JPH0934998A - On-line handwritten character recognizing method and device for the same - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide on-line handwritten character recognizing method and device for the same with which even double writing/stroke cut/character describing a stroke from the opposite direction can be recognized by utilizing a recognizing method not depending on the input state of strokes consisting of a character. SOLUTION: A binary pattern composed of continuous lines is provided by performing the processing of interpolation between the writing points of the stroke consisting of a character to a handwritten input pattern 1 by preprocessing 30. The character line of this binary pattern is made thicker by thickening processing and sent to feature extracting processing 40 and a feature describing the character is extracted from the sent binary pattern. While using this extracted feature, the category to which the input pattern 1 belongs is discriminated by identifying processing 60 and a character candidate is outputted. Thus, since on-line handwritten character recognition is performed without depending on the input state of strokes, even double write/stroke cut/ character describing the stroke from the opposite direction can be recognized as well.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to online handwritten character recognition in which a handwriting of a character is represented by a time series of writing point coordinates at the time of writing and the character is recognized from the time series of the obtained position coordinates. More specifically, the present invention relates to an on-line handwritten character recognition method and an apparatus therefor capable of recognizing double strokes / stroke breaks / strokes written from the opposite direction.

[0002]

2. Description of the Related Art A conventional online handwritten character recognition method extracts strokes forming a written character, and associates each stroke between the stroke group and a stroke group forming a standard dictionary stored in advance. Character recognition is performed by performing an attachment process and obtaining the sum of the distances between the associated strokes or the degree of similarity.

[0003]

However, in the above-mentioned conventional online handwritten character recognition method, there is a problem in that it is difficult to recognize double strokes / stroke breaks / strokes written in the opposite direction. there were.

The present invention has been made in view of the above problems in the prior art, and an object of the present invention is to use a recognition method that does not depend on the input state of the strokes that form a character, so that double strokes / stroke breaks can be achieved. Another object of the present invention is to provide a method and an apparatus for making it possible to recognize a character whose stroke is written from the opposite direction, and to relax the input condition for handwritten character recognition.

[0005]

In order to achieve the above object, in the first invention of the present invention, a handwriting of a handwritten input character is selected from a series of writing point coordinate time series from a pen down point to a pen up point when writing. And a step of generating a binary pattern consisting of black and white binary pixels from the expressed stroke, and a step of extracting features from the generated binary pattern to recognize the character. And an on-line handwritten character recognition method characterized by having:

In the above-described online handwritten character recognition method, at the stage of generating a binary pattern, a binary pattern composed of continuous lines is generated by interpolation processing between each writing point of strokes forming a character. This is suitable for correcting the fact that the brush point coordinate sequence becomes discrete due to the difference between the speed of inputting characters and the speed of inputting an input pattern and improving recognition ability.

Further, in the above online handwritten character recognition method, at the stage of generating a binary pattern, generating a binary pattern composed of a line thickened by stroke thickening processing facilitates feature extraction. It is suitable above.

Similarly, in the second aspect of the present invention, the input means for detecting the coordinates of the stroke of the character overlaid on the screen of the display means and input by handwriting, and the stroke of the character for which the coordinates have been detected are detected. A pre-processing unit that generates a binary pattern from a brush point, a feature extraction unit that extracts a feature from the generated binary pattern, and a category of the input character is identified using the extracted feature. An on-line handwritten character recognition device, characterized by comprising: an identification unit.

In the above-mentioned on-line handwritten character recognition device, a pre-processing unit or, independently of the pre-processing, inter-writing point interpolating means for interpolating the writing points of the strokes of a character to form a stroke consisting of continuous lines is provided. It is preferable to have the configuration to improve the recognition ability by correcting the fact that the brush point coordinate sequence is discrete due to the difference between the handwriting character input speed and the input pattern input speed.

Further, in the above online handwritten character recognition device, in the preprocessing section or independently of the preprocessing,
It is preferable to have a configuration for thickening the character line segment of the binary pattern generated by the preprocessing unit in order to facilitate the feature extraction.

In the above-described on-line handwritten character recognition method and apparatus of the present invention, a binary pattern composed of continuous lines obtained by interpolation processing between the writing points of strokes of characters input by handwriting, or the binary pattern. By performing character recognition using the features extracted from the binary pattern obtained by thickening the character lines, it is possible to perform online handwritten character recognition that does not depend on the input state of strokes. / Enables recognition of strokes written in the opposite direction.

[0012]

Next, an embodiment of the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of an online handwritten character recognition method of the present invention, in which 1 is an input pattern, 30 is preprocessing, 40 is feature extraction processing, 60 is identification processing, 70 is an identification result.

The input pattern 1 is represented by a stroke composed of a series of writing point coordinate time series from a pen-down point to a pen-up point at the time of writing a handwriting of a character handwritten by the input means. Sent. The input pattern 1 is binary-patterned by this preprocessing 30,
The features that describe the character are extracted by the feature extraction process 40 and sent to the identification process 60. In the discrimination processing 60, the characteristic is used to perform the discrimination processing for determining the character category to which the input pattern 1 belongs, and the discrimination result 70 is output.

FIG. 2 is a block diagram showing in detail the preprocessing 30 of FIG. 1, in which 31 is a noise removing process, 32 is a smoothing process, 33 is a position normalizing process, and 34 is a size normalizing process. The process, 35 is the inter-stroke interpolation process, and 36 is the thickening process.

The noise removal processing 31 is performed at time t when the writing point at time t and the writing point at time t + 1 in the writing point time series of the input pattern 1 are separated by a predetermined threshold value or more.
This is the first processing to remove the +1 writing point.

Smoothing processing 32 following noise removal processing 31
Is a process for smoothing a state in which the coordinate series includes a large number of irregularities, for example, when a pen shakes due to handwriting.

Position normalization process 3 following the smoothing process 32
3 is defined as the center of gravity of the character by calculating the average value of each of the x and y coordinate values of all the writing points forming the character, and then the input pattern is parallel so that the center of gravity is at the center of the character frame. This is a process of moving.

A size normalization process 34 following the position normalization process 33 is performed so that the average value of the distances from the barycenter of the character obtained above to each writing point becomes equal to a predetermined normalization radius. This is a process for uniformly enlarging / reducing a character pattern around the center of gravity.

The size normalization process 34 is followed by a writing point interpolation process 35, which will be described in detail with reference to FIG.

When interpolating between consecutive writing points a and b in the same stroke, first, the x-axis projection length Δx (= ‖x _b −x _a ‖) between the writing points a and b is calculated. y-axis projection length Δy
(= ‖Y _b -y _a ||) is calculated. However, ‖‖ represents the absolute value.

When Δx ≧ Δy, c = Δy / Δx,
x _b> x _a and y _b> y _a, then the writing point sequence to be interpolated are (x
_{a +1, [c × 1 +} y a +0.5]), (x a +2, [c
_{× 2 + y a +0.5])} ..., (x a + Δx-2, [c ×
_{(Δx-2) + y a} +0.5]), (x a + Δx-1,
It is [c × (Δx-1) + y a +0.5]). However, [] represents a Gaussian function.

[0023] Similarly, x _b> x _a and y _b <y _a, then the writing point sequence to be interpolated are _{(x a +1, [c ×} (-1) + y a +
_{0.5]), (x a +2} , [c × (-2) + y a +0.
5]) ..., (x _a + Δx−2, [c × (−Δx + 2) +
_{y a +0.5]), (x} a + Δx-1, [c × (-Δx +
1) a + y _a +0.5]).

[0024] If x _b <x _a and y _b> y _a, writing point sequence to be interpolated are _{(x a -1, [c ×} 1 + y a +0.5]), (x
_{a -2, [c × 2 +} y a +0.5]) ..., (x a -Δx +
2, [c × (Δx- 2) + y a +0.5]), (x a -Δ
x + 1, which is [c × (Δx-1) + y a +0.5]).

[0025] x _b <x If _a and y _b <y _a, writing point sequence to be interpolated are _{(x a -1, [c ×} (-1) + y a +0.
_{5]), (x a -2} , [c × (-2) + y a +0.5])
_{..., (x a -Δx + 2} , [c × (-Δx + 2) + y a +
0.5]), (x _a −Δx + 1, [c × (−Δx + 1)
+ A y _a +0.5]).

When Δx <Δy, c = Δy / Δx
Then, interpolation is performed by the same process as above with x and y reversed.

FIG. 4 shows an example in which a continuous writing point series is obtained from a discrete writing point series by interpolating the writing point in the stroke in the character "TE".

Next, a detailed description of the thickening process 36 following the inter-stroke interpolating process 35 will be given with reference to FIG.

The binary pattern is composed of black dots representing a character line segment and white dots representing a character background portion. When the brush point interpolation processing is performed, a character pattern composed of line segments having a width of 1 mesh is generated. A part of the character pattern after the interpolation processing is shown in FIG.

First, a method for thickening a character line by 3 meshes will be described. Up / down / at point P
If the four meshes adjacent to the left / right are white points, the white points are converted to black points. At the point P, the points adjacent to the top / left / bottom are white points and are therefore changed to black points that represent the character line segment, but the points that are adjacent to the right are originally black points and the above processing is not performed. The state of the processing is shown in FIG. Next, pay attention to the adjacent point P ′. At the point P ′, since the mesh adjacent to the upper / left side is originally a black point, pre-processing is not performed,
The mesh adjacent to the bottom / right side is a white point, so a new black point is created. The state of the processing is shown in FIG. The thickening process is performed on all the meshes of the character line segments that exist in the same manner.

Similarly, when thickening to a width of 5 meshes, the thickening process is efficiently performed by focusing on only the mesh newly added in the first thickening process and performing the same process. It becomes possible. Further, for thickening processing beyond that, it is possible to repeat the above method.

In the thickening process described above, the number of adjacent meshes of interest is 4 meshes. However, there are 4 meshes adjacent to top / bottom / left / right and 4 meshes adjacent to upper right / lower right / lower left / upper left. It is also possible to pay attention to the 8 meshes that are combined. An example in the case of paying attention to adjacent 8 meshes is shown in FIG. 5 (d) and FIG. 5 (e).

FIG. 6 shows an example in which the thickening process is performed on the character "F". FIG. 6A shows a character pattern after the inter-stroke interpolation processing is performed, and FIG. 6B shows a character pattern after being thickened to a width of 3 mesh when focusing on 4 meshes of upper / lower / left / right. Character pattern, FIG. 6 (c) is a character pattern after the character pattern is thickened to a width of 5 mesh, FIG. 6 (d).
Is a character pattern after being thickened to a width of 3 mesh in the case of focusing on 8 meshes of top / bottom / left / right / bottom right / bottom left / bottom left, and FIG. 6 (e) shows the character pattern of 5 meshes. Represents a character pattern after it is widened to the width.

FIG. 7 is a diagram showing a configuration example of the on-line handwritten character recognition apparatus of the present invention. 10 is an input integrated display section, 11 is an input section, 12 is a display section, 20 is a stylus pen, and 30a is 30a. A preprocessing unit, 40a is a feature extraction unit, 50 is a dictionary, and 60a is an identification unit.

The operation of the apparatus configured as above will be described.

In this configuration example, the case where the extended contour contribution rate feature is used as a feature for describing a character will be described as an example.

The input-integrated display unit 10 is constituted by stacking the input unit 11 and the display unit 12, and for example, the input unit 11 is an electromagnetic induction tablet,
A liquid crystal display device is used as the display unit 12.

Characters are input to the input unit 11 by the stylus pen 20 as individual strokes. The stroke group input to the input unit 11 is cut out character by character and sent to the preprocessing unit 30a.

The stroke groups sent to the preprocessing unit 30a are subjected to the preprocessings 31 to 36 in the procedure shown in FIG. A well-known pre-process is a noise removal process. The noise-removed stroke group is subjected to smoothing processing, and then to position normalization processing and size normalization processing.

Further, the preprocessing unit 30a of the present invention performs the inter-stroke interpolation processing in addition to the above-mentioned preprocessing. The reason for performing this processing is that when the writing speed input by the stylus pen 20 is not sufficiently fast for the input unit 11 to take in as a writing point coordinate sequence, the writing point coordinate sequence becomes discrete. This is to increase the cognitive ability correspondingly. As the interpolation method, for example, the interpolation method shown in FIG. 3 may be used.

By performing the inter-stroke interpolating process, 2
A thickening process for thickening the line width of the character line segment is then performed on the value-patterned input pattern. The reason for performing this process is to facilitate the feature extraction in the feature extraction unit 40a. As a thickening processing method, for example, a thickening processing as shown in FIG. 5 can be considered. By performing the pre-processing, the input pattern is written twice
It becomes possible to convert the stroke into a character pattern that does not depend on the input state of the stroke, such as writing from the opposite direction.

The pre-processed character pattern is sent to the feature extraction unit 40a. The feature extraction unit 40a performs a process of extracting the extended contour direction contribution rate feature from the character pattern sent from the preprocessing unit 30a. The extracted features are sent to the identification unit 60a.

The identification unit 60a, which has received the character pattern subjected to the characteristic process, calculates the distance between the characteristic of each standard pattern extracted from the dictionary 50 and the characteristic of the input pattern. The character category having the smallest distance is sent to the display unit 11 as an identification result. Further, the character categories from the character having the second smallest distance to the character having the nth smallest distance are sent to the display unit 11 as a candidate character group. The value of n is set in advance.

With the above operation, it is possible to perform online handwritten character recognition independent of strokes.
It becomes possible to recognize a character written twice from strokes / stroke breaks / strokes in opposite directions.

FIG. 8 shows an example in which the character "YO" written twice can be recognized. In FIG. 8A showing the input pattern, it is difficult to correctly recognize it by the conventional method because the horizontal line is written in three strokes by rewriting it later. An enlarged view of the portion where the writing is performed twice is shown in FIG. However, since the recognition method of the present invention does not depend on the stroke order / stroke count by forming a binary pattern,
It becomes recognizable because it is less susceptible to double writing. An enlarged view after the binary patterning is shown in FIG.

FIG. 9 shows an example in which the character "4", which is out of stroke, can be recognized. In the input pattern FIG. 9A, since the stroke of the first stroke is broken, the conventional method has a very high possibility of being erroneously recognized as “thousand” of three strokes.
The standard pattern of "4" is shown in FIG. 9B, and the standard pattern of "thousand" is shown in FIG. 9C. However, in the recognition method of the present invention, the binary patterning allows the recognition to be performed because it is not affected by the fluctuation in the number of strokes due to the stroke break. Figure 9 shows the input pattern after binary patterning.
9D shows the standard pattern of "4" and FIG. 9F shows the standard pattern of "1000".

FIG. 10 shows an example in which the character "Noh" whose stroke is written in the opposite direction can be recognized. In FIG. 10A showing the input pattern, strokes are written from the opposite direction. An enlarged view of the portion where the stroke is written from the opposite direction is shown in FIG. In the conventional method, since the distances are calculated in order of the stroke input direction, the distance value between the standard pattern and FIG. 9B becomes large, and the probability of erroneous recognition increases. An enlarged view of the standard pattern corresponding to the enlarged view is shown in FIG. But,
In the recognition method of the present invention, the binary pattern is used, so that it can be recognized because the influence of the stroke input direction is eliminated. An enlarged view of the input pattern after binary patterning is shown in FIG. Standard pattern 2
An enlarged view after the value patterning is shown in FIG.

In the embodiment described above, the extended contour contribution feature is used as the feature extraction method, but it is also possible to use other feature extraction methods.

Further, in the present embodiment, an example in which both the inter-stroke point interpolation processing and the thickening processing are executed is shown, but the thickening processing may not be necessary depending on the feature extraction processing method. In that case, as shown in FIG. 2, the binary pattern may be sent from the inter-stroke interpolating process 35 to the feature extracting process 40. In addition, when the brush point coordinate sequence can be captured at high speed, there may be a case where inter-brush point interpolation processing is not required.

Further, in the present embodiment, the inter-stroke interpolating section and the thickening processing section are provided inside the pre-processing section, but a circuit different from the pre-processing section may be used.

Although the present invention has been specifically described based on the embodiments of the present invention, the present invention is not limited to the embodiments described above, and various modifications can be made without departing from the spirit of the present invention. Needless to say.

[0052]

As described above, according to the on-line handwritten character recognition method and apparatus of the present invention, recognition can be performed without depending on the stroke input state. Stroke order / specific to the device
It is possible not only to input the stroke order / strokes freely by eliminating the writing stroke limit, but also to recognize characters written from double strokes / stroke breaks / reverse strokes. There is an advantage that it is possible to realize an input method and its device that are extremely resistant to writing characters in a bad environment such as.

Further, in the case where the inter-writing point interpolation is performed in the pre-processing, especially when the writing point coordinate sequence of the input character becomes discrete due to the difference between the writing speed and the input velocity of the writing point coordinate sequence. In addition, the stroke can be corrected to be a continuous line, which can contribute to the improvement of the recognition ability.

When stroke thickening processing is performed in the preprocessing, feature extraction for recognition can be particularly facilitated, which contributes to improvement of recognition ability.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of an online handwritten character recognition method of the present invention.

FIG. 2 is a block diagram showing details of preprocessing in the above embodiment.

FIG. 3 is a diagram showing a state of interpolating between stroke writing points in the preprocessing.

FIG. 4 is a diagram showing a process of interpolating between stroke writing points in the preprocessing.

FIG. 5 is a diagram illustrating thickening processing of a line segment in a binary pattern in the preprocessing.

FIG. 6 is a diagram showing an example of the thickening process.

FIG. 7 is a diagram illustrating a configuration example of the present invention.

FIG. 8 is a diagram showing an example in which writing is performed twice.

FIG. 9 is a diagram showing an example of stroke shortage.

FIG. 10 is a diagram showing an example in which strokes are written from the opposite direction.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 ... Input pattern 10 ... Input integrated display part 11 ... Input part 12 ... Display part 20 ... Stylus pen 30 ... Preprocessing 30a ... Preprocessing part 40 ... Feature extraction 40a ... Feature extraction part 50 ... Dictionary 60 ... Identification process 60a ... Identification unit 70 ... Identification result 31 ... Noise removal processing 32 ... Smoothing processing 33 ... Position normalization processing 34 ... Size normalization processing 35 ... Inter-brush point interpolation processing 36 ... Thickening processing

Front Page Continuation (72) Inventor Kazumi Otaka 1-1-6 Uchisaiwaicho, Chiyoda-ku, Tokyo Nihon Telegraph and Telephone Corporation

Claims (6)

[Claims] 1. A step of expressing a handwriting of a handwritten input character by a stroke consisting of a series of writing point coordinate time series from a pen down point to a pen up point at the time of writing, and a black and white binary pixel from the expressed stroke. An online handwritten character recognition method comprising: a binary pattern consisting of: and a step of extracting a feature from the generated binary pattern to recognize the character. 2. The online pattern according to claim 1, wherein, in the step of generating a binary pattern, a binary pattern composed of continuous lines is generated by an interpolation process between each writing point of strokes forming a character. Handwriting recognition method. 3. The step of generating a binary pattern, wherein the strokes are thickened by thickening 2
3. The online handwritten character recognition method according to claim 1, wherein a value pattern is generated. 4. Input means for detecting the coordinates of strokes of characters to be handwritten and input, which is superimposed on the screen of the display means, and a binary pattern is generated from the writing points of the strokes of the characters for which the coordinates have been detected. A preprocessing unit; a feature extraction unit that extracts features from the generated binary pattern; and an identification unit that identifies the category of the input character using the extracted features. A unique online handwriting recognition device. 5. The pre-processing unit or independently of the pre-processing unit has a writing point interpolating means for interpolating between the writing points of the strokes of a character to form a stroke composed of continuous lines. Item 5. The online handwritten character recognition device according to Item 4. 6. The pre-processing unit or independent of the pre-processing unit has means for thickening the character line segment of the binary pattern generated by the pre-processing unit. 5. The online handwritten character recognition device described in 5.