JPS6215681A - Recognizing system for on-line handwritten character - Google Patents

Abstract

PURPOSE:To decrease the cases of misrecognition among similar characters without increasing both the arithmetic quantity and the memory capacity, by selecting a standard character pattern having rationality in the form of Roman characters. CONSTITUTION:A CPU 3 selects a standard character pattern which produces the minimum difference and a difference with which the difference from and the ratio to said minimum difference is kept within the prescribed threshold value out of the degrees of difference between the input characters stored in a memory 4 and all standard character patterns. Then only a standard character pattern having rationality as a Roman character sentence is selected out of those selected standard character patterns in consideration of the result of recognition of said input characters. In case just a single pattern is selected, this pattern is delivered as the result of recognition to the input character. While the pattern having the minimum difference from the input character is delivered as the result of recognition when plural patterns are selected.

Description

[Detailed Description of the Invention] [Field of Application of the Invention] The present invention is an online handwritten character recognition system that sequentially inputs handwriting information of handwritten characters and immediately recognizes what character the input characters are based on this information. Regarding the method.

[Background of the invention]

Online handwritten character recognition will be explained with reference to FIG.

FIG. 1 is a block diagram showing an example of an online handwritten character recognition system, in which 1 is a writing device, 2 is a first interface circuit, 3 is a CPU, 4 is a first memory, and 5 is a second
6 is a second interface circuit, and 7 is an output terminal.

In the figure, the writing instrument 1 sequentially outputs information such as the handwriting coordinates of the currently written character and the state in which the top of the writing instrument 1' is downward at small time intervals (for example, about 10 [mJ]). Yes, for example, a tablet. below,
The information output from the writing device 1 may also be referred to as writing information.

The CPU 3 sequentially inputs handwritten information via the first interface circuit 2, and performs recognition calculation processing described below based on this information.

Then, it is determined what kind of character the written character is, and the sono character code is outputted from the output terminal 7 via the second interface circuit 6.

FIG. 2 is a flowchart illustrating the recognition calculation process performed by the CPU 3.
The operation of the diagram will be explained.

The CPU 3 via the first interface circuit 2
The coordinates of the writing point and the information of the top and bottom of the writing instrument 1' are sequentially inputted from the writing instrument 1 and recorded on the first memory 4. The first memory 4 is a rewritable memory that temporarily records data, and is, for example, a general-purpose RAM. cpr
ys first performs preprocessing operations described below.

First, one stroke of one character is detected based on the information on the rise and fall of the marking tool 1' in the first memory 4. That is, the period from when the writing instrument 1' goes down from the top to when it goes up again from the bottom is detected as one stroke of the character.

Next, the CPU 5 performs point paste sampling on each side of the character. That is, based on the coordinates of the handwriting on each surface, the coordinates of points arranged at equal intervals on each surface are calculated. Next, the CPU
3 normalizes the position and size of input characters.

Position normalization is an operation in which, for example, the center of gravity of a character is calculated, and the coordinates of each point are translated in parallel so that this becomes the coordinate origin. Finally, calculate y) from the average value of the z and y coordinates of each point obtained by the above resampling, and calculate the x of each point.
, respectively from the y coordinate; , y. Normalization of size means, for example, the distance of each point from the center of gravity (for example,
The average value 7 of the Euclidean distance) is calculated, and the input characters are enlarged and reduced so that 7 becomes a predetermined value. Thth
, when the predetermined value is C, a process is performed in which the +y coordinate value of each point is multiplied by e/1.

Next, approximate representation of each of the three CPU images is performed. Then, based on the coordinates of the points obtained by the resampling, the coordinates of the points dividing each surface into one equal square are calculated. is generally a small number such as 3.6. , (above, step 5-1 in Figure 2).

Second, the CPU 3 performs a matching operation between the input character and each character standard pattern. In other words, the degree of difference from the standard pattern for all characters recorded in advance on the second memory 5 is calculated, and the degree of difference is stored in the first memory 4.
Record above. The second memory 5 is a memory that records data semi-permanently, and is, for example, a general-purpose FROM. On the second memory 5, a standard character pattern is recorded in which all the characters to be recognized are subjected to the above-mentioned preprocessing calculations. For example, the CPU 3 calculates the distance (for example, Euclidean distance) between corresponding points of the corresponding strokes of each character standard pattern having the same number of strokes as the input character,
The sum of the distances between the points (for example, for a character with m strokes, the sum of the distances between m x 1 points) is used as the degree of difference between the input character and the standard character pattern, and this is stored in the first memory. 4 Record on top. The above is step (S-2) in FIG. CPU
5, thirdly, selects the minimum value from among the dissimilarity degrees recorded on the first memory 4. Then, the character code of the standard character pattern giving the minimum value is outputted to the output terminal 7 via the second interface circuit 6. The above is step (S-3) in FIG.

However, according to this method, for example, the uppercase letter r
□ Misrecognition occurs every once in a while between characters that are similar in shape, such as J and "C." For example, an input character written as l□J was written with some deformation, so
” is different from the standard character pattern, and [C]
This is the case where the degree of difference from the standard character pattern becomes the minimum.

In order to reduce such erroneous recognition due to the similarity of character shapes, various similar character identification methods have been developed, and have achieved the effect of reducing the erroneous recognition rate. For example, JP-A-55-13817
As stated in Publication No. 2.

We extract the time series of four types of features as character features: the straightness of the stroke, the relative length ratio, the direction angle from the start point to the end point, and the angle change indicating the degree of curvature of the stroke, and use these to create a character standard. There is a method that performs a matching operation with a pattern. According to this method, compared to the method using the sum of the distances between points.

Since more information about the shape of each stroke is included in the matching operation, it is possible to reduce recognition errors due to the presence of similar characters. However, the amount of calculation required to extract the above features and perform matching calculations is much larger than the amount of calculation required to calculate the sum of distances between points, and the amount of calculations required to extract the above feature amounts and perform matching operations is much larger than that required to calculate the sum of distances between points. The disadvantage is that the amount of memory required to do this is much greater than that required to record the coordinates of several feature points on each surface.

[Purpose of the invention]

An object of the present invention is to provide an online handwritten character that eliminates the drawbacks of the above-mentioned prior art and reduces erroneous recognition due to similar characters without increasing the amount of calculations and memory for recognition processing when inputting Roman characters through online handwritten character recognition. Provides a character recognition method.

[Summary of the invention]

In order to achieve this objective, the present invention performs a matching operation between handwritten input Roman characters and each character standard pattern, and the minimum dissimilarity and the difference or ratio with the minimum dissimilarity are within a certain threshold. The feature is that a character standard pattern that gives a degree of dissimilarity is selected, and from among the character stacking patterns, a character standard pattern that is reasonable as a royal character is selected by considering the recognition result of the previous input character. There is.

If there are multiple character standard patterns that are reasonable as Roman characters, the one that provides the smaller degree of difference is adopted. For example, an input character written as "1°0",
As a result of the matching operation with all alphanumeric standard patterns, the character standard pattern "C" gives the minimum dissimilarity, and the character standard pattern (
When the ratio of OJ to the minimum dissimilarity is within a certain threshold, the letter rOJ is adopted as the recognition result since the letter 1 "C" is not used in the Roman alphabet. .

[Embodiments of the invention]

Embodiments of the present invention will be described below with reference to FIGS. 1 and 2.

This will be explained using FIGS. 3, 4, and 5.

FIGS. 1 and 2 are a block diagram and a flowchart showing an example of the online handwritten character recognition method used to explain the conventional example, and the explanation of each block will be omitted. FIG. 3 is a flowchart showing an example of a method for selecting recognition results from candidate characters, FIG. 4 is an example of a Roman character connection table, and FIG. 5 is a table showing examples of erroneous recognition of alphanumeric characters.

In FIG. 1, the CPU 5 first performs a series of recognition arithmetic processing described in the order of the background for the handwritten Romaji literati. ') Then, the writing information of the written characters is inputted sequentially from the writing instrument 1, and a series of recognition calculation processes including preprocessing calculations, matching calculations, and selection of the minimum degree of difference are performed.

The operation of the CPU 3 will be explained below with reference to the flowchart shown in FIG.

The CPU 5 determines the minimum degree of difference and the difference or ratio between the input characters and the standard pattern of all characters from among the degrees of difference recorded in jlEIK, 1st tD) ie:S4-H, and a predetermined constant value. A character standard pattern that provides a degree of dissimilarity within a threshold value is selected (1-1).

Second, the CPU 3 selects only those that are reasonable as Roman characters from among the character standard patterns, taking into account the recognition results of the previous input characters.

A specific example of the method will be explained with reference to FIG. FIG. 4 is an example of a Roman character connection table listing all cases in which preceding and succeeding characters in a Roman character can be connected, and this content is recorded in the second memory 5 in advance. C
The PU 3 first inputs the recognition result of the previous input character recorded on the first memory 4, refers to the row corresponding to the recognition result of the previous input character in the Roman character connection table, and selects candidate characters. From among these, only the character standard patterns written in the associated connecting character column are selected (t-2). If there is only one selected character standard pattern, it is immediately output as the recognition result for the input character. If there are multiple selected character standard patterns, the one with the smallest degree of difference from the input character is output as the recognition result (t-
3).

Hereinafter, as a specific example, the operation of the CPU 3 in the case of an input character written as "O" will be explained using FIGS. 2 and 3.

When the handwriting information of input characters is input one after another, the CPU 3
First, a series of preprocessing calculations are performed (S-1), and since the input character "0" is detected as one stroke, the degree of difference from the standard pattern of all alphanumeric characters of one stroke is calculated. For example, one stroke is approximately expressed at 6 points, the sum of the distances (for example, Euclidean distance) between the 6 corresponding points between the input character and each character standard pattern is calculated, and this is used to calculate the distance between the input character and the character standard pattern. This is the degree of difference from the pattern. Then, the degree of difference is recorded on the first memory 4 (5-2).

The CPU 3 first selects the minimum value from among the degrees of difference. Here, the degree of difference from the standard character pattern "C" is 1.
Suppose that it is the minimum at 00 (S-3).

Next, the CPU 5 selects a character standard pattern that provides a degree of difference whose ratio to the minimum degree of difference is less than a certain threshold value from among the degrees of difference recorded on the first memory. For example, if the constant threshold value is 162, a character standard pattern that gives a degree of difference of 120 or less is selected. As a result, character standard patterns r□J, b and 11 that give a degree of difference of 110 are obtained.
Suppose that "U", which gives 5, is selected (t-1).

Here, let rKJ be the recognition result of the previous input character. C.P.
U 3 was selected in this way by referring to the Roman character connection table recorded on the second memory 5 [cJ
, r□J, and rUJ, only "O" and "U" which come immediately after rKJ and are reasonable as Roman letters are selected (t-2).

Then, among these, rOJ having a smaller degree of similarity to the input character is adopted as the recognition result for the input character (t-5).

° Next, as an example, out of 36 alphanumeric characters for 21 authors,
A total of 650 characters, excluding characters that are not in the Roman alphabet such as LJ, ll'J, and misspelled characters, are input, and the recognition calculation process shown in Figure 2 is performed to recognize the standard character pattern that provides the minimum degree of dissimilarity. As a result, the situation of misrecognition in this case is shown in FIG. In the point approximation of the image during the preprocessing operation (S-1), L=6, and the distance between points in the matching operation 3i (S-2) is the city distance. The figure shows the number of character standard patterns that give the minimum degree of difference only when they differ from the input characters. The overall misrecognition rate is (37/63o) x 100 = 5
．． It is 9%. According to the online handwritten character recognition method described in the non-implemented example, characters that are reasonable as Roman characters are selected from among the standard character patterns that are candidates for recognition results, taking into account the recognition results of the previous input characters. Therefore, the 21 misrecognized characters marked with a circle in the diagram will be resolved. Therefore, the overall misrecognition rate is (16/6+0
) X 100 = 2.5%.

In addition, in this example, we have explained a method of testing the rationality of a Roman alphabet by considering only the immediately preceding input character, but we list candidates for recognition results for several consecutive characters, and If you select a combination that is reasonable as a Roman alphabet from among them, you can reduce the misrecognition rate.
Of course, this will bring about even greater effects.

〔Effect of the invention〕

As explained above, according to the present invention, in Roman character writing through online handwritten character recognition, not only character standard patterns that provide the minimum degree of dissimilarity from input characters, but also characters that provide a degree of dissimilarity that is extremely close to the minimum degree of dissimilarity from input characters. Standard patterns are included as candidates for recognition results, and from among these candidates, only those that are reasonable as Roman characters are selected, taking into consideration the recognition results of previous input characters, and are used as recognition results for the input characters. Therefore, the rate of misrecognition due to the similarity of character shapes is significantly reduced without using a complicated similar character identification method] 5, it is possible to input rational Roman characters, and this eliminates the drawbacks of the above-mentioned conventional technology. It can provide an online handwritten character recognition method with excellent functions.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an example of an online handwritten character recognition method, FIG. 2 is a flowchart explaining recognition calculation processing, FIG. 3 is a flowchart showing an example of a method for selecting recognition results from candidate characters, and FIG. The figure is an example of a Roman character connection table. FIG. 5 is a table showing examples of incorrect recognition of alphanumeric characters. 1... Writing device 2... First interface circuit 3... CPU 4... First memory 5... Second memory 6... Second interface circuit 7... Output Terminal Figure 2 Figure 3 Figure 4

Claims (1)

[Claims] In an online handwritten character recognition method that sequentially inputs handwriting information of a character currently being written and recognizes the input character written based on this, means for calculating the degree of difference between the input character and a standard pattern of all characters; Means for selecting a character standard pattern that provides a minimum value of the degree of difference and a degree of difference such that the difference or ratio with the minimum value is within a predetermined threshold, and a connectable combination of preceding and succeeding characters in a Roman alphabet sentence. means for recording a Roman character connection table indicating the input character, and if there are a plurality of the selected character standard patterns for the input character of the Roman character, With reference to the recognition results and the Roman character connection table, select a standard character pattern that is reasonable as a Roman character, and if there are multiple standard character patterns that have the rationality, An online handwritten character recognition method characterized in that, among certain standard character patterns, a character standard pattern that gives a minimum value of the degree of difference from the input character is used as a recognition result for the input character.