JPS5929246Y2 - Online recognition processing device for handwritten characters - Google Patents

Description

[Detailed description of the invention] Description of the field to which the invention pertains This invention is an online character recognition processing device that sequentially performs recognition processing using information from the process of handwriting characters. This invention relates to a recognition processing system that switches the number of feature points to be extracted in each stroke based on a relatively simple character and a relatively simple character.

Description of the Prior Art This type of online character recognition processing device recognizes each stroke constituting a character by multiple points (for example, the starting point of the stroke,
(midpoint, end point) is written using feature points, and the recognition of the distance between these feature points is sought between standard characters prepared in advance.
Some methods use a method that recognizes the input character as a standard character that provides the minimum value.

This recognition method only needs to use a relatively small number of characteristic characters to recognize kanji and katakana characters whose main constituents are straight lines, but it is useful for recognizing hiragana and numbers whose main constituents are curved lines. requires the use of a relatively large number of feature points.

Therefore, in order to use this recognition method to configure a recognition device that uniformly recognizes a mixture of kanji, katakana, hiragana, numbers, etc., it is necessary to use an arbitrary number of feature points depending on the number of strokes of the input character. A method can be considered in which a feature point approximation means for extraction is provided and the number of feature points is switched depending on the number of strokes of the input character.

However, in this case, there was a drawback that a feature point approximation device with a complicated configuration was required.

Purpose of the invention In order to eliminate such drawbacks, the present invention provides a classification means for classifying input characters into groups including kana characters, numeric symbols, etc. and other groups based on the number of strokes of the characters. For characters belonging to the latter group, which have a relatively large number of strokes (4 strokes or more), the number of feature points is uniformly set to, for example, 3 points (start point, middle point, end point), and a means dedicated to 3-point approximation is used. For characters in the former group with a relatively small number of strokes (3 or less strokes), each stroke is treated as six feature points, for example, and the feature points are extracted using a means dedicated to six-point approximation. It was designed to be extracted.

Therefore, the present invention classifies input characters into two groups, and provides a dedicated point approximation means that can be configured relatively easily for each group. This method has the advantage that the configuration of the entire device can be simplified compared to preparing a single point approximation means that can perform extraction.

This will be explained in detail below using the drawings. Explanation of the constituent buttons and functions of the invention In Figure 1, for characters with a stroke count of 4 or more, the stroke is described by 3 feature points, and for characters with a stroke count of 3 or less, the stroke is described by 6 feature points. 1 shows a block diagram of an embodiment of the present invention in a case where:

In the figure, 1 is a character information input means for detecting the coordinates of a writing point represented by a data tufflet, 2 is a preprocessing means, and 3 is a screen)
Loaf number classification means, 4 is a stroke 3-point approximation means, 5 is a stroke 6-point approximation means, 6 is an inter-pattern distance calculation means, 7 is a standard pattern storage means, 8 is a minimum distance detection means,
9 is a switching means.

The operation of the present invention is as follows.

First, the characters from the character information input means 1 are processed by the preprocessing means 2.
is normalized to an appropriate size.

Thereafter, the input characters are classified into two groups by the stroke number classification means 3 according to the number of strokes.

That is, when the number of strokes of a character is four or more, the three-point stroke approximation means 4 approximates each stroke with three characteristic points: the start point, the middle point button, and the end point of the stroke.

Here, the starting point is the point at which the stroke begins, the midpoint is the point corresponding to 1 of the total length of the stroke, and the end point is the point at the end of the stroke.

Also, if the number of strokes of the input character is 3 or less, hiragana,
Since this includes numerical symbols, the six-point stroke approximation device 5 approximates each stroke with six feature points that divide the stroke length into five.

Therefore, a character with a stroke count of 4 or more can be represented by 3n feature points, where the stroke count is D.

Also, characters with a stroke count of 3 or less can be represented by 6 feature points.

In this way, the point approximation means can be selected by the switching means 9 according to the number of strokes of the two groups of input characters, and the feature points can be uniquely set.

Next, the characteristic pattern of this input character is combined with a standard pattern previously stored in the standard pattern storage means 1 by approximating each stroke with 3 points (the number of strokes is 4 or more) or 6 points (the number of strokes is 3 or less). The inter-pattern distance calculation means 6 calculates the inter-pattern distance as follows.

That is, when the recognition target category is represented by θ and the inter-pattern distance is D(, the distance D( is given by).

Here, (X2+Xi) is the coordinate of the ii-th feature point of the input pattern, (xiθ, yiθ) is the coordinate of the first feature point of the standard pattern belonging to category θ, and σixθ is X1
The variance of θ, σiyO represents the variance of yiθ, and M represents the total number of feature points of the input pattern and standard pattern.

When the number of strokes is n, M=3n for characters with a stroke count of 4 or more, and M26n for characters with a stroke count of 3 or less.

Among the inter-pattern distances D() calculated in this way, the minimum value is detected by the minimum distance detecting means 8, and the category θ indicating the minimum value is recognized as an incoming pattern pattern.

The buttons, components 1, 2, 4, 6, 7 and 8 of this device are based on Japanese Patent Application No. 52-083733 (Japanese Unexamined Patent Publication No. 54-18633).
The configuration is similar to that shown in the embodiment.

Figure 2 shows hiragana characters with a stroke count of 3 or less.
It shows the change in recognition rate with respect to the number of stroke feature points.

This experimental data uses 8 training samples per category.
0, and the number of unknown samples per category was 30. This was obtained as a result of an experiment.

From this experimental data, it can be seen that the number of stroke feature points of 6 is appropriate for characters such as hiragana whose main constituent elements are curved lines.

At the same time, it can also be seen that a relatively high recognition rate can be obtained even when a stroke is described using four or five feature points.

On the other hand, 3-point approximation and 5-point approximation processes repeat the calculation of the midpoint L
7, it requires less calculation and can be configured more easily than 4-point or 6-point approximation processing.

L7 uses easy point approximation processing, sets the number of strokes and classifies input characters so that the number of greeps is as small as possible without reducing the recognition rate, and uses special feature point approximation for each group. By providing a point approximation means capable of extracting an arbitrary number of feature points, it is possible to construct an online character recognition processing device that is more efficient than a recognition device using a stand-alone window.

Description of Effects As explained above, the online recognition processing device for handwritten characters according to the present invention groups input characters into characters with a large number of strokes and characters with a small number of strokes, depending on the number of strokes of the input characters. By setting the number of feature points that approximate each stroke of input characters for each group, in addition to characters whose main components are M lines such as kanji and katakana characters, curved lines such as hiragana and numbers can be It is possible to have sufficient recognition ability even for characters whose main components are , and the structure of the feature point approximation means and the structure of the entire recognition device can be simplified.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of one embodiment of the present invention,
FIG. 2 is an explanatory diagram illustrating the recognition rate according to hiragana characters with a stroke count of 3 or less. In the figure, 1 is a character information input means, 2 is a preprocessing means, 3 is a stroke number classification means, 4 is a stroke 3-point approximation means, and 5
6 represents a stroke 6-point approximation means, 6 represents an inter-pattern distance calculation means, 7 represents a standard pattern storage means, 8 represents a minimum distance detection means, and 9 represents a switching means.

Claims (1)

[Scope of utility model registration request] Characters including at least hiragana and kanji are input while being written using a character writing device, the strokes of the input characters are traced, and the coordinates of multiple writing points of each stroke are extracted as feature points to create standard characters prepared in advance. a character information input means for inputting characters while detecting the coordinates of a writing point using an online recognition processing device for handwritten characters that recognizes standard characters as input characters based on distances between the standard characters and the respective feature points; A stroke number classification means for classifying input characters into characters with a large number of strokes and characters with a small number of strokes, and in the case of characters with a large number of strokes, feature points are extracted for each stroke and a turn is made based on the characteristics of the input character. a first point approximation means for forming a character, and a second point approximation means for forming a characteristic pattern of an input character by extracting feature points in units of strokes using more feature points than the first point approximation means in the case of a character with a small number of strokes. 2 point approximation means, the first point approximation means and the second point approximation means;
a standard pattern storage means for storing a standard pattern formed from each feature point of the standard character; and a standard pattern storage means for storing a standard pattern formed from each feature point of the standard character; inter-pattern distance calculation means for calculating the distance between the feature pattern and each standard pattern of the standard characters, and a standard that detects a minimum value from among the calculation results of the inter-pattern distance calculation means, and which becomes the minimum value. What is claimed is: 1. An online recognition processing device for handwritten characters, comprising: minimum distance detection means for recognizing characters as input characters.