KR100236247B1 - On-line character recognition method and apparatus - Google Patents

Abstract

It was not possible to perform high-precision character recognition at high speed, which is hardly influenced by fluctuations in the handwriting of the handwritten characters, variations in the number of strokes, and deformation of the characters of the local handwritten characters.

Input means for inputting handwriting information of characters, character shape recognition means for selecting a plurality of candidate characters with high similarity from standard characters stored in a character form librarian, based on the global shape characteristics of the handwriting character, and each stroke of the handwritten character A handwriting correspondence recognition means for performing correspondence recognition in consideration of the handwriting order for candidate characters selected by the character shape recognition means based on a feature, and a candidate selected by the character shape recognition means based on each stroke feature of the handwritten character; And a stroke number corresponding recognition means for performing correspondence recognition in consideration of the number of strokes for the characters, and display means for displaying the recognition result of the essential correspondence recognition means or the stroke number corresponding recognition means.

Description

Online Character Recognition Method and Online Character Recognition Device

The present invention relates to an online character recognition method and apparatus for accurately recognizing handwritten characters.

Conventional on-line character recognition apparatuses include an on-line character recognition apparatus for recognizing characters by absorbing fluctuations of the order and an on-line character recognition apparatus for recognizing characters by absorbing variations in stroke numbers. Hereinafter, each on-line character recognition device will be described separately.

[Prior Example 1]

In a conventional on-line character recognition device that recognizes characters by absorbing fluctuations in the order of order, a distance is obtained for each combination of strokes between all the strokes of the standard pattern in the librarian and all the strokes of the input pattern, and is applied to each stroke of the standard pattern. By correlating the strokes of the input pattern with the smallest distance, it is possible to cope with fluctuations in the input pattern order (Special Element 54-061146), or to store information such as the method of changing the order of the order in the standard pattern in advance. There is a case where the correspondence between the strokes of the input pattern and the standard pattern is restricted and the throughput of the correspondence is reduced (particularly 57-178579).

For example, FIG. 26 is a block diagram showing the configuration of a conventional on-line character recognition device that absorbs the necessary order fluctuations disclosed in Japanese Patent Application Laid-Open No. 57-178579.

In Fig. 26, reference numeral 1 denotes a pattern matching device for calculating the distance between a stroke of an input pattern and a stroke of a standard pattern, and reference numeral 2 denotes a standard pattern in which each character stores a stroke string of a standard pattern in advance. The memory device 3 denotes a combination order memory device which stores the combination order method and the like, and the reference numeral 4 indicates the standard pattern stored after the standard pattern storage device 2 based on the information of the essential order combination memory device 3. The standard pattern reconstruction apparatus 5 for changing the order of the strokes of the apparatuses 5 is an identification apparatus that determines the recognition result based on the distance output from the pattern matching apparatus 1.

Hereinafter, the operation of the conventional on-line character recognition apparatus will be described with reference to FIG. 26.

First, input pattern information X is input to the pattern matching apparatus 1. The input pattern information (X) consists of n strokes in the case of n stroke characters, and the strokes are in order of X1, X2,... … , Xn,

X = (X1, X2, ……, Xn)

Is represented by a multidimensional vector space.

Further, a standard pattern composed of n strokes stored in the standard pattern storage device 2 is Sk (k represents a category name of the standard pattern), and each stroke of Sk is Sk1, Sk2,... … , Skn, the standard pattern Sk,

Sk = (Sk1, Sk2, ……, Skn)

Is represented by a multidimensional vector space.

Next, the pattern matching device 1 calculates the distance between the input pattern X and the standard pattern Sk. However, the pattern matching device 1 cannot be limited to the fact that the input pattern X is written in the correct order. By simply matching each element of the standard pattern in order, it is not possible to correctly recognize a character written in a different order from that of the standard pattern.

Thus, the standard pattern reconstruction device 4 constructs a standard pattern Skm (m = 1, 2, ..., P) to change the order of writing based on the combination information of the writing order stored in the writing order combination storage means 3, and pattern matching. Transmit to device 1 in turn. Here, P is the total number of necessary fluctuation combinations of the standard pattern.

Next, the pattern matching apparatus 1 obtains the distance Pxkm with respect to the input pattern X with respect to the standard pattern sequentially sent from the standard pattern reconstruction apparatus 4 by the following formula | equation.

Pxkm = | X1-Skm1 | + | X2-Skm2 | +… … + | Xn-Skmn |

Next, the identification device 5 identifies the minimum distance among the P distances sent from the pattern matching device 1, and sets the minimum distance as the distance Pxk between the input pattern X and the standard pattern Sk. The above process is performed for having n standard patterns of all the categories stored in the standard pattern storage device, and finally, the category which becomes the smallest distance is output as a recognition result.

[Conventional Example 2]

In the conventional on-line character recognition apparatus which recognizes a character by absorbing stroke number variation, DP matching is performed at a plurality of feature points of an input pattern stroke and a feature point of a standard pattern stroke, and the feature points are matched so as to have the smallest distance. In order to improve the recognition accuracy of the stroke variation pattern, there is described a method of connecting strokes in standard patterns (Japanese Patent Laid-Open No. 60-79485, Japanese Patent Laid-Open No. 60-126783).

For example, Fig. 28 is a block diagram showing the configuration of a conventional on-line character recognition apparatus that absorbs the stroke number fluctuations disclosed in the special office 62-62394. Fig. 27 is a view for explaining the recognition procedure of the conventional on-line character recognition device that absorbs stroke number variation.

Fig. 27A shows a standard pattern, and Fig. 27B shows an input pattern. 27C and 27D show the correspondence between the respective strokes of the standard pattern and the respective strokes of the input pattern.

In Fig. 27, reference numerals 30 to 39 denote strokes of the input pattern, and reference numerals 10 to 21 denote strokes of the standard pattern.

In Fig. 28, reference numeral 40 denotes a character stroke detection unit, reference numeral 41 denotes a librarian address generator, reference numeral 42 denotes a librarian, reference numeral 43 denotes a feature extractor, and reference numeral 44 denotes The DP matching section, reference numeral 46 is a determining section, and reference numeral 47 is a range designating section.

Next, the operation of the conventional on-line character recognition device that absorbs stroke number variations will be described with reference to FIGS. 27 and 28.

First, the input pattern is input to the character stroke number detection unit 40 and the feature extraction unit 43 to extract the number of strokes and the feature points of each stroke. The character stroke count detection unit 40 outputs to the librarian 13 the range of character strokes (n + α) to (n-β) to be combined with respect to the number of character strokes n of the input pattern. In the case of the letter " 도시 "

As shown in Fig. 27B, since the input pattern has five strokes, there are ten feature points, and since the standard pattern has six strokes, there are 12 feature points. Therefore, if the correspondence is simply taken from the beginning, the non-corresponding feature points remain as shown in Fig. 27C, so that the stroke is not correctly performed, and the distance becomes a large value.

Therefore, in the conventional example, correspondence with different stroke numbers is performed by using DP (Dynamic Programming) matching to minimize the distance. Here, specifically, DP matching is performed as follows.

Each characteristic point of an input pattern is shown as follows.

A = a1, a2,... … , ai,… … , aI

And the feature point of the librarian's standard pattern is expressed as follows.

B = b1, b2,... … , bj,… … , bJ

At this time, the distance d (i, j) of ai, bj can be obtained as follows.

[Equation 1]

d (i, j) = | ai-bj |

(1) If i = 1, j = 1

g (1,1) = d (1,1)

(2) When i = 1, j ≠ 1

g (1, j) = g (1, j-1) + d (1, j)

(3) When i ≠ 1, j = 1

g (i, 1) = g (i-1,1) + d (i, 1)

(4) other

G (I, J) (initial condition) is calculated | required by the above ignition formula, and distance S (A, B) of the feature point sequence of A and the feature point sequence of B is calculated | required by the following formula. And it is set as the identification result that this distance is minimum.

S (A, B) = 1 / (I + J) * g (I, J)

The DP matching section 44 sequentially reads feature points of the standard pattern of the character stroke number area from the librarian 13, obtains the distance between the feature points of the input pattern and the feature points of the standard pattern by the DP matching, and determines the judging section 46. ) The determination unit 46 outputs the character of the smallest distance among the distances as a recognition result.

In this example, even with only the input pattern as shown in FIG. 27B, the feature points are correctly matched as shown in FIG. 27D, and the distance is smaller than that in FIG. 27C.

As described above, in the on-line character recognition apparatus such as the conventional example 1 which calculates the distance for all combinations of the standard pattern and the input pattern and absorbs the change in the order of order, since the distance is calculated for all the combinations, the computational amount is increased to recognize the recognition in real time. It was difficult to carry out.

In addition, in the on-line character recognition apparatus described in the conventional example 1 which describes a method of matching the order of the mandatory patterns in the standard pattern, only the fluctuation of the hand order prepared in the standard pattern can be dealt with beforehand. There was a problem that is increased.

In addition, in the apparatus corresponding to the necessary order variation of the prior art example 1, it is possible to cope with the necessary order variation, but there is a problem that the recognition accuracy of continuous characters or the like is lowered because it cannot cope with the variation in stroke count.

In addition, in the on-line character recognition apparatus like the conventional example 2 which corresponds to the number of strokes by DP matching, it is necessary to perform n × m distance calculation in order to find the distance between one standard pattern n stroke and the input pattern m stroke, This calculation needs to be performed for all standard patterns in the range of strokes from (m + α) to (n-β). Considering the case where 11 characters are input as the input pattern, m = 11, and as shown in the example, when α = 2 and β = 1, it is necessary to calculate a distance with respect to 10 to 13 standard patterns. There is. Even in the case of the first level of JIS, 10 strokes have 289 characters, so 289 × 10 × 11 = 31,790 times, and 11 strokes have 300 characters, so 300 × 11 × 11 = 36,300 times, 12 strokes. Since 297 characters exist, 297 × 12 × 11 = 39,204 times, and 13 strokes have 242 characters, so 242 × 13 × 11 = 34,606 distance calculations are required. Therefore, a total of 141,900 distance calculations are required, and the calculation amount becomes very large, making it difficult to recognize in real time.

In addition, in an online character recognition device like the conventional example 2 having information such as a joining method in a librarian, it is necessary to have information of a connection method for each stroke of each character, making it difficult to prepare a librarian, and the librarian capacity becomes very large. In addition, there is a problem that the recognition accuracy is lowered because it is not possible to cope with continuous characters not prepared in advance.

In addition, the on-line character recognition apparatus corresponding to the variation in the number of strokes according to the conventional example 2 has a problem in that it is possible to cope with the variation in the stroke number, but not to the variations in the stroke order, resulting in a decrease in the recognition accuracy of characters having different stroke orders.

In addition, since the on-line character recognition apparatuses of the above-described examples 1 and 2 actively recognize and use the information on the stroke of the character, although the overall shape of the character is clean, it is possible to select a character in which a part is continuously written or a character in a different order. It was difficult to read with high precision.

In addition, since the recognition process is performed regardless of the quality of the characters, it requires the same processing time as the dirty handwritten characters even when relatively carefully handwritten, and all the characters are recognized if the processing is complicated to improve the recognition accuracy. There was a problem that time was reduced.

SUMMARY OF THE INVENTION The present invention has been made to solve such a problem, and an object thereof is to obtain an on-line character recognition method and an on-line character recognition apparatus that can realize high-precision recognition and can recognize them at high speed.

Specifically, the first object is to perform high-speed and high-precision recognition by reducing the calculation amount of necessary order recognition recognition.

Further, the second object is to perform high speed and high precision recognition by reducing the calculation amount of stroke number corresponding recognition.

In addition, the third object is to recognize at a high speed, to cope with fluctuations in the order of maneuver and to change in the number of strokes, and to perform recognition with higher precision.

Further, even if the writing order of the written characters differs from the writing order of the standard characters, the fourth object is to accurately perform recognition.

In addition, the fifth object of the present invention is to provide high-speed recognition with high flexibility in responding to the fluctuation in the required order for characters with a high likelihood of correct answers, and to not change the required order for the characters with a low likelihood of correct answers.

In addition, when the recognition result of the character form recognition is good, the sixth object is to output the recognition result at high speed.

In addition, the seventh object is that the recognition rate is deteriorated by making a correspondence which is not possible in actual writing.

1 is a diagram showing the configuration of an on-line character recognition device according to the first embodiment.

2 is a flowchart showing the operation of the control means 59 of the first embodiment.

Fig. 3 shows an input pattern written in the input means 50 of the first embodiment.

4 is a flowchart showing the operation of the character form recognition means 52 according to the first embodiment.

Fig. 5 is a diagram showing the types of assignment direction codes of the character form recognition means 52 of the embodiment.

FIG. 6 is a diagram showing an input pattern after performing a process of assigning a direction code of the first embodiment; FIG.

Fig. 7 is a view in which the direction codes of the first embodiment are divided into areas, and direction codes are counted for each area.

Fig. 8 is a flowchart showing the operation of the necessary sequence correspondence recognition means 54 of the first embodiment.

Fig. 9 is a diagram showing the direction in the case of assigning the direction of a stroke and the direction of a virtual stroke in the essential correspondence recognition means 54 of the first embodiment.

Fig. 10 is a diagram showing the stroke characteristics of the first stroke of the input pattern in the necessary order recognition means 54 of the first embodiment.

Fig. 11 is a diagram showing a rule for correspondence of strokes in the necessary order correspondence recognition means 54 of the first embodiment.

Fig. 12 is a diagram showing a stroke string of the standard pattern of the letter “亞” in the stroke feature librarian of Embodiment 1;

Fig. 13 is a diagram showing an input pattern in which the letter “亞” of Embodiment 1 is written in different order of times;

FIG. 14 is a diagram showing a procedure of changing the order of input patterns in the order of note correspondence recognition means 54 of the first embodiment; FIG.

Fig. 15 is a diagram showing an input pattern of continuous characters for explaining the operation of the stroke number corresponding recognition means 57 of the first embodiment.

Fig. 16 shows the result of correspondence between the standard pattern and the continuous character in the stroke number correspondence recognition means 57 of the first embodiment.

Fig. 17 is a diagram showing the configuration of the on-line character recognition device of the second embodiment.

18 is a flowchart showing the flow of operation of the control means 59 of the second embodiment.

Fig. 19 is a diagram showing the configuration of the on-line character recognition device of the third embodiment.

20 is a flowchart showing the flow of operation of the control means 59 of the third embodiment.

Fig. 21 is a diagram showing the configuration of the on-line character recognition device of the fourth embodiment.

Fig. 22 is a flowchart showing the flow of operation of the control means 59 of the fourth embodiment.

FIG. 23 is a diagram showing a path where the distance between the DP matching result of the input pattern and the standard pattern in the DP matching means 130 according to the fourth embodiment is minimized; FIG.

Fig. 24 is a diagram showing a path in the case where correspondence that cannot be achieved in the actual writing of the input pattern and the standard pattern in the DP matching means 130 of the fourth embodiment is shown.

Fig. 25 shows the correspondence of the stable strokes detected by the stable stroke verification means 132 of the fourth embodiment.

Fig. 26 is a diagram showing the configuration of the on-line character recognition device of the conventional example 1.

FIG. 27 is a diagram showing an input, a pattern, a standard pattern, and corresponding results of the conventional example 2. FIG.

Fig. 28 is a diagram showing the configuration of the on-line character recognition apparatus of the conventional example 2.

* Explanation of symbols for main parts of the drawings

1: pattern matching device 2: standard pattern storage device

3: essential order combination memory device 4: standard pattern reconstruction device

5: identification device 10-17: characteristic point of standard pattern

30 ~ 39: Characteristic point of input pattern 40: Character stroke count detection unit

41: librarian address generator 42: librarian

43: feature extraction section 44: DP matching section

46: determination unit 47: range designation

50: input means 51: character form librarian

52: character form recognition means 53: stroke feature librarian

54: necessary order recognition means 55: required priority output means

57: stroke number corresponding recognition means 59: control means

60 to 66: Stroke of input pattern 70: Vertical direction code (V)

71: horizontal direction code (H) 72: right upward direction code (R)

73: Left upward direction code (L) 80 to 86: Stroke of the standard pattern

90 ~ 96: Stroke of input pattern with different order

100 to 105: Stroke of the input pattern of consecutive characters

110: character type priority output means 120: weighted mandatory correspondence recognition means

130: DP matching means 131: stroke response test means

132: stable stroke test means

Means to solve the problem

According to a first aspect of the present invention, an online character recognition method includes a handwriting information input step of inputting handwriting information according to a handwriting operation of a character, extracting a feature representing an entire shape of the handwritten character from the handwriting information, and extracting the entire shape. A character form recognition step of selecting a plurality of standard characters similar to the handwritten characters from a librarian based on the displayed characteristics, and extracting a first stroke characteristic representing a feature of a stroke constituting the handwritten character from the handwriting information; A plurality of standards selected in the character form recognition step, taking into account the stroke order of the written character corresponding to the stroke of the handwritten character and the stroke of the standard character selected in the character shape recognition step based on the one stroke characteristic; A necessary order recognition step of calculating a similarity degree of the written character of the character, And a first output step of outputting a standard character selected in the character shape recognition step according to the similarity calculated in the necessary order correspondence recognition step.

Here, the writing information input step corresponds to a step (S1 in FIG. 2) of writing a character in a tablet or the like in the embodiment to be described later. In addition, a 1st stroke characteristic is corresponded to the direction from the time of a stroke to an end point in the below-mentioned embodiment, the width | variety, the height of an external spherical shape of a stroke, etc. In FIG. The similarity with respect to the handwritten character of the plurality of standard characters selected in the character form recognition step corresponds to the distance between the standard character and the handwritten character in the embodiment to be described later.

In addition, the online character recognition method of the second invention includes a handwriting information input step of inputting handwriting information according to a handwriting operation of a character, extracting a feature representing the overall shape of the handwritten character from the handwriting information, and the overall shape. A character form recognition step of selecting a plurality of standard characters similar to the handwritten character from a librarian on the basis of a feature representing a second character, and a second stroke characteristic representing a feature of a stroke constituting the handwritten character from the handwriting information; The correspondence between the stroke of the handwritten character and the stroke of the standard character selected in the character shape recognition step is performed in consideration of the number of strokes of the handwritten character based on the second stroke feature, and is selected in the character shape recognition step. A stroke number for calculating a similarity degree to the written character of a plurality of standard characters And a second output step of outputting a standard character selected in the character shape recognition step according to the correspondence recognition step and the similarity calculated in the stroke number correspondence recognition step.

Here, the second stroke features correspond to the width, the height, and the like of the stroke in the embodiments to be described later.

In addition, the online character recognition method according to the third aspect of the present invention includes a handwriting information input step of inputting handwriting information according to a handwriting operation of a character, extracting a feature representing the overall shape of the handwritten character from the handwriting information, and the overall shape. A character form recognition step of selecting a plurality of standard characters similar to the handwritten character from a librarian on the basis of the characteristic representing the first character, and a first stroke characteristic representing a characteristic of the stroke constituting the handwritten character from the handwriting information; A plurality of items selected in the character form recognition step, taking into account the order of the written characters corresponding to the strokes of the written character and the stroke of the standard character selected in the character form recognition step based on the first stroke feature; Calculation of the similarity of the written characters of the standard characters of When it is determined that the first discrimination step and the first discrimination step determine whether or not the standard character is output by comparing the recognition step with the similarity calculated in the necessary order correspondence recognition step and a predetermined value. A first output step of outputting a standard character selected in the character shape recognition step according to the similarity calculated in the necessary order correspondence recognition step, and from the handwriting information when it is determined that it is not output in the first discrimination step. Extracting a second stroke characteristic representing a feature of the stroke constituting the written character, and matching the stroke of the written character with the stroke of the standard character selected in the character form recognition step based on the second stroke characteristic; Is carried out in consideration of the number of strokes of the written character, and is elected in the character form recognition step. A stroke number corresponding recognition step of calculating a similarity degree of the plurality of standard characters with respect to the written character, and a second output step of outputting a standard character selected by the character shape recognition step according to the similarity calculated in the stroke number correspondence recognition step; To have.

Here, the 1st determination step is corresponded to determination of the presence or absence of the output by the order of priority output means 55 in embodiment mentioned later (S4 of FIG. 2).

An online character recognition apparatus according to a fourth aspect of the present invention comprises handwriting information input means for inputting handwriting information according to a handwriting operation of a character, and extracts a feature representing the overall shape of the handwritten character from the handwriting information, thereby extracting the overall shape. Character form recognition means for selecting a plurality of standard characters similar to the handwritten character from a librarian on the basis of the characteristic to be indicated; and extracting a first stroke characteristic representing a feature of the stroke constituting the handwritten character from the handwriting information, and Matching the stroke of the handwritten character with the stroke of the standard character selected by the character shape recognition means on the basis of the first stroke characteristic in consideration of the order of the handwritten character, and elected by the character shape recognition means. Calculating a similarity degree to the written character of the plurality of standard characters According to the degree of similarity calculated in the order recognition means corresponding to the stroke orders corresponding recognition means to have a first output means for outputting the standard character chosen in the character-recognition means.

Here, the handwriting information input means corresponds to the input means 50 in the embodiment described later. In addition, 1st output means is corresponded to the display means 58 in embodiment mentioned later.

In addition, the on-line character recognition apparatus of the fifth aspect of the present invention provides handwriting information input means for inputting handwriting information according to a handwriting operation of a character, and extracts a feature representing the overall shape of the handwritten character from the handwriting information. Character form recognition means for selecting a plurality of standard characters similar to the handwritten character from a librarian based on the feature representing the shape, and a second stroke characteristic representing a feature of the stroke constituting the handwritten character from the handwriting information; And associating the stroke of the written character with the stroke of the standard character selected by the character shape recognition means based on the second stroke feature in consideration of the handwritten character and the number of strokes, and elected by the character shape recognition means. For calculating the similarity of the written characters of the plurality of standard characters According to the degree of similarity calculated in the response recognition means and the recognition means strokes corresponding to have a second output means for outputting the standard character chosen in the character-recognition means.

Here, it corresponds to the display means 58 in embodiment mentioned later of a 2nd output means.

In addition, the on-line character recognition apparatus of the sixth invention extracts a feature representing the overall shape of the handwritten character from the handwriting information input means for inputting the handwriting information according to the handwriting operation of the character, and the entirety. Character shape recognition means for selecting a plurality of standard characters similar to the handwritten character from a librarian based on the feature representing the shape, and a first stroke characteristic representing a characteristic of a stroke constituting the handwritten character from the handwriting information; And matching the stroke of the handwritten character with the stroke of the standard character selected by the character shape recognition means on the basis of the first stroke feature in consideration of the order of the handwritten character, and elected by the character shape recognition means. To calculate similarity for the written characters of the plurality of standard characters A first discriminating means for discriminating whether or not the standard character is output by comparing the corresponding recognition means with similarity calculated by the necessary order correspondence means and a predetermined value, and when it is determined that the first discriminating means is output. The first output means for outputting the standard character selected by the character shape recognition means according to the similarity calculated by the necessary order correspondence means, and the handwriting information when it is determined that the first character is not output by the first discriminating means. Extracts a second stroke characteristic representing a feature of the stroke constituting the handwritten character, and corresponds to the stroke of the handwritten character and the stroke of the standard character selected by the character shape recognition means based on the second stroke characteristic; In consideration of the number of strokes of the written character, A second output for outputting a stroke number corresponding recognition means for calculating a similarity with respect to the written characters of the plurality of standard characters outputted, and the standard characters selected by the character shape recognition means according to the similarity calculated by the stroke number corresponding recognition means; It has a means.

Embodiment 1

The on-line character recognition apparatus in this embodiment extracts candidate characters by character form recognition, performs necessary order recognition recognition on the candidate characters, and also performs stroke number recognition, which will be described below with reference to FIGS. 1 to 16. .

1 is a configuration diagram of an on-line character recognition device in this embodiment.

In Fig. 1, reference numeral 50 denotes an input means for inputting handwriting information such as a tablet, and reference numeral 51 denotes a character form in which an international shape characteristic of a character that does not depend on the number of strokes and the handwriting order for each character is previously stored. I'm a librarian.

Reference numeral 52 is a character form recognition means for recognizing by extracting a large shape feature of a character that does not depend on the number of strokes and the order of writing from the handwriting information obtained from the input means 50 and comparing it with the character form librarian 51. . The general shape features of the characters that do not depend on the number of strokes and the order of stroke are described later.

Reference numeral 53 is a stroke feature librarian which stores, in advance, the number of strokes, the stroke order, and the stroke characteristics for each stroke constituting the character for each character.

In this specification, the stroke refers to each character stroke constituting any one character.

Reference numeral 54 extracts the stroke feature of the input from the handwriting information obtained from the input means 50, and recognizes the order of correspondence recognized by matching the features of each character stored in the stroke feature librarian 53 in consideration of the order of handwriting. Means.

Reference numeral 55 is a priority order output means for determining whether or not to output the recognition result as the final recognition result based on the information of the recognition result obtained in the necessary order correspondence recognition means 54.

Reference numeral 57 denotes stroke number corresponding recognition means for recognizing that the stroke feature of the input pattern corresponds to the stroke of the standard pattern in the stroke feature librarian 53 in consideration of the stroke count.

Reference numeral 58 denotes display means for displaying the result of recognition by the necessary order recognition means 54 or the result of recognition by the stroke number corresponding recognition means 57.

Reference numeral 59 is a control means for controlling the processing operation of each of the means 50 to 57 in the character recognition apparatus, and is connected to each of the means 50 to 57. FIG.

Next, the operation of the on-line character recognition device in this embodiment will be described with reference to FIG.

FIG. 2 is a flowchart showing the flow of processing by the control means 59 of the on-line character recognition apparatus in this embodiment.

As the description in this embodiment, the character recognition procedure when the character "문자" is input will be described.

First, in S1, the control means 59 instructs the input means 50 and obtains an input pattern. The input pattern is information including the progress of the input resulting from the write operation to the input means 50. This information is referred to as handwriting information.

FIG. 3 is a diagram showing an input pattern "Y" written in the input means 50, and reference numerals 60 to 66 are stroke strings input in chronological order. The order of reference numeral 60-66 shows the order of input of the stroke to write, ie, the order of writing. In this example, since seven strokes were written, it consists of seven strokes.

Therefore, in S1, seven strokes 60 to 66 are obtained as handwriting information.

Proceeding to S2, the control means 59 sends the stroke string of the input pattern to the character shape recognition means 52, and instructs it to perform character shape recognition.

In this example, the strokes 60 to 66 of the input pattern of Fig. 3 are sent to the character shape recognition means 52, and the character shape recognition means 52 is from the strokes 60 to 66 of the input pattern. Recognition is performed by extracting international shape features of characters that do not depend on the number of strokes and the order of stroke.

The flow of S2 which is the recognition process of the character form recognition means 52 is demonstrated in detail using FIG.

4 is a flowchart showing the flow of processing by the character form recognition means 52. As shown in FIG.

First, in S11 of FIG. 4, the character form recognition means 52 assigns four types of direction codes to each of the plurality of sampled points for each stroke of the input pattern.

The four types of direction codes are four types of direction codes shown in Fig. 5, specifically, vertical (V: 70), horizontal (H: 71), upper right (R: 72), and upper left 73 directions. The code exists.

In each direction, the range from the arrow direction to the dotted line is the same direction.

A plurality of points are sampled for one stroke, and either one is assigned to the four kinds of direction codes for each point.

However, normally, the sampled points are not at regular intervals due to fluctuations in the writing speed or the sampling rate, and the points are not necessarily closely arranged.

Therefore, if the direction code is assigned to each point as it is in this state, even if the characters of the same molecular form, the number of direction codes changes depending on the writing speed. In addition, since the score varies depending on the character size, the score is normalized by normalizing to an appropriate size and interpolating between points.

A direction code is assigned to each point by determining the direction from the current point to the next point for each normalized point, and determining which direction corresponds to the direction shown in FIG. Here, the end point of the stroke is assigned the same direction as the previous point because the next point does not exist. Fig. 6 shows the results of normalizing the magnitudes and interpolating the strokes 60 to 66 of the input pattern and assigning direction codes to the points.

FIG. 6A shows the result of assigning the horizontal direction code H, and FIG. 6B shows the result of assigning the vertical direction code V. FIG. In the pattern of this example, since the direction of upper right R and left upper L does not exist, the direction is abbreviate | omitted.

Next, the flow advances to S12 in Fig. 4, where the whole character is divided into 4x4 areas, and the number of direction codes is counted for each area.

Fig. 7 is a diagram in which the entire character is divided into 4x4 areas, and the number of direction codes in each area is counted.

FIG. 7C is a result of counting the codes in the horizontal direction of FIG. 6A, and FIG. 7D is a result of counting the codes in the vertical direction in FIG. 6B.

As a result, the four-way direction code feature F extracted from the input pattern is expressed as a four-dimensional 16-dimensional vector in each direction.

F = (FH, FV, FR, FL)

FH = FH1, FH2,... , FH16

FV = FV1, FV2,... , FV16

FR = FR1, FR2,... , FR16

FL = FL1, FL2,... , FL16

FH1, FH2,... , FH16 represents the number of counts of the horizontal direction code in the 4x4 area. The same applies to FV, FR, and FL.

Subsequently, the process proceeds to S3 in Fig. 4, where the distance D between the character form librarian and the characteristic of the input pattern which stores the direction code distribution characteristic of the standard pattern of each character is calculated.

The distribution characteristic S of the direction code of a standard pattern can be expressed as follows.

S = (SH, SV, SR, SL)

SH = SH1, SH2,... , SH16

SV = SV1, SV2,... , SV16

SR = SR1, SR2,... , SR16

SL = SL1, SL2,... , SL16

In this case, the distance D between the input pattern and the standard pattern may be expressed as follows.

D = (DH, DV, DR, DL)

DH = | FH1-SH1 | + | FH2-SH2 | +… + | FH16-SH16 |

DV = | FV1-SV1 | + | FV2-SV2 | +… + | FV16-SV16 |

DR = | FR1-SR1 | + | FR2-SR2 | +… + | FR16-SR16 |

DL = | FL1-SL1 | + | FL2-SL2 | +… + | FL16-SL16 |

The above formula calculates the distance between the standard pattern and the input pattern in the character form librarian, and outputs 50 candidate letters in the order of recognition result from the smallest distance.

In the above, the processing from S11 to S13 in Fig. 4 is the recognition processing based on the global shape feature of the characters that do not depend on the number of strokes and the order of the strokes.

After fifty candidate characters are selected, they are moved to S3 in FIG.

Subsequently, the flow advances to S3 in FIG. 2, and the control means 59 sends the 50 rear characters obtained by the character shape recognition means 52 to the handwriting correspondence recognition means 54, and instructs to perform handwriting correspondence recognition.

The operation of the necessity corresponding recognition means 54 will be described in detail with reference to FIG.

8 is a flowchart showing the operation of the necessary order correspondence recognition means 54.

First, the necessary sequence recognition means 54 proceeds to S20 in FIG. 8 to extract the stroke feature of each stroke from the input pattern.

The stroke features include the shape characteristics of the stroke for each stroke, the direction from the start of the stroke to the end point, the width of the circumscribed rectangle of the stroke, the height of the circumscribed rectangle of the stroke, the direction of the imaginary stroke, the width of the circumscribed rectangle of the virtual stroke, Seven kinds of heights of the circumscribed rectangle of the virtual stroke are extracted.

The circumscribed sphere of the stroke refers to a sphere circumscribed with respect to the stroke.

The virtual stroke is a straight line connecting the end point of the current stroke and the start point of the next stroke. In the case of the final stroke, the next stroke is the first stroke. The circumscribed spherical shape of the virtual stroke is a spherical shape that makes the virtual stroke diagonal.

Further, the square of each stroke is quantized in eight directions as shown in FIG. The width and height are values obtained by normalizing the width and height of the circumscribed rectangle of the entire character as 100, respectively. Fig. 10 shows an example in which seven types of stroke characteristics are extracted from the stroke 60 of the first stroke of the input pattern.

Next, the combination of the stroke of the input pattern and the standard pattern in the stroke feature librarian of the 50 candidate characters obtained by the character form recognition means 52 is performed. If the number of strokes of the pattern and the number of strokes of the standard pattern do not match, it cannot be matched.

Here, since the character form recognition means 52 does not use the information on the number of strokes, the characters of the various strokes are contained in the candidate characters. Therefore, first in S21, a combination of the number of strokes of the input pattern and the standard pattern is performed. If the number of strokes does not coincide, it becomes "NO" in S21, so that the correspondence with the standard pattern of the current candidate character is not performed, and the correspondence with the next character is performed. If it matches in S21, it becomes "YES" and advances to S22.

In S22, correspondence using the stroke feature of the standard pattern of the stroke feature librarian 53 and the stroke feature of the input pattern is performed. Here, as shown in Fig. 12, it is assumed that the stroke features are arranged in the correct order of writing or the order of writing by many people. In addition, the determination of whether or not the strokes correspond to each other is performed by referring to the corresponding rule described later.

An example of a rule corresponding to FIG. 11 is shown. In the example, the characteristic to test for each shape of a stroke is described, and the input pattern and a standard pattern are tested with respect to the characteristic. For example, in the corresponding test of the horizontal lines, the test is performed only for the width of the stroke, which is an important feature of the horizontal line strokes.

Here, the test regarding the direction determines that the case where the difference in the direction is ± 2 or more does not correspond,

If the difference in width and height is 70 or more for each of the width and height of the circumscribed rectangle of the entire character, it is determined that the test is not compatible with the width and length. However, in the case where the strokes correspond to the short strokes rather than the same standard, the direction is not more stable than that in the case of the long stroke, so that the corresponding allowable range is widened.

In addition, the direction, the width, and the height of the virtual stroke are used as a measure of the positional relationship between the strokes. Therefore, the test is performed regardless of the shape of the stroke. Here, the above correspondence is performed while the correspondence is successful, and when the correspondence fails or all the strokes correspond, the flow advances to S23.

Next, in S23, it is checked whether all the strokes of the strokes of the input pattern have succeeded. If all correspondences are made, the process proceeds to S26, and if it fails in the middle, the process proceeds to S24. In this example, since the input patterns are written in the correct order, all strokes correspond to S26 in correspondence.

Next, in S26, the distance between the stroke feature of the input pattern and the stroke feature of the standard pattern is normalized for each feature for every stroke, and then the distance of each feature is weighted to find the final distance from the current standard pattern. The weight is set in advance to increase the recognition rate, but the weight may be changed by the number of strokes or the like.

However, the stroke shape is not used for distance calculation.

Here, the distance Dd after normalization of the stroke direction, the distance Dw after normalization of the stroke width, the distance Dh after normalization of the stroke height, the distance Dvw after normalization of the virtual stroke width, the distance Dvh after normalization of the virtual stroke height, and the normalization of the virtual stroke direction. From later distance Dvd, distance D with standard pattern

D = (Wd × Dd + Ww × Dw1 + Wh × Dh + Wvw × Dvw + Wvh × Dvh + Wvd × Dvd). Here, Wd represents the weight in the stroke direction, Ww represents the weight of the stroke width, Wh represents the weight of the stroke height, Wvw represents the weight of the virtual stroke width, Wvh represents the weight of the virtual stroke height, and Wvd represents the weight in the virtual stroke direction. These relationships can be obtained as follows.

Wd + Ww + Wh + Wvw + Wvh + Wvd = 1

This weight is set in advance to increase the recognition rate. The weight may be set in accordance with the number of strokes of the characters to be recognized.

Next, the operation in the case of failing to correspond in S23 will be described. It is assumed that the letter "문자" is written in a different order from the standard pattern as shown in FIG. In this case, although the second stroke 91 of the input pattern and the second stroke 81 of the standard pattern are the same vertical bars, it is determined that the stroke height test does not correspond. Therefore, the process proceeds to S24 at the end of the correspondence to the second stroke.

Next, in S24, it is determined that the order of the input patterns is different, and the strokes of the input patterns which have failed to be corresponded to the strokes of the input patterns which have not yet been used for correspondence (assuming that they correspond to the kth stroke of the current standard pattern) Of the strokes other than the input strokes used for the k-th stroke up to the k-th stroke) are not replaced by the k-th stroke).

However, if only the last stroke does not correspond, or if the unused stroke does not exist with respect to the stroke of the standard pattern currently being applied, it is judged that there is an error in the correspondence up to that time, and corresponding to the previous stroke. Unused input strokes are made to be corresponded.

In this example, the non-corresponding stroke 91 is replaced with the next stroke 92 as shown in Fig. 14A.

Subsequently, the procedure proceeds to S25 and the number of retry attempts (mandatory replacement frequency) of the input strokes up to now is compared with the predetermined upper limit number of retry attempts, and when it does not exceed, it returns to NO and returns to S22. The correspondence resumes from the stroke of the standard pattern which did not correspond.

In S22, the rearranged input stroke 92 does not correspond to the second stroke 81 of the standard pattern, and as described above, the second stroke of the standard pattern fails to proceed to S24.

In S24, the next unused input stroke 93 is brought to the second stroke, but the input strokes 91 and 92 which fail to correspond to the second stroke of the standard pattern return in the order of the original input pattern.

Fig. 14B shows the input stroke string after the movement.

14C to 14F show the necessary replacement procedure until the above process is repeated to finally correspond to all strokes. In this example, the response succeeds with six retries, and the procedure proceeds to S26 as in the case of writing in the correct order, and the distance to the standard pattern is calculated. In the input stroke whose handwriting is changed in accordance with the replacement of the handwriting, the directions, widths, and heights of the virtual strokes indicating the positional relationship between the strokes are recalculated.

When the corresponding is finished for all the recognition candidate characters obtained by the character form recognition means 52, the process proceeds to S4 of FIG.

In S4, the control means 59 sends the distance of the character which became the minimum distance among the characters with which the correspondence succeeded by the correspondence order recognition means 54 to the priority order output means 55. As shown in FIG.

The priority order output means 55 compares the distance of the character which succeeded in correspondence with a predetermined threshold, and determines that the result is output when the distance is shorter than the threshold, and does not output the result when the distance is larger than the threshold. Determine.

When the control means 59 determines that the priority order output means 55 outputs, among the 50 characters selected by the character recognition means 52, the characters of the short distance calculated by the necessary order correspondence means 54 are sequentially selected. The predetermined number is sent to the display means 58, and the display means 58 displays characters.

In the case of the input pattern of Fig. 3 and the input pattern of Fig. 13, since the correspondence of the necessary order is successful at the same time, the result distance is also sufficiently smaller than the threshold value, and the priority order output means 55 determines that the output is necessary, and the necessary order correspondence recognition means. The recognition result at 54 is finally output.

If it is determined that the output means 55 does not output first, the flow advances to S5 of FIG.

Next, in order to explain the operation of S5 in FIG. 2, the processing in the case of the pattern in which the letter "亞" is written in six strokes as shown in FIG. 15 is considered.

The input pattern of FIG. 15 is written in six strokes, and since the stroke feature librarian 53 has only seven standard patterns of "k", the necessary order recognition means 54 compares with the input pattern of FIG. Response is impossible.

Therefore, even if the six-stroke characters succeed, for example, the distance from the input pattern becomes very large because they correspond to the other characters, and the output means 55 does not become a value smaller than the threshold value. It is determined that no output is made, and the process proceeds to S5.

In S5, the stroke count correspondence recognition means 57 performs matching of the strokes having different stroke numbers to the minimum of 50 characters selected by the character recognition means 52 by DP matching as in the conventional example 2. The stroke of the input pattern is shown as follows.

A = a1, a2,... , ai,… , aI (I is the number of strokes in the input pattern)

The standard pattern stroke of the librarian is indicated as follows.

B = b1, b2,... , bj,… , bJ (J is the number of strokes in the standard pattern)

The distance d (i, j) of ai, bj can be calculated as follows.

d (i, j) = | Wai-Wbj | + | Hai-Hbj |

Here, W represents the width of the stroke and H represents the height of the stroke.

Equation 2

If d (i, j) = | a1-bj |

(1) If i = 1, j = 1

g (1,1) = d (1,1)

(2) When i = 1, j ≠ 1

g (1, j) = g (1, j-1) + d (1, j)

(3) i = 1, j = 1

g (i, 1) = g (i-1,1) + d (i, 1)

(4) other

G (I, J) (initial condition) is calculated | required by the ignition formula shown in the following, and distance S (A, B) of the stroke row of A and the stroke row of B is calculated | required by the following formula.

S (A, B) = 1 / (I + J) * g (I, J)

The DP matching is performed on the candidate characters of the fifty recognition results obtained by the character form recognition means 52, and the distances S (A, B) are output to the display means 58 as the recognition results in order from the least.

In the display means 58, characters are displayed in order starting from the character whose distance calculated by DP matching is small.

In this example, by DP matching, the input pattern of FIG. 15 is matched with the stroke sequence of the standard pattern of FIG. 12 as in FIG.

In addition, in the present embodiment, the direction code feature is used as the feature of the character presence recognition means 52. However, the peripheral distribution feature, the line density of the stroke, etc., which are characterized by the area from the circumscribed rectangle of the character to the stroke of the character. It may be a feature that reflects the overall character form of the character.

In addition, in this embodiment, although the area | region division was divided into 4x4 as a direction code characteristic, you may change the area division number according to a librarian capacity and identification capability.

In the present embodiment, 50 is fixed as the number of candidate characters output from the character form recognition means 52, but this value may be changed in accordance with the identification capability of the character form recognition means 52.

In the present embodiment, the same number of candidate characters are output from the character form recognition means 52 to the corresponding recognition means 54 and the stroke number correspondence recognition means 57 in response to the change in order of the necessary order. The number of candidate characters to be output may be changed.

The effect of the on-line character recognition apparatus in this embodiment is stated.

As described above, according to the present embodiment, only the candidate characters of the result recognized by the character of the international character are recognized according to the order of handwriting and the recognition according to the number of strokes. Since no correspondence using the detailed features is performed, high-speed and high-precision recognition can be realized.

Specifically, when handwriting characters are recognized by characters, up to some characters of the second level of JIS, it is necessary to recognize about 4000 kinds of characters at high speed. If it responds to the number of strokes, it is difficult to make practical processing time. Therefore, in this embodiment, the character recognition by the character shape recognition means 52 is the first process, and the character which performs the essential order recognition and stroke number correspondence recognition performed in the next step using the approximate broad feature of a character is made into high speed. To compress.

In addition, since characters similar in character form are output as recognition candidate characters, the impression given to the user at the time of misrecognition is improved, and the ratio of correct answer characters in the recognition candidate characters is also improved.

In addition, since the correspondence order recognition means 54 precisely performs stroke correspondence with a plurality of features, it is possible to accurately recognize characters written according to the number of strokes, and the number of strokes in which a standard pattern such as a continuous character does not exist is changed. This generated input pattern can be sent accurately to the stroke counterpart recognition means 57, so that continuous characters can also be recognized with high accuracy.

As the character display in this embodiment, there may be a case where the recognition result by the priority order output means 55 is displayed, and the recognition result by the stroke correspondence recognition means 57 may be displayed.

Therefore, the output to the display means 58 is comprised in two steps. Therefore, when it is determined that output from the priority order output means 55 is determined, the display time can be displayed at an early stage, so that the recognition time can be increased.

In this embodiment, character type recognition is performed first, mandatory correspondence recognition is performed second, and stroke number correspondence recognition is performed third. However, if the priority order is more important than the continuation character, the continuation character is more important than the essential order. For example, the recognition order after the second may be changed according to the situation.

Therefore, the following pattern can be considered as a character recognition order.

A: Recognition of character form → stroke recognition → stroke order recognition

B: character type recognition → mandatory correspondence recognition

C: character type recognition → stroke number recognition

Embodiment 2

The online character recognition device in this embodiment determines whether or not to output candidate characters extracted by character shape recognition, and performs necessary order recognition and stroke number recognition when it is determined that the character is not output. 17 and FIG. 18 will be described.

In Fig. 17, reference numerals 50 to 59 are the same as those in the first embodiment, and the description thereof is omitted.

Reference numeral 110 is character type priority output means for determining whether or not to output the recognition candidate character of the character type recognition means 52 based on the information of the recognition candidate character obtained in the character shape recognition means 52.

Next, the operation of the on-line character recognition apparatus according to this embodiment will be described with reference to FIG. 18.

Fig. 18 is a flowchart showing the flow of processing by the control means 59 of the on-line character recognition device in this embodiment.

The processing up to S2 is the same as that in the first embodiment, and the description thereof is omitted.

By performing the processing of S1 and S2, the control means 59 sends the character of the minimum distance among the 50 recognition candidate characters obtained by the character form recognition means 52 to the character form priority output means 110. FIG.

Next, the flow advances to S30 in FIG. 18 to determine whether the character type priority output means 110 outputs the recognition result of the character type recognition means 52. The determination finds the similarity (eg, normalization distance) between the input pattern and the standard pattern with respect to the character of the minimum distance sent by the character shape recognition means 52, and returns the result if the similarity is larger than the threshold prepared for each character in advance. It is determined to output, and when it is small, it is determined not to output the result.

If the vector of the input pattern is f and the vector of the standard pattern is g, the similarity s (f, g) is obtained by the following equation.

s (f, g) = (f, g) / ∥ f∥-∥ g∥

Here, the threshold value of the letter A is obtained by obtaining the similarity by the direction code feature between the standard pattern in the letter form librarian 51 of the letter A and the handwriting pattern other than the letter A in the handwriting pattern including all the character patterns to be recognized. A large similarity (the similarity of the pattern other than the letter A closest to the standard pattern of the letter A) is set. As a result, the threshold value of the letter A in which there are no letters having similar letter shapes becomes a small value, and the ratio at which the result of the letter shape recognition means 52 is output as it is is increased.

In addition, since characters having characters of similar character forms exist, the threshold value is increased, so that the output ratio is small, and the recognition by detailed stroke characteristics can be performed.

When it is determined in S30 that the character type priority output means 110 does not output the result, the 50 recognition candidate characters obtained by the character type recognition means 52 are displayed in order from the smallest distance calculated in S2.

When it is determined in S30 that the character type priority output means 110 does not output the result, the process after S3 is performed similarly to the first embodiment on the fifty recognition candidate characters obtained by the character shape recognition means 52.

In addition, in this embodiment, although the normalization distance was used as the similarity degree for the output determination by the character-form priority output means 110, you may use a Euclidean distance, a Mahalanobis distance, etc.

The effect of the on-line character recognition apparatus in this embodiment is demonstrated.

In the online character recognition apparatus of this embodiment, in the case of a character in which characters having similar character shapes do not exist, the rate at which the recognition result is output by the character shape recognition means 52 is increased, and the order of change, stroke number variation, and locality of the characters are increased. It can be recognized with high precision without being influenced by general deformation. Further, when the recognition result is output from the character form recognition means 52, the recognition can be performed at a very high speed.

Further, according to this embodiment, high precision can be obtained by the character shape recognition means with respect to the carefully handwritten characters, and since they are displayed at high speed without the necessary correspondence recognition recognition, the case of normal human recognition Similarly, polite characters can be recognized early, improving the user's impression.

In the online character recognition device according to the present embodiment, when the result of the first character type recognition is highly reliable, the recognition result is displayed at this stage. Therefore, a character with respectfully handwritten characters can be recognized at high speed. Done.

Embodiment 3

The on-line character recognition device in this embodiment sets a large number of retries of necessary correspondence recognition for a character having a high probability of correct answer character among candidate characters extracted by character shape recognition, which will be described below with reference to FIGS. 19 to 20. do.

In Fig. 19, reference numerals 50 to 59 and 110 are the same as in the above embodiment, and thus description thereof is omitted. Reference numeral 120 denotes a corresponding, weighted mandatory correspondence recognition means for performing correspondence between the stroke of the input pattern and the standard pattern.

The operation of the online character recognition apparatus in this embodiment will be described with reference to FIG.

20 is a flowchart showing the flow of processing by the control means 59 of the on-line character recognition apparatus in this embodiment.

The processing from Fig. 20 to S30 is the same as that of the second embodiment, and the description thereof is omitted.

The process of S1, S2 S30 is performed, and it progresses to S40. In S40, the control means 59 sends the 50 candidate characters obtained in the character shape recognition means 520 to the weighted handwriting correspondence recognition means 12, as in the first embodiment, and the weighted handwriting correspondence recognition means 120 receives the input pattern and the standard. The strokes of the patterns are matched and recognized.

The operations corresponding to the detailed strokes are the same as those in the first embodiment of FIG. 8, but in the first embodiment, the number of reordering times (retry times) of the necessary order is obtained for the candidate characters obtained by the character shape recognition means 52 in the first embodiment. Compared with the above, in this embodiment, the candidate character of the character type recognition means 52 is changed by the rank (the rank which descended in order from small distance).

Specifically, the higher the number of letters among the candidate letters in the recognition result of the character form recognition means 52, that is, the letters that are more likely to be correct letters, increases the threshold of the number of retries, and the lower the number of letters, that is, the more correct letters. The lower the letter, the smaller the threshold number of retries.

In addition, according to this embodiment, since the number of necessary order rearrangements is reduced for the characters with a low probability of correct answer and no correspondence is necessary, it is possible to recognize them at a high speed, and the required order of characters with a lower probability of correct answer can be recognized. Misunderstanding due to rearrangement can be reduced.

Embodiment 4

When the on-line character recognition device in this embodiment recognizes stroke number correspondence, the on-line character recognition apparatus extracts a stable point corresponding to the input pattern and the standard pattern in a one-to-one correspondence, and performs each stroke to the stable point. A description will be given based on FIG.

In Fig. 21, reference numerals 50 to 59 are the first embodiment, reference numeral 110 is the second embodiment, and reference numeral 120 is the same as the third embodiment, and the description thereof is omitted.

Reference numeral 130 denotes a DP matching means, reference numeral 131 denotes a verification means corresponding to a stroke, and reference numeral 132 denotes a stable stroke corresponding means.

The operation of the on-line character recognition device in this embodiment will be described based on FIG. 22.

Fig. 22 is a flowchart showing the flow of processing by the control means 59 of the on-line character recognition apparatus in this embodiment.

Operations from Fig. 22 to S4 are the same as in the first to third embodiments described above, so explanation is omitted and the operation after S50 will be described.

First, when it is determined in S4 that the priority order output means 55 does not output, the control means 59 outputs all 50 candidate characters of the recognition result obtained by the character shape recognition means 52 to the DP matching means 130. Send to.

In S50, the control means 59 instructs the DP matching means 130 to perform correspondence between the input pattern and the standard pattern with respect to 50 candidate characters of the recognition result obtained by the character shape recognition means 52.

Here, the test means 131 corresponding to the stroke tests the path corresponding by the DP matching means 130, and penalizes for the non-existent correspondence to prevent the selection of that path.

The correspondence between the input pattern of FIG. 15 and the standard pattern of FIG. 12 will be described as an example for the corresponding path by the DP matching means 130. FIG. The input pattern of FIG. 15 corresponds to the standard pattern of FIG. 12 as shown in FIG. 16, but FIG. 23 shows the corresponding path in two dimensions.

In Fig. 23, the horizontal axis represents the stroke sequence of the standard pattern, and the vertical axis represents the stroke sequence of the input pattern. The thick line in the figure is the corresponding path for DP matching.

Although it does not have a problem when it responds correctly like FIG. 24, in some cases, it responds like FIG.

FIG. 24 is a diagram showing a path in the case where correspondence that cannot be achieved in the actual writing of the input pattern and the standard pattern in the DP matching means 130 is performed.

In the case of Fig. 24, the third stroke of the input pattern corresponds to the third, fourth, and fifth strokes of the standard pattern. Such a corresponding state is called a combination of strokes, and shows a state in which the strokes to be originally described in two strokes are continuously described in one stroke.

The fifth stroke of the standard pattern corresponds to the third and fourth strokes of the input pattern. Such a state is called separation of strokes, and a state in which a stroke originally to be described in one stroke is divided into two strokes is described.

In other words, the stroke of the fifth stroke of the standard pattern occurs at the same time as the strokes are combined and separated. Since such a phenomenon does not occur in the writing of ordinary characters, such an inconsistent correspondence is prevented by the stroke correspondence verification means 131.

That is, the stroke correspondence test means 131 performs the DP matching means 130 with respect to the correspondence which cannot be achieved in normal writing, such as a path that goes from side to side and a path that goes from side to side when the DP matching is performed. Adding a sufficiently large penalty to the distance value when the route is selected prevents the route from being selected.

Next, the flow advances to S51, and the control means 59 sends the route information obtained by the DP matching means 130 to the stable stroke verification means 132. The route information is information expressed by lines corresponding to the respective strokes shown in FIGS. 23 to 25.

The stable stroke verification means 132 extracts the point as a stable point when the front and rear paths are the slope paths at each point on the path. However, at the start of the path, if the path is inclined immediately, the path end point is set as a stable point in the case of the path where the previous path is inclined.

Fig. 25 is a diagram showing the stable point extracted by the stable stroke verification means 132;

In Fig. 25, the stable point extracted by the stable stroke extracting means 132 in the path corresponding to the input pattern of Fig. 15 and the standard pattern of Fig. 12 is indicated by ○. In this example, five stable points are extracted.

Next, the stable stroke verification means 132 assumes that the stroke of the input pattern and the standard pattern has been performed in a one-to-one correspondence with respect to the extracted stable point. Test against the stroke.

If it is determined that at least one of the strokes of the stable point does not correspond to this correspondence, it is determined that the correspondence in the DP matching has failed, and the process proceeds to the correspondence of the candidate character of the next character form recognition means 52.

In the example of Fig. 25, the verification of the correspondence with the stroke of the standard pattern is performed for the strokes other than the combined stroke 102 among the strokes of the input pattern.

The above process is performed for all 50 candidate characters of the character form recognition means 52, and a predetermined number is output as the final recognition result in order from the character with the smallest distance among the characters which have successfully corresponded.

The effect of the on-line character recognition apparatus in this embodiment is stated.

As described above, according to this embodiment, since the path is controlled so that the correspondence at the time of DP matching does not occur in actual writing, high-precision recognition can be realized.

In addition, according to this embodiment, the strokes corresponding to one-to-one correspondence as a result of DP matching are tested to match the strokes with a plurality of features, so that incorrect correspondence can be prevented and high precision recognition can be realized.

Since this invention is comprised as mentioned above, it shows the effect shown below.

Since the first invention performs necessary order correspondence recognition on the standard characters selected in the character form recognition step, the number of standard characters that perform the necessary order correspondence recognition can be reduced, so that high speed and high accuracy can be recognized.

In the second invention, stroke number recognition is performed on the standard characters selected in the character shape recognition step, so that the number of standard characters for stroke number recognition can be reduced, so that high-speed and high precision recognition is possible.

The third invention compares the similarity calculated in the mandatory correspondence recognition step with a predetermined value, and when it is determined that the first determination step determines whether or not to output a standard character, and outputs in the first determination step, Since it has a first output step for outputting the standard character selected in the character shape recognition step according to the similarity calculated in the necessary order correspondence recognition step, if the recognition result in the essential order correspondence recognition step is good, the recognition result is output at high speed. can do.

Since the fourth invention makes correspondence between the strokes of the written characters and the strokes of the standard characters in the order of the written characters, and if the correspondence is not possible, the fourth stroke changes the order of the corresponding strokes so that the correspondence is performed again. Even if the writing order of the written characters differs from the writing order of the standard characters, the recognition can be performed with high accuracy.

Since the fifth invention performs the number of correspondences in which the order of strokes in the necessary order correspondence recognition step is changed according to the similarity calculated in the character form recognition step, it is possible to flexibly respond to the required order fluctuation for the characters with high probability of correct answers. For a character with a low probability of correct answer, it is possible to avoid changing the order of the mandatory, so that high-speed and high-precision recognition is possible.

The sixth invention determines whether or not to output the standard character by comparing the similarity between the standard character selected in the character form recognition step and the predetermined value after the character shape recognition step. A second discrimination step,

In the case where it is determined that the output is performed in the second discrimination step, since the third output step outputs the standard character according to the similarity calculated in the character shape recognition step, the recognition result of the character shape recognition step is good. The recognition result can be output at high speed.

The seventh invention extracts a stable stroke in which a stroke of a written character and a stroke of a standard character correspond one-to-one from a corresponding stroke in a DP matching step, and extracts the stable stroke of the standard character with respect to the written character for the stable stroke. Since it has a stable stroke detection step of calculating the similarity degree, it is possible to prevent the recognition rate from deteriorating because there is a correspondence that cannot be found in actual writing.

In the eighth invention, since the necessary correspondence recognition is performed on the standard characters selected by the character shape recognition means, the number of standard characters that perform the necessary correspondence recognition can be reduced, so that high-speed and high-precision recognition is possible.

In the ninth invention, stroke number recognition is performed on the standard characters selected by the character shape recognition means, so that the number of standard characters for stroke number recognition can be reduced, so that high-speed and high precision recognition is possible.

The tenth invention compares the similarity calculated by the mandatory correspondence means with a predetermined value, whereby the first discriminating means for discriminating whether or not to output standard characters and the first discriminating means for outputting is determined, Since it has a first output means for outputting the standard character selected by the character shape recognition means according to the similarity calculated by the necessary order correspondence recognition means, the recognition result is output at high speed when the recognition result in the essential order correspondence recognition means is good. can do.

The eleventh invention extracts a stable stroke in which a stroke of a written character and a stroke of a standard character correspond one-to-one from a stroke corresponding to a DP matching means, and the standard character for the written character with respect to the stable stroke is extracted. Since it has a stable stroke detection means for calculating the similarity of, it is possible to prevent the recognition rate from deteriorating due to the existence of a correspondence which is not possible in actual writing.

The twelfth invention includes a second discrimination means for determining whether to output the standard character by comparing the similarity of the standard character selected by the character recognition means with respect to the written character and a preset value;

In the case where it is determined that the output is made by the second discriminating means, since the third output means outputs the standard character according to the similarity calculated by the character shape recognition means, the recognition result of the character shape recognition means is good. The recognition result can be output at high speed.

Claims (6)

(Correction) a handwriting information input step of inputting the handwriting information in accordance with a writing operation of the character; A character shape recognition step of extracting a feature representing the overall shape of the written character from the handwriting information, and selecting a plurality of standard characters similar to the written character from a librarian based on the feature representing the overall shape; Extracting a first stroke feature representing a feature of the stroke constituting the written character from the handwriting information, and extracting the stroke of the written character and the standard character selected in the character form recognition step based on the first stroke feature; A correspondence order recognition step of matching strokes in consideration of the handwriting order of the handwritten characters, and calculating a similarity degree to the handwritten characters of a plurality of standard characters selected in the character shape recognition step; A first judging step of judging whether or not to output the standard character by comparing the similarity calculated in the necessary sequence correspondence recognition step with a predetermined value; A first output step of outputting a standard character selected in the character form recognition step in accordance with the similarity calculated in the necessary order correspondence recognition step when it is determined that the output is made in the first determination step; In the case where it is determined that the output is not performed in the first determination step, a second stroke feature representing a feature of the stroke constituting the written character is extracted from the handwriting information, and the written character is based on the second stroke feature. The correspondence between the strokes of the letter and the stroke of the standard character selected in the character shape recognition step is performed in consideration of the number of strokes of the written character, and the similarity of the plurality of standard characters selected in the character shape recognition step with respect to the handwritten character. A stroke number corresponding recognition step of calculating a value; And a second output step of outputting a standard character selected in the character shape recognition step in accordance with the similarity calculated in the stroke number correspondence recognition step. Handwriting information input means for inputting the handwriting information according to a writing operation of a character; Character shape recognition means for extracting a feature representing the overall shape of the handwritten character from the handwriting information and selecting a plurality of standard characters similar to the handwritten character from a librarian based on the feature representing the overall shape; Extracting a first stroke feature representing a feature of the stroke constituting the written character from the handwriting information, and extracting the stroke of the written character and the standard character selected in the character form recognition step based on the first stroke feature; Matching order of strokes in consideration of the handwriting order of the written characters, and calculating the similarity order of the plurality of standard characters selected by the character shape recognition means to the similarity of the written characters; First discriminating means for discriminating whether or not to output the standard character by comparing the similarity calculated in the necessary order correspondence means with a predetermined value; First output means for outputting standard characters selected by said character form recognition means in accordance with said similarity calculated by said necessary order correspondence means when it is determined that the output is made by said first discrimination means; In the case where it is determined that the first discriminating means does not output, a second stroke feature indicating a feature of the stroke constituting the written character is extracted from the writing information, and the written character is based on the second stroke feature. The correspondence between the strokes of the letter and the stroke of the standard character elected by the letter shape recognition means is performed in consideration of the number of strokes of the written character, and the similarity of the plurality of standard letters elected by the letter shape recognition means to the handwritten character. Stroke number corresponding recognition means for calculating a; And second output means for outputting standard characters selected by said character shape recognition means in accordance with said similarity calculated by said stroke correspondence recognition means. (Newly formed) The character form recognition step according to claim 1, wherein the character form recognition step calculates a similarity degree of the selected standard character with respect to the handwritten character, and the mandatory correspondence recognition step changes the order of the strokes into the character form. On-line character recognition method characterized in that the number of times according to the similarity of the standard character calculated in the recognition step. (New) The method according to claim 1, wherein after the character form recognition step, whether or not to output the standard character by comparing the similarity and a predetermined value with the standard character calculated in the character form recognition step for the written character. A second discrimination step of determining whether or not there is; And a third output step of outputting the standard character selected in the character shape recognition step according to the similarity calculated in the character shape recognition step, when it is determined that the output is made in the second determination step. And if it is determined that no output is made in the judging step, the necessary correspondence recognition step is executed. (New) The second stroke according to claim 1, wherein the stroke number correspondence recognition step is composed of a DP matching step and a similarity calculation step, wherein the DP matching step is a second stroke indicating the characteristics of the stroke constituting the handwritten character from the handwriting information. Extracting a feature and matching the stroke of the written character with the stroke of the standard character selected in the character shape recognition step based on the second stroke feature is performed by DP matching, and the similarity calculating step is performed by the DP matching. In the step, a stable stroke in which the stroke of the written character and the stroke of the standard character correspond one-to-one is extracted from the corresponding stroke, and the written character of the selected standard character is selected in the character shape recognition step with respect to the stable stroke. Calculate a similarity with respect to a character, and the second output step is similar to the similarity The dependence on the degree of similarity, the online character recognition method, characterized in that it has for outputting the standard character chosen in the character-recognition step calculated by the calculating step. The display apparatus according to claim 2, further comprising: second discrimination means for discriminating whether or not to output the standard character by comparing a similarity of the standard character calculated by the character shape recognition means with respect to the handwritten character and a predetermined value; And third output means for outputting the standard characters selected by the character shape recognition means in accordance with the similarity calculated by the character shape recognition means when the output is determined by the second discrimination means. And the recognizing means carries out the processing when it is determined that the second discriminating means does not output.