear strokes classified by the classification step with the characteristics of the non-linear strokes constituting the recognition target character stored in the stroke characteristic dictionary; and The classified continuous strokes are divided into the linear strokes or the non-linear strokes, and the characteristics of the divided linear strokes or the non-linear strokes and the linear strokes or the non-linear strokes that constitute the recognition target characters stored in the stroke characteristic dictionary. Based on a continuation character stroke correspondence step of performing correspondence with a characteristic of a straight stroke, based on the correspondence result obtained by the straight stroke correspondence step, the non-linear stroke correspondence step or the continuation character stroke correspondence step. , The handwritten sentence Those with a character evaluation step recognizes.

According to a sixth aspect of the present invention, there is provided a stroke associable area determining step of determining an existing area of each stroke of the handwritten character on the basis of the feature extracted in the stroke feature extracting step. The step, the non-linear stroke associating step or the continuation character stroke associating step includes recognizing the recognition target character stored in the stroke feature dictionary corresponding to the existence area of each stroke determined by the associable area determining step. This is for associating with the features of the constituted linear stroke or non-linear stroke.

In a seventh aspect, the non-linear stroke associating step associates the recognition target characters associated by the linear stroke associating step.

According to an eighth aspect, in the continuation character stroke associating step, the recognition target characters associated by the linear stroke associating step or the non-linear stroke associating step are associated with each other.

[0026]

BEST MODE FOR CARRYING OUT THE INVENTION

Embodiment 1 FIG. Hereinafter, Embodiment 1 will be described with reference to the drawings. FIG. 1 is a configuration diagram of a character recognition device of the present invention,
In the figure, reference numeral 1 denotes input means for inputting characters by handwriting on a tablet-type input device, 2 denotes stroke feature extracting means for extracting the characteristics of the strokes constituting the handwritten characters from the handwritten characters input by the input means 1, and 3 denotes For a plurality of characters to be recognized, a stroke characteristic dictionary in which stroke characteristics of straight strokes and non-linear strokes constituting a standard pattern, which is each character to be recognized written in the correct stroke order, are stored in advance.
Is a stroke classifying means for classifying each stroke of a handwritten character into one of a linear stroke, a non-linear stroke, and a continuation character stroke based on the features extracted by (1). 5 is a linear stroke associating means for associating the characteristics of the linear strokes classified by the stroke classifying means 4 with the characteristics of the linear strokes constituting the standard pattern stored in the stroke characteristic dictionary 3; A stroke associable area determining means for determining an existing area of each stroke of the handwritten character based on the features extracted by the means; Non-linear stroke associating means for associating with the features of the non-linear strokes constituting the stored standard pattern, 8 divides the continuation stroke classified by the stroke classifying means into a linear stroke or a non-linear stroke, The characteristics and strokes of this divided straight stroke or non-linear stroke A continuation character stroke associating means for associating with a feature of a linear stroke or a non-linear stroke constituting a standard pattern stored in the Loke feature dictionary 3, and 9 is a partial stroke creating means for creating a partial stroke from the continuation character stroke is there. Reference numeral 10 denotes a character evaluation unit that obtains an evaluation value of a character from the association result obtained by the linear stroke association unit 5, the non-linear stroke association unit 7, and the continuation character stroke association unit 8, and recognizes a handwritten character. Control means for controlling each means.

FIG. 2 is a flowchart showing the flow of processing of the character recognition device shown in FIG. An example of a processing procedure will be described with reference to a flowchart showing the flow of processing of the character recognition device in FIG. First, in step S1, characters are handwritten by the input means 1 to the tablet type input device with a pen. FIG. 3 shows an input pattern "A" of a handwritten character input to the input means 1, and reference numerals 60 to 65 denote strokes input sequentially in time. In this example, both the stroke order and the number of strokes are not regular, and are composed of six strokes. Proceeding to step S2, the stroke feature extracting means 2 extracts the features of each stroke constituting the input pattern of the handwritten character. In this example, the stroke features of the strokes 60 to 65 of the input pattern in FIG. 3 are extracted.

FIG. 4 shows the stroke features of the input pattern extracted by the stroke feature extraction means 2, and shows the stroke features obtained from the strokes 60 to 65 of the input pattern of FIG. In this stroke feature, the direction of the stroke and the direction of a virtual stroke that is not actually written connecting one stroke to the next are assigned in order to quantize the stroke shown in FIG. 5 into 16 directions. It is determined using a 16-direction distribution map.

Next, proceeding to step S3, the stroke classifying means 4 classifies the stroke of the input pattern according to its shape. FIG. 6 is a flowchart showing the flow of the processing of the stroke classification means 4. Stroke classification means 4
Will be described in detail with reference to the flowchart of FIG.

First, in step S10 of FIG. 6, the stroke classifying means 4 determines whether the current stroke is a linear stroke in order from the top stroke. If the current stroke is a linear stroke, the process proceeds to step S11. Otherwise, the process proceeds to step S13.
Proceed to. In the input pattern of FIG. 3, five strokes 60, 61, 62, and 6 are obtained from the shape of the stroke feature of FIG.
Steps S11 and S65 are determined to be linear strokes.
Proceed to.

In step S11, the stroke classifying means 4
Extracts the directional component distribution of the stroke. The directional component distribution of the stroke is obtained from the directional distribution of each straight line connecting the points obtained by sampling the input stroke at appropriate intervals. Here, when the stroke classification means 4 obtains the direction distribution, the eight-direction component distribution shown in FIG. 7 is used.

In step S12, the stroke classifying means 4 checks whether the obtained directional component is only in a fixed direction. If the directional component is only in a fixed direction, the process proceeds to step S15, where the stroke component is determined to be a linear stroke. Otherwise, the process proceeds to step S13. .
In the example of the input pattern shown in FIG. 3, since all the linear strokes have components only in a certain direction, step S1 is executed.
The process proceeds to 5 and is determined to be a straight stroke.

If the direction component is not only a fixed direction, in step S13, the stroke classifying means 4 determines whether the stroke shape existing in the stroke feature dictionary 3 is for a straight stroke having a plurality of direction components and a stroke other than a straight line. If the stroke is not found in the dictionary, the process proceeds to step S16 to determine that the stroke is a character stroke. If the stroke exists in the dictionary, the process proceeds to step S14, where it is determined that the stroke is a non-linear stroke. In the example of the input pattern of FIG. 3, the unknown stroke 64 is a stroke that does not exist in the dictionary, and thus the process proceeds to step S16.
It is determined to be a continuation character stroke.

Next, in step S4 of FIG. 2, the control means 11 sends the stroke determined as a straight stroke by the stroke classification means 4 to the straight stroke associating means 5, and the straight stroke of the input pattern and the stroke feature dictionary. 3 is associated with the linear stroke of the standard pattern. In the example of the input pattern shown in FIG. 3, the linear strokes 60, 61, 62, 63, 65 are sent to the linear stroke associating means 5.

FIG. 8 is a flow chart showing the flow of the processing of the linear stroke associating means. The operation of the linear stroke associating means 5 will be described with reference to the flowchart of FIG. First, the linear stroke associating means 5
In step S21, each straight stroke is classified into three types (long, middle, and short) according to the length. The length of the stroke is obtained from the Euclidean distance between the start point and the end point of the stroke. In this example, a long stroke is defined as 70 or more, a short stroke is defined as 30 or less, and an intermediate stroke is defined as an intermediate stroke. From the length information of the input stroke shown in the stroke feature of the input pattern in FIG.
Strokes 62 and 65 are classified as long strokes, and stroke 63 is classified as short strokes.

In step S22, the linear stroke associating means 5 determines the area in which the stroke is present from the direction of the stroke, the coordinates of the start and end points of the stroke, and the length of the stroke by the stroke associable area determining means 6. FIG. 9 shows an example of area division used in associating linear strokes. Here, an example is shown in which a stroke existence area is determined by an area obtained by dividing a character into four parts in a horizontal direction and a vertical direction by a circumscribed rectangle.

FIG. 10 shows information used for determining an area, which is used by the stroke assignable area determining means 6. The stroke direction in FIG. 10 indicates a value obtained by quantizing the direction from the start point to the end point of the stroke in the 16-direction distribution diagram of FIG. For example, for the horizontal straight stroke in FIG.
This indicates that there is a possibility of being associated with the horizontal stroke of the standard pattern in the directions of 6, 12, 13, and 14. Also, the horizontal straight stroke existing in the area 1 is
This indicates that there is a possibility of being associated with a horizontal straight stroke existing in the area 2.

As shown in FIG. 10, the stroke associable area determining means 6 determines that the input pattern has a relatively small variation in the distance between the stroke direction and the vertical direction and is relatively stable. I do. However, when a plurality of regions are straddled, a region adjacent to the plurality of regions is set as an existing region. FIG. 11 shows a region where a stroke exists and a region that can be associated with the input pattern of FIG.

Proceeding to step S23, the linear stroke associating means 5 determines the stroke of the input pattern.
The stroke of the standard pattern of the stroke feature dictionary 3 to be associated is determined. FIG. 12 shows a standard pattern stroke of the character “A” stored in the stroke feature dictionary 3. FIG. 13 shows the stroke features of the standard pattern of FIG. 12 stored in the stroke feature dictionary 3. Conditions that can be associated are 1)
Stroke shape condition, 2) Stroke length condition, 3)
The stroke of the standard pattern to be associated is determined based on the three conditions that can be associated with the stroke.

FIG. 14 shows whether or not the correspondence can be made according to the stroke shape. The stroke shapes used by the linear stroke associating means 5, the non-linear stroke associating means 7, and the continuation character stroke associating means 8 are shown. This determines whether or not the corresponding association is possible. More specifically, the stroke shape condition is such that, in the correspondence relationship shown in FIG. 14, the stroke shape of the input pattern and the stroke shape of the standard pattern can be associated with each other with a circle, and x must be associated with each other. I do. Also, the stroke length condition is
A long stroke can be associated with a long stroke or an intermediate stroke, an intermediate stroke can be associated with a long stroke and a short stroke, and a short stroke can be associated with an intermediate stroke or a short stroke. In addition, the area condition that can be associated with the stroke is
The result of the stroke assignable area determining means 6 is used.

In this example, the stroke feature of FIG. 13 is stored in the stroke feature dictionary 3 with respect to the standard pattern of FIG. Based on the stroke characteristics, a stroke of a standard pattern that can be associated with each stroke of the input pattern is determined. In the present embodiment, a stroke of a standard pattern that can be associated with each stroke of the input pattern of FIG. 3 is determined.

FIG. 15 shows how the strokes of the standard pattern that can be associated are narrowed according to the stroke shape condition, the stroke associable area condition, and the stroke length condition. Here, the narrowing down by each condition is sequentially performed only on the narrowed down ones.

As shown in FIG. 15, the stroke 60 corresponds to the strokes 70 and 73, and the stroke 61 corresponds to the strokes 74 and 75.
, Stroke 62 becomes stroke 75, stroke 6
3 is reduced to a stroke 71, and the stroke 65 is reduced to strokes 73 and 76. Here, since there is no other correspondence for the stroke narrowed to one,
The narrowed standard stroke cannot be used for associating with another input stroke. In this case, since the strokes 71 and 75 are reduced to one, the strokes 71 and 75 are reduced from other input stroke candidate strokes.
except for. As a result, the stroke 61 becomes the stroke 74
Finally, only the strokes 60 and 65 have two stroke candidates.

Next, proceeding to step S24, the linear stroke associating means 5 performs detailed correspondence between the strokes of the input pattern and the standard pattern candidate strokes, and returns all the results of the successful correspondence to the linear strokes. Assume the association result. Here, the detailed correspondence check is performed by comparing the coordinate values of the stroke direction, the stroke width, the stroke height, and the start and end points of the stroke with a predetermined correspondence threshold value. Check if is possible. At this time, if the input stroke is associated with a continuous stroke of the standard pattern, the association is verified using the association result of the virtual stroke. Also, in order to absorb the input pattern stroke order variation, all the strokes of the standard pattern that can be associated with the strokes of each input pattern so that the strokes can be associated even when the input pattern is different from the standard pattern stroke order. The candidate strokes are associated.

More specifically, a comparison is made between the stroke characteristics of the input pattern shown in FIG. 4 and the stroke characteristics of the standard pattern shown in FIG. In this example, stroke 6 of the input pattern
For 1, 62 and 63, the candidate for association has already been 1
And the strokes are correct, so that the correspondence is successful. Since there are two candidates for the strokes 60 and 65, the standard pattern candidate stroke is associated with each input pattern stroke. That is, the stroke 60 of the input pattern is replaced with the stroke 7 of the standard pattern.
Stroke 73 and stroke 76 are associated with 0, stroke 73, and stroke 65.

First, the process of associating the strokes 60 of the input pattern will be described. The stroke 60 is associated with the strokes 70 and 73 of the standard pattern.
Here, when the stroke 60 and the stroke 73 are associated with each other, the difference between the Y coordinate values of the start and end points becomes large. Also,
When the correspondence result is checked using the virtual stroke indicating the positional relationship between the strokes, in the case of the stroke 73, the direction of the virtual stroke to the next stroke 74 is
It becomes “16” from FIG. In the case of stroke 60,
Since the stroke of the input pattern corresponding to the standard stroke 74 of the dictionary is the stroke 61, the stroke 6
The direction of the virtual stroke of 0 is the direction from the end point of the stroke 60 to the start point of the stroke 61, and is “13”.
Therefore, the difference in the virtual stroke direction is 16-13 = 3. On the other hand, in the association between the stroke 60 and the stroke 70, the difference between the coordinate values of the start and end points is small, and the difference in the direction of the virtual stroke is small (the distance of the association result is small).

Next, the stroke 65 and the standard pattern strokes 73 and 76 are associated with each other. Stroke 6
When 5 and the stroke 73 are associated with each other, the difference between the coordinate values of the start and end points (the difference between the stroke characteristics) increases. On the other hand, when the stroke 65 is associated with the stroke 76,
The difference between stroke features is small and correct association is achieved.

As a result, the correspondence of the linear stroke is
The stroke 60 of the input pattern corresponds to the stroke 70 of the standard pattern, the stroke 61 corresponds to the stroke 74, the stroke 62 corresponds to the stroke 75, the stroke 63 corresponds to the stroke 71, and the stroke 65 corresponds to the stroke 76. Can be As in this example,
By the above-described associating process, it is possible to make an association even when the stroke order of the strokes of the input pattern is different from the stroke order of the standard pattern.

Proceeding to step S25, the linear stroke associating means 5 proceeds to step S27 if the linear stroke of the input pattern corresponds to one of the strokes of the standard pattern, and sets the association to OK (OK). If there is any unsupported stroke, step S2
8 and the association with the current standard pattern is not possible (N
G). In this example, the straight strokes of all the input patterns were correctly associated with the strokes of the standard pattern.

Next, proceeding to step S5 in FIG. 2, the control means 11 checks the result of associating the linear strokes, and if the association is not possible (NG), proceeds to step S8 to associate with the next standard pattern. Do. If the association is successful, the process proceeds to step S6, where the association with the non-linear stroke is performed.

FIG. 16 is a flowchart showing the flow of the processing of the non-linear stroke associating means 7. The operation of the non-linear stroke associating means 7 will be described in detail with reference to the flowchart of FIG. In step S30, the non-linear stroke associating means 7 determines the stroke of the standard pattern to be associated with the stroke of the input pattern by the stroke associable area determining means 6 for the non-linear stroke as in the case of the linear stroke. This association uses the information diagram for determining the area in FIG. 10 similarly to the linear stroke association. Here, the correspondence between the non-linear stroke of the input pattern and the stroke of the standard pattern is performed only for the standard pattern having the linear stroke associated with the linear stroke associating means 5.

In step S31, the non-linear stroke associating means 7 associates the input pattern with the corresponding stroke of the standard pattern for each stroke shape. First, a correspondence between strokes of the input pattern obtained in the process of step S30 and a stroke that can be associated with the standard pattern is checked by a stroke shape. The rules for associating stroke shapes are shown in FIG.
4 is a diagram showing the possibility of association in accordance with the stroke shape. Next, for strokes that are determined to be able to be associated, a detailed stroke association check is performed in the same manner as for a straight stroke.

Proceeding to step S32, if it is determined that the stroke of the input pattern is associated with one of the strokes of the standard pattern, the non-linear stroke associating means 7 proceeds to step S34 and determines that the association is successful. . If there is no correspondence, it is assumed that there is no match with the stroke of the standard pattern, and the process proceeds to step S33, where it is determined that the stroke is a continuous character stroke.

Proceeding to step S35, the non-linear stroke associating means 7 checks whether or not all the non-linear strokes have been associated with each other.
Proceeding to 30, the next non-linear stroke is associated.
When all the associations have been completed, the association of the non-linear stroke association means 7 is completed.

Next, proceeding to step S7 in FIG. 2, the control means 11 causes the continuous character stroke associating means 8 to associate continuous character strokes.

FIG. 17 is a flowchart showing the flow of the processing of the continuation character stroke associating means 8. Hereinafter, the operation of the continuation character stroke associating means 8 will be described with reference to the flowchart of FIG. First, step S50
Then, the continuation character stroke associating means 8 determines the continuation character stroke associable area of the input pattern by the stroke associable area determining means 6, and the standard pattern stroke which can be associated with the continuation character stroke of the input pattern. Ask for a candidate. Here, the continuation stroke correspondence of the input pattern is performed only for the standard pattern in which the linear stroke and the non-linear stroke associated by the linear stroke associating means 5 and the non-linear stroke associating means 7 exist.

FIG. 18 is a diagram showing an example of area division used by the stroke associable area determination means 6 in the case of associating other than a straight stroke. The stroke associable area determination means 6 determines a stroke associable area using the area division shown in FIG. 18 for a stroke other than the strokes shown in the information diagram for associable area determination in FIG. The stroke associable region covers up to a region outside the region including the circumscribed rectangle of the stroke of the input pattern. In the example of the input pattern shown in FIG. 3, the stroke 64 is determined to be a continuation character stroke. Since the stroke is not the stroke shape shown in FIG. 10, first, an area including the stroke 64 is determined.

FIG. 19 shows an example in which the input pattern shown in FIG. 3 is divided by the area division shown in FIG. 18, and a circumscribed rectangle 80 of the stroke 64 is obtained. It is a figure which shows where in the area | region used by (1). From this figure, it can be seen that the circumscribed rectangle 80 of the stroke 64 is included in the regions A to H, and the regions that can be associated with the strokes are the regions A to L.

Next, the continuation character stroke associating means 8
Calculates the stroke of the standard pattern existing in the determined assignable area. In the case of the example of the stroke of the standard pattern shown in FIG. 12, four strokes 70, 71, 72, and 73 are included in the areas A to L. Among them, the straight stroke associating means 5 and the non-linear stroke Strokes of the standard pattern that are associated one-to-one with the input strokes by the stroke association means 7 are excluded because the association has been determined. Therefore, in this example, the strokes 70 and 71 are excluded because they are determined by the linear stroke association means. As a result, the strokes 72 and 73 are selected as candidate strokes.

Proceeding to step S57, the continuous character stroke associating means 8 checks whether or not an associative area has been determined for all continuous character strokes. If so, the flow advances to step S51 to make the determination. If not, the process returns to step S50. In the input pattern of FIG. 3, since there is only one continuous character stroke, the process proceeds to step S51 to determine a standard pattern candidate stroke to be a start stroke and an end stroke of the continuous character stroke of the input pattern.

More specifically, for the start and end points of the continuation character stroke of the input pattern, the area where the start and end points exist is obtained by the area division shown in FIG. It is a possible area. Next, a stroke including the start point of the standard pattern corresponding to the associable area where the start point of the continuation character stroke of the input pattern exists is selected as a start stroke candidate. Similarly, a stroke including the end point in the stroke of the standard pattern corresponding to the associable area in which the end point of the input continuation character stroke exists is selected as an end stroke candidate.

In the example shown in FIG. 19, since the starting point of the continuation character stroke 64 of the input pattern exists in the area A, the areas to which the starting point can be associated are the areas A, B, E, and F.
Since the starting points of the strokes 72 and 73, which are the candidates for the standard pattern obtained in the process of step S50, are both included in this area, the starting stroke candidate is the stroke 7
2, 73. Similarly, the continuation character stroke 64
Is located in the area H, the associable area is
The areas are C, D, G, H, K, and L. Thus, the end point stroke candidates are also the strokes 72 and 73.

Proceeding to step S52, the continuation character stroke associating means 8 determines whether the total number of strokes of the starting and ending point candidates determined in step S51 (however, the same stroke is not counted repeatedly) is three or less. If the number is three or less, the process proceeds to step S54. If the number is three or more, the process proceeds to step S53. In the case of the continuous stroke in FIG. 3, the number of candidate strokes is two, so the process proceeds to step S54.

In step S54, a continuation character stroke of the input pattern is associated with a standard pattern candidate stroke. FIG. 20 is a flowchart showing the flow of the association processing in step S54. With reference to the flowchart of FIG. 20, the association of the continuation character stroke will be described in detail.

First, in step S60, the continuation character stroke associating means 8 determines the stroke order of the standard pattern candidate strokes. Here, in step S60, the number of standard pattern candidate strokes is always reduced to three or less. Since the continuation character stroke is composed of at least two strokes, the combination of the stroke order when there are three stroke candidates of the standard pattern is 3C2 × P2 + P3 = 12. Similarly, when there are two standard pattern candidate strokes, there are two types.

Since there are 12 cases in the worst case, the combination can be sufficiently processed in real time.
Limit the number of combinations by one constraint. 1) Candidates for starting stroke determined in step S51,
Restriction by End Stroke Candidate 2) Restriction by Undeletable Stroke Here, the non-deletable stroke refers to a standard pattern stroke that corresponds only to the current input pattern stroke candidate. Since the candidates of the standard pattern strokes that can be associated with the strokes of each input pattern are determined in the process of step S50, the deletion stroke can be determined based on this information.

In the case of the stroke 64 of the input pattern,
There are two candidate stroke numbers, and both strokes are candidates for the start stroke and the end stroke.
Since it is not a deletion stroke, two stroke orders are used. In this example, first, the correspondence is performed in the stroke order of the strokes 72 and 73.

In step S61, the continuation character stroke associating means 8 associates the continuation character stroke of the input pattern with the stroke of the standard pattern. The association is performed at the feature point level, and the stroke feature point information of the standard pattern includes only the start point and the end point in the case of a straight stroke, and the feature point exists near the start point, the end point, and the bending point of the stroke other than the straight stroke. This feature point stored in the feature dictionary 3 is also used. Also, the stroke feature point information of the input pattern is assumed to be sampling points expressed at appropriate intervals after the preprocessing.

According to the stroke characteristics of the standard pattern shown in FIG. 13, the characteristic point information of the stroke 72 has three points, namely, the starting point, the characteristic point in the stroke, and the ending point. The characteristic point information of the stroke 73 has two points, the starting point and the ending point. You can see it exists. Therefore, there are five stroke feature points of the standard pattern. Stroke feature points of the obtained standard pattern are shown by circles in FIG. In addition, characteristic points of the continuation character stroke 64 of the input pattern are indicated by circles in FIG.

Next, the continuation character stroke associating means 8
Performs association in which the distance between the stroke feature point of the input pattern and the stroke feature point of the standard pattern is minimized. In general, the correspondence between the stroke A of the input pattern and the stroke B of the standard pattern is determined from the set {0, 1,..., I} of the element sequence of the characteristic point of the stroke A of the input pattern. The mapping to the set of element sequences {0, 1,..., J} ω: {0, 1, 2,..., I} → {0, 1, 2,. . Where ω is the stroke A
And a mapping that matches both end points of the stroke B, ω (0)
= 0, ω (I) = J (that is, the start point and the end point coincide)
Is satisfied, and the order of the components is a monotonous mapping that does not allow inversion, and is therefore called an expansion mapping. Also, stroke A
Is the ith element of ai, the jth element of the stroke B is bj, and the distance between ai and bj is d (ai, bj): feature point a
Assuming that the Euclidean distance between i and bj, the accumulated distance to the i-th element of the stroke A, and the j-th element of the stroke B are g (i, j), the recurrence formula 1 ≦ i ≦ I, 0 ≦ j ≦ J
G (i, j) = d (ai, bj) + min {g (i-1, k) | 0 ≦ k ≦ j} (Equation 2) where g (0,0) = d (a0 , B0), g (0, j)
= ∞ (when j> 0) holds. Thereby, the stroke A of the input pattern
G of correspondence between stroke and standard pattern stroke B
(I, J) can be obtained by dynamic programming.
Here, if the value of k that gives min {g (i-1, k) | 0 ≦ k ≦ j} is recorded, information on the optimal correspondence between feature points can be obtained. That is, it is possible to determine which feature point of the continuation character stroke of the input pattern corresponds to each feature point of the stroke of the standard pattern.

When the feature points of the stroke of the standard pattern of FIG. 21 and the stroke feature points (sampling points) of the input pattern of FIG.
The evaluation result of the association and the association result indicated by the arrow in FIG. 23 are obtained.

In step S62, the continuation character stroke associating means 8 determines the stroke characteristics of the divided strokes from the obtained associating result by the stroke characteristic extracting means 2. In order to obtain the stroke feature, first, the continuation character stroke of the input pattern is divided by the characteristic points corresponding to the start and end points of the stroke of the standard pattern, and the stroke sandwiched between the start and end points is extracted as a divided stroke. Next, the stroke feature of the divided stroke is extracted by the stroke feature extracting means 2. FIG. 24 is an example in which the dividing strokes 130 and 131 are obtained from the stroke 64 of the input pattern.

Proceeding to step S63, the continuous character stroke associating means 8 checks whether the current candidate stroke has been associated with the continuous character stroke of the input pattern in all stroke orders. As a result of the check, if the association by all the strokes has not been performed, the process proceeds to step S60, and the association is performed for another stroke order.

In the case of the stroke 64 of the input pattern,
Proceeding to step S60, association is performed in the stroke order of strokes 73 and 72. The associating process is performed for stroke 7
2 and 73, the description is omitted here. Here, the case of the stroke order of the strokes 72 and 73 is as follows.
Since the distance between the stroke feature point of the input pattern and the stroke feature point of the standard pattern becomes smaller, it is adopted as a result of associating correct partial strokes.

When the continuation character stroke associating means 8 completes associating all the stroke orders with the continuation character stroke of the current input pattern, step S56 in FIG.
To check if all continuation character strokes have been matched, and if the matching has been completed,
The processing of the continuation character stroke associating means 8 is completed, and FIG.
Go to step S8. If the association has not been completed, the process returns to step S51 in FIG. In the input pattern of FIG. 3, the condition is satisfied because there is one continuous stroke, and the process proceeds to step S8.

In step S8, the control means 11 checks whether or not matching has been performed for all the standard patterns in the stroke feature dictionary 3. If the matching has been completed, the process proceeds to step S9. Proceeding to S4, association with the next standard pattern is performed.
Regarding the input pattern of FIG. 3, since the association with all the standard patterns has been completed, the process proceeds to step S9.

In step S 9, the control unit 11 sends the matching result obtained by the straight stroke matching unit 5, the non-linear stroke matching unit 7, and the continuation character stroke matching unit 8 to the character evaluation unit 10. Character evaluation means 10
Calculates the evaluation value of the character based on the sent association result, and recognizes the recognition target character (standard pattern) having the shortest distance.
Is selected as the final recognition result.

The evaluation value D of the character is obtained by the following equation (3). D = (Wd × Dd + Ww × Dw + Wh × Dh + Wvw × Dvw + Wvh × Dvh + Wvd × Dvd) / k (Equation 3) Here, the distance Dd after the normalization of the stroke direction, the distance Dw after the normalization of the stroke width, and the stroke height. Distance Dh after normalization, and distance Dv after normalization of virtual stroke width
w, the distance Dvh after the virtual stroke height is normalized, the distance Dvd after the virtual stroke direction is normalized, Wd is the weight in the stroke direction, Ww is the weight in the stroke width, Wh
Is the weight of the height of the stroke, Wvw is the weight of the width of the virtual stroke, Wvh is the weight of the height of the virtual stroke, Wv
d indicates the weight of the direction of the virtual stroke, and k indicates the number of strokes of the current standard pattern. It is assumed that Wd + Ww + Wh + Wvw + Wvh + Wvd = 1.

However, when there are a plurality of correspondence results of the linear stroke, the non-linear stroke, and the continuation stroke, there is a stroke of an unsupported standard pattern which does not correspond to any of the input patterns. Consideration is made so that there is no contradiction that the stroke of one standard pattern corresponds to the stroke of a plurality of input patterns.

Next, the partial stroke extracting means 9 will be described. FIG. 25 is an example of an input pattern including continuation character strokes in which seven strokes of a character "word" are written in four strokes. Hereinafter, the operation of the partial stroke extracting means 9 will be described using the input pattern of FIG. 25 as an example. In this example, stroke 14
1, 142 and 143 are linear stroke associating means 5
It is assumed that they are associated with each other by the non-linear stroke associating means 7, and the description thereof is omitted. Stroke 140
Is a continuation character stroke, so that the process proceeds to step S7 in FIG.

First, proceeding to step S50 in FIG. 17, the continuation character stroke associating means 8 determines the associative area of the continuation character stroke of the input pattern by the stroke associable area determining means 6, and determines the continuation character stroke. A stroke candidate of a standard pattern that can be associated is obtained.

Stroke associable area determining means 6
Determines an associable area using the area division information in FIG. The associable area covers up to an area outside the area including the circumscribed rectangle of the stroke of the input pattern. FIG. 26 shows an example in which the area division shown in FIG. 18 is performed on the input pattern of FIG. 26 that the circumscribed rectangle 144 of the stroke 140 includes the regions A to L. As a result, the stroke associable regions of the stroke 140 are all regions A to P.

The continuation character stroke associating means 8 obtains a stroke of the standard pattern existing in the determined associable area. FIG. 27 shows a standard pattern of the character “word”, in which all strokes are included in the areas A to P. Of these strokes, the linear stroke associating means 5 and the non-linear stroke associating means 7 input the input pattern. The strokes of the standard pattern that are associated one-to-one with the strokes are excluded because the association has been determined. In this example, the strokes 154 and 156 are determined by the linear stroke associating means 5, and the stroke 155 is determined by the non-linear stroke associating means 7. As a result, four strokes 150, 151, 152, and 153 are selected as candidate strokes.

Next, the continuation character stroke associating means 8
Proceeds to step S57 in FIG. 17, checks whether or not an associative area has been determined for all continuation character strokes, and if determined, proceeds to step S51;
If not, the process proceeds to step S50. In the input pattern of FIG. 25, since there is only one continuous character stroke, the process proceeds to step S51 to determine a standard pattern candidate stroke to be a start stroke and an end stroke of the continuous character stroke. 26, since the starting point of the stroke 140 of the input pattern exists in the area B, the associable areas are the areas A, B, C, E, F, and G, and the stroke 15 of the standard pattern including the starting point in this area is shown.
0, 151, and 152 are candidates for the start stroke.
Similarly, the strokes 152 and 153 are candidates for the end stroke.

In step S52, the continuation character stroke associating means 8 checks the number of stroke candidates in the standard pattern. In this example, the number of candidate strokes is four, so the flow proceeds to step S53. In step S53, the continuation character stroke association means 8 causes the partial stroke extraction means 9 to extract a partial stroke from the continuation character stroke of the input pattern.

FIG. 28 is a flowchart showing the flow of the process of associating partial strokes in step S53.
The operation of the partial stroke extracting means 9 will be described with reference to the flowchart shown in FIG. In step S70, the partial stroke extracting means 9 extracts a bending point of the stroke of the input pattern as a partial stroke dividing candidate point, and sets the starting point of the partial stroke as the starting point of the stroke of the input pattern. FIG. 29 is a diagram illustrating a breakpoint candidate of a continuation character stroke of the input pattern. In FIG. 29, 160-1
65 is a candidate point for partial stroke division of the stroke 140 of the input pattern, 166 is a starting point of the partial stroke, 167
Is the end point of the partial stroke.

In step S71, the partial stroke extracting means 9 traces the dividing point in the reverse direction from the end point of the stroke of the input pattern, and the number of candidate strokes from the starting point of the current partial stroke to the dividing point is three or less. Is determined, and the divided point is set as the end point of the partial stroke. In the example of FIG. 29, first, from the starting point 166, the division candidate points 160, 161, 162, 163, 164, 165.
Are selected in the regions A, B, C, D, E, F, G, I, J, and K, so that the assignable regions are all regions,
Since the number of candidate strokes remains four, the process proceeds to the division candidate point 164 before the division candidate point 165. Similarly, the dividing point is sequentially moved in the direction of the starting point, a partial stroke is obtained, and the candidate stroke is extracted. In this example, since the number of candidate partial strokes from the starting point 166 to the division candidate points 160, 161, and 162 is three, this stroke is selected as a partial stroke.

In step S72, the continuation character stroke associating means 8 associates the obtained partial strokes with the candidate strokes. FIG. 31 is a flowchart showing the flow of the process of step S72. Step S7
The process 2 will be described with reference to the flowchart of FIG. In step S80, candidates that can be associated are narrowed down by the circumscribed rectangle of the partial stroke and the circumscribed rectangle of each candidate stroke. Specifically, a candidate stroke including a circumscribed rectangle in an area obtained by expanding the circumscribed rectangle of the partial stroke by a predetermined threshold value is selected. Reference numeral 180 in FIG. 30 denotes a circumscribed rectangle of the extended partial stroke. In this example, stroke 1 of the candidate strokes 150, 151, 152
52 is removed from the candidates.

In step S81, the continuation character stroke associating means 8 determines the stroke order of the standard pattern candidate strokes as in the case where the number of candidate strokes is three or less. Here, the determination of the combination of the stroke order also uses information that is a candidate of a start stroke when the current partial stroke includes a start point of a continuation character stroke of the input pattern, and information that is a candidate of an end stroke when the current partial stroke includes an end point. Since there are two candidate stroke numbers for the current partial stroke, and all of the strokes satisfy the condition of the start stroke, the stroke order is two. Here, first, association is performed in the stroke order of strokes 150 and 151.

In step S82, the continuous character stroke associating means 8 associates the partial strokes with the standard pattern candidate strokes. The method of associating is the same as step S61 in FIG. 20 (similar to the case of associating the continuation character stroke). Step S6 in FIG.
In FIG. 1, since the strokes 150 and 151 are both straight strokes, the starting point and the ending point are feature points, and the correspondence with the feature points of the partial strokes is obtained by dynamic programming.

In step S83, the continuation character stroke associating means 8 divides the partial stroke from the obtained associating result, and the stroke feature extracting means 2 determines the stroke characteristics of the divided stroke. FIG.
Is the dividing stroke 19 from the partial stroke in FIG.
This is an example in which 0 and 191 are obtained.

In step S84, the continuation character stroke associating means 8 checks whether the current candidate stroke has been associated with the partial stroke of the input pattern in all stroke orders. As a result of the check,
If all the combinations have not been performed, the process proceeds to step S81, and another stroke order is associated.
In this example, the process proceeds to step S81 and the stroke 15
Correspondence in another stroke order of 1, 150 is performed. The associating process is the same as in the case of the strokes 150 and 151, and the description is omitted. Here, strokes 150, 15
In the case of the first stroke order, the distance between the stroke feature point of the input pattern and the stroke feature point of the standard pattern becomes smaller, so that it is adopted as the result of associating correct partial strokes.

Next, the continuation character stroke associating means 8
Proceeds to step S73 in FIG. 28, and determines whether or not the association of the current continuation character stroke is completed. If completed, step S75 is performed. If not completed, step S75 is performed.
Proceed to 74. Since the processing is not completed in the example of FIG. 29, the process proceeds to step S74. In step S74, the end point of the current partial stroke is set as the start point of the next partial stroke. Here, the division candidate point 162 in FIG. 29 is the start point of the next partial stroke.

In step S71, the continuation character stroke associating means 8 uses the partial stroke extracting means 9 to create the next partial stroke. Specifically, the partial stroke extracting means 9 calculates a new partial stroke
The dividing point is traced in the reverse direction from the end point of the partial stroke, and a dividing point at which the number of candidate strokes from the starting point of the current partial stroke to the dividing point becomes 3 or less is determined, and the dividing point is set as the end point of the partial stroke. . In the example of FIG. 29, the strokes 150 and 151 that have already been determined are excluded from the candidates.
One of 3 Therefore, the division candidate point 162,
163, 164, 165 and the end point 167 of the partial stroke
All remaining partial strokes consisting of are selected.

In step S72, the continuation character stroke associating means 8 associates the obtained partial strokes with the candidate strokes. The operation of associating partial strokes will be described with reference to the flowchart in FIG. In step S80, candidates that can be associated are narrowed down by the circumscribed rectangle of the partial stroke and the circumscribed rectangle of each candidate stroke. In this example, both the candidate strokes 152 and 153 are targets of association satisfying the condition.

Next, proceeding to step S81, the continuous character stroke associating means 8 determines the stroke order. Since there are two candidate stroke numbers for the current partial stroke, and both strokes satisfy the condition of the end stroke, two stroke orders are obtained. Here, first, the correspondence in the stroke order of the strokes 152 and 153 is performed.

Proceeding to step S82, the continuation character stroke associating means 8 associates the partial strokes with the standard pattern candidate strokes. Fig. 2
0, as in step S61,
Since both 153 are linear strokes, their starting and ending points are feature points, and the correspondence with the feature points of the partial strokes is performed by dynamic programming. Here, since the starting point 162 of the current partial stroke coincides with the end point of the partial stroke up to that point, a virtual stroke component that is not actually written may exist between the strokes. Therefore, both the association starting from the division candidate point 162 and the association starting from the next division candidate point 163 are performed, and the smaller distance is determined as the final result. In this example, the association starting from the division candidate point 163 is selected.

In step S83, the continuation character stroke associating means 8 divides the partial stroke from the obtained association result, and the stroke feature extracting means 2 obtains the stroke feature. FIG. 33 is an example in which the dividing strokes 192 and 193 are obtained from the partial strokes.

Proceeding to step S84, the continuous character stroke associating means 8 checks whether the current candidate stroke has been associated with the partial stroke of the input pattern in all stroke orders. As a result of the check,
If all the combinations have not been performed, the process proceeds to step S81, and another stroke order is associated.
In this example, the process proceeds to step S81 and the stroke 15
Correspondence is performed in the stroke order of 3, 152. The associating process is the same as in the case of the strokes 152 and 153, and a description thereof will be omitted. Here, in the case of the stroke order of the strokes 152 and 153, the distance between the stroke feature point of the input pattern and the stroke feature point of the standard pattern becomes smaller.
This is adopted as a result of associating correct partial strokes.

Next, proceeding to step S73 in FIG. 28, the continuation character stroke associating means 8 determines whether or not the association of the current continuation character stroke is completed. Proceeds to step S74. In this example, since the processing has been completed, the process proceeds to step S75.

In step S75, the continuation character stroke associating means 8 obtains a continuation character stroke association result from each partial stroke association result. Here, since two strokes of the standard pattern are associated with each of the partial strokes, a stroke obtained by arranging the two in the order of the partial strokes is adopted as the final continuous stroke.

In this example, there is no uncorrelated stroke in the strokes of the standard pattern. However, if there is any uncorresponding stroke, the correspondence with the stroke obtained by connecting the continuous partial strokes is determined. The stroke is performed including the stroke not attached, and if the correspondence distance is reduced, the stroke is adopted. At this time, the number of strokes of the standard pattern inserted between each partial stroke is only one. Through the above processing, the stroke association by extracting the partial strokes from the continuous character strokes is completed, the character evaluation means 10 calculates the final character evaluation value, and the recognition result is determined.

In this embodiment, coordinate information of feature points is used for associating continuation strokes. However, strokes are sampled at regular intervals, and association is performed by using a direction code between the sample points. You may.

Also, in the present embodiment, in order to determine a stroke-corresponding area, characters are set to 4 × 1, 1 ×
Although the regions divided into 4, 4 × 4 (representing the divided regions in the horizontal direction × the vertical direction, respectively) are used, the division number may be any number.

Further, in the present embodiment, the candidate points for dividing the continuous character stroke are set as the bending points, but points sampled at regular intervals may be used.

As described above, according to the present embodiment, strokes that are not continuously written are associated only with stroke characteristics, and strokes that are not associated are associated using fine feature point information only. , And finally the evaluation is performed using only the stroke features, so that a high-speed character recognition can be performed and a recognition method that is resistant to character deformation can be realized.

In addition, not only strokes that have not been continuously written but also strokes that have been continuously written are associated with the stroke order, so that patterns in which both the number of strokes and the stroke order fluctuate can be accurately recognized. Is possible.

[0108] Strokes written extremely continuously are divided into partial strokes and associated, so that it is possible to associate in consideration of stroke order variation without increasing the processing time.

The input strokes, which are handwritten characters, are classified into linear strokes, non-linear strokes, or continuation character strokes, and are correlated in the order of linear stroke, non-linear stroke, and continuation stroke. For the standard pattern for which the matching has failed in the matching, the subsequent matching is not performed. Therefore, the matching with the standard pattern different from the correct answer is not performed, and the efficient matching process is performed. And character recognition is possible at high speed.

Further, since the correspondence of the continuous character strokes is performed at the feature points at the start point and the end point of the stroke, the correspondence can be performed at a high speed and the capacity of the stroke feature dictionary can be reduced. .

[0111]

Since the present invention is configured as described above, it has the following effects. In the first or fifth invention, strokes are classified into linear strokes, non-linear strokes, and continuation strokes based on the stroke shape of a handwritten character (input pattern) written on a tablet, and each stroke characteristic is defined as a standard pattern ( Since it is made to correspond to the stroke feature of the (character to be recognized), character recognition can be performed at high speed and with high accuracy even for an input pattern in which the stroke order and the number of strokes fluctuate.

According to the second or sixth aspect of the present invention, the existence area of each stroke of the input pattern is determined, and the stroke of the standard pattern corresponding to the existence area of the stroke is associated with the stroke of the input pattern. Can recognize characters.

According to the third or seventh invention, the correspondence between the non-linear strokes of the input pattern and the strokes of the standard pattern is performed for the standard pattern associated with the linear strokes. it can.

In the fourth or eighth invention, the correspondence between the continuation character stroke of the input pattern and the stroke of the standard pattern is performed for the standard pattern associated with the linear stroke and the non-linear stroke. , High-speed character recognition.

[Brief description of the drawings]

FIG. 1 is a configuration diagram of a character recognition device according to a first embodiment.

FIG. 2 is a flowchart showing a processing flow of the character recognition device according to the first embodiment.

FIG. 3 is a diagram showing an input pattern “A” written on the input unit 1 according to the first embodiment.

FIG. 4 is a diagram showing stroke features of an input pattern extracted by a stroke feature extraction unit 2 according to the first embodiment.

FIG. 5 is a diagram showing assignment of directions for quantizing the direction of the stroke and the direction of the virtual stroke in the stroke characteristics in the first embodiment into 16 directions.

FIG. 6 is a flowchart showing a flow of processing of a stroke classification unit according to the first embodiment.

FIG. 7 is a diagram illustrating an eight-direction component distribution used by the stroke classification unit of the first embodiment to determine a direction component distribution.

FIG. 8 is a flowchart showing a flow of processing of a linear stroke association unit 5 according to the first embodiment.

FIG. 9 is a diagram illustrating an area division used by the stroke associable area determination unit 6 according to the first embodiment when associating a linear stroke.

FIG. 10 is a diagram showing information for determining a correspondence area used by a stroke correspondence possible area determination unit 6 according to the first embodiment.

FIG. 11 is a diagram showing an area where input strokes can be associated according to the first embodiment;

FIG. 12 is a diagram showing a stroke sequence of a standard pattern of the character “A” in the stroke feature dictionary according to the first embodiment.

FIG. 13 is a diagram showing stroke features in the standard pattern of the character “A” in the stroke feature dictionary according to the first embodiment.

FIG. 14 is a diagram showing whether or not correspondence can be made according to the stroke shape used by the linear stroke associating means 5, the non-linear stroke associating means 7, and the continuation character stroke associating means 8 of the first embodiment.

FIG. 15 is a diagram showing a state of narrowing down candidate strokes in the linear stroke associating means 5 according to the first embodiment.

FIG. 16 is a flowchart illustrating a flow of processing of a non-linear stroke association unit 7 according to the first embodiment.

FIG. 17 is a flowchart showing a processing flow of a continuation character stroke association unit 8 of the first embodiment.

FIG. 18 is a diagram illustrating an area division used by the stroke associable area determining unit 6 according to the first embodiment in the case of associating other than a straight stroke.

FIG. 19 is a diagram showing where each stroke of the input pattern according to the first embodiment exists in an area used by the stroke associable area determination means 6;

FIG. 20 is a flowchart showing a detailed processing flow of processing in step S54 of the first embodiment.

FIG. 21 is a diagram showing characteristic points in a stroke of a standard pattern according to the first embodiment.

FIG. 22 is a diagram illustrating sampling points obtained by sampling strokes of the input pattern according to the first embodiment at regular intervals.

FIG. 23 shows feature points of the standard pattern of the first embodiment,
FIG. 7 is a diagram illustrating a result of associating input patterns with sampling points.

FIG. 24 is a diagram showing strokes obtained by dividing the input pattern of the first embodiment in accordance with the strokes of the standard pattern.

FIG. 25 is a diagram illustrating each stroke of a character “word” written on the input unit 1 according to the first embodiment.

FIG. 26 is a diagram showing where each stroke of the input pattern according to the first embodiment exists in an area used by the stroke associable area determination means 6;

FIG. 27 is a diagram showing a standard pattern of a character “word” in the stroke feature dictionary 3 according to the first embodiment.

FIG. 28 is a flowchart showing a detailed processing flow of associating partial strokes according to the first embodiment;

FIG. 29 is a diagram showing division candidate points of a continuation character stroke of the input pattern according to the first embodiment;

FIG. 30 is a diagram showing a partial stroke according to the first embodiment.

FIG. 31 is a flowchart showing a detailed flow of a process in step S72 of FIG. 28.

FIG. 32 is a view showing a dividing stroke obtained from the partial stroke of FIG. 30;

FIG. 33 is a diagram showing a dividing stroke obtained from a partial stroke.

FIG. 34 is a configuration diagram of a conventional online character recognition device.

[Explanation of symbols]

1 input means, 2 stroke feature extracting means, 3 stroke feature dictionary, 4 stroke classifying means, 5 linear stroke associating means, 6 stroke associable area determining means, 7 non-linear stroke associating means, 8 continuous character stroke associating Means, 9 partial stroke extraction means, 10 character evaluation means, 11 control means.

Claims (8)

[Claims] An input unit for inputting handwritten characters online; a stroke feature dictionary in which features of linear strokes and non-linear strokes constituting each of the plurality of recognition target characters are stored in advance; Stroke feature extraction means for extracting the features of the strokes constituting the handwritten characters from the handwritten characters input by the method, and converting each stroke of the handwritten characters into the straight line based on the features extracted by the stroke feature extraction means. Stroke classifying means for classifying any one of a stroke, the non-linear stroke, and a continuous character stroke which cannot be associated with a stroke constituting the recognition target character; and characteristics of the linear stroke classified by the stroke classifying means. Characters to be recognized stored in the stroke feature dictionary Linear stroke associating means for associating with the features of the constituted linear strokes; nonlinear stroke features classified by the stroke classifying means and non-linear strokes constituting recognition target characters stored in the stroke feature dictionary A non-linear stroke associating means for associating the continuous stroke classified by the stroke classifying means with the linear stroke or the non-linear stroke, and dividing the divided linear stroke or the non-linear stroke Continuation character stroke associating means for associating a characteristic of a character with a straight stroke or a non-linear stroke constituting a recognition target character stored in the stroke characteristic dictionary; Correlation means or said continuation stroke A character recognition device, comprising: character evaluation means for recognizing the handwritten character based on a result of association obtained by the association means. 2. A method according to claim 1, further comprising: a stroke-corresponding-region determining unit that determines an existing region of each stroke of the handwritten character based on the feature extracted by the stroke-characteristic extracting unit. The straight stroke associating means or the continuation character stroke associating means comprises a straight stroke constituting a recognition target character stored in the stroke feature dictionary corresponding to an area where each stroke is determined by the associable area determining means. 2. The character recognition device according to claim 1, wherein the character recognition device performs correspondence with a feature of a non-linear stroke. 3. The non-linear stroke associating means,
2. The character recognition device according to claim 1, wherein the recognition is performed on the recognition target characters associated by the linear stroke association unit. 4. The continuation character stroke associating means,
2. The character recognition apparatus according to claim 1, wherein the recognition target character associated by the linear stroke association unit or the non-linear stroke association unit is associated. 5. A character recognition apparatus having a stroke feature dictionary in which features of a linear stroke and a non-linear stroke constituting each of the plurality of recognition target characters are stored in advance. A stroke feature extraction step of extracting a feature of a stroke constituting the handwritten character from the handwritten character input in the input step; and a step of extracting the handwritten character based on the feature extracted in the stroke feature extraction step. A stroke classification step of classifying each stroke into one of the linear stroke, the non-linear stroke, and a continuation character stroke that cannot be associated with the stroke constituting the character to be recognized; Special features of straight stroke A linear stroke associating step of associating the stroke with the features of the linear strokes constituting the recognition target character stored in the stroke feature dictionary; the non-linear strokes classified by the stroke classifying step and the stroke feature dictionary A non-linear stroke associating step of associating with the features of the non-linear strokes constituting the character to be recognized stored in the stroke, the continuous stroke classified by the stroke classifying step, the linear stroke or the non-linear stroke And a continuation character stroke associating step of associating the characteristics of the divided linear stroke or non-linear stroke with the characteristics of the linear stroke or non-linear stroke constituting the recognition target character stored in the stroke characteristic dictionary. And the straight stroke Based on the matching result obtained in the matching step, the non-linear stroke matching step or the continuation character stroke matching step,
A character recognition method comprising a character evaluation step of recognizing the handwritten character. 6. A stroke associable area determining step of determining an existing area of each stroke of the handwritten character based on the feature extracted in the stroke feature extracting step, The straight stroke matching step or the continuation character stroke matching step is a straight stroke forming the recognition target character stored in the stroke feature dictionary corresponding to the existing area of each stroke determined in the matching area determining step. 6. The character recognition method according to claim 5, wherein the correspondence with the feature of the non-linear stroke is performed. 7. The character recognition method according to claim 5, wherein the non-linear stroke association step associates the recognition target characters associated by the linear stroke association step. 8. The continuation character stroke associating step, wherein the recognition target characters associated by the linear stroke associating step or the non-linear stroke associating step are associated with each other. Character recognition method.