JPS63307593A - Continuous character recognizing device - Google Patents

Abstract

PURPOSE:To reduce the arithmetic quantity which follows up a recognition, by dividing a character-string into plural partial character-strings. CONSTITUTION:The titled device is provided with a partial character-string dividing part 300 for dividing an input pattern from an input means 100, into plural partial character-strings, based on character pitch information of said input pattern, giving them to a recognizing means 500. By the input means 100, a character-string data is converted to an input pattern, and this input pattern is divided into plural partial partial character-strings by using the charac ter pitch information by the partial character-string dividing part 300 and sent to the recognizing means 500. In the recognizing means 500, patterns which are held in an isolated character pattern storage part 600 and an inter-character stroke pattern storage part 700 are synthesized, pattern is recognized. In such a way, the arithmetic quantity can be reduced remarkably.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a means for recognizing one or more consecutively written characters based on handwriting of the characters manually drawn.

〔overview〕

The present invention is a means for recognizing a continuously written character string based on a composite pattern of an isolated character pattern and an inter-character stroke pattern. It is also possible to reduce the

, [Prior Art] A conventional device that recognizes a string of consecutively written characters without the user instructing the end of one character includes, for example, the device disclosed in Japanese Patent Application No. 61-053397. There was a continuous character recognition device described in "No." (hereinafter referred to as Document (1)). Hereinafter, a conventional continuous character recognition device will be explained using this device as an example.

This device uses continuously written characters, including those written without breaks, as an input pattern, and pattern-matches this pattern with a standard pattern stored in the recognition device in advance to recognize the characters. do. As a method for recognizing isolated characters written individually one by one using pattern matching, for example, the Institute of Electronics and Communication Engineers Technical Research Report PRL83-29 (August 27, 1983)
There is a method shown in a paper published on pages 1 to 8 entitled "Online handwritten character recognition using stacked DP matching" (hereinafter referred to as document (2)).

In this method, first, a human-written character is created using a time-series pattern of direction angles of line segments that make up the character A=(ai11≦i≦■
) is converted to Here, ai represents the direction angle, and ■ represents the number of line segments. Let this time-series pattern be input pattern A. The standard pattern is stored in advance in the recognition device as a time-series pattern with the same direction angle as the input pattern.

This is the standard pattern Bk = (bj l 1≦J≦
Jk).

Here, k (1≦≦K) is the distance between standard patterns Lbj
represents the direction angle, and Jk represents the number of line segments of the standard pattern of category 'Jk. The distance between the i-th data of input pattern A and the j-th data of standard pattern Bk is determined by the direction angles a1 and b.
Defined by the angle between J.

Let this distance be d (i, j). The inter-pattern distance Dk between the input pattern A and the standard pattern Bk is this distance d
Let this be the cumulative value of the distance d(i, j) when input pattern A and standard pattern Bk are aligned on the time axis so that the cumulative value of (i, j) in the time axis direction is the minimum. This matching on the time axis is performed as follows using the DP matching method described in document (2).

The recurrence formula for each time l of input pattern A is 1≦J≦J
By solving for k, the minimum cumulative distance up to time i can be found. Here, g (i, j) represents the recurrence formula value at one time lS J in the category. However, as the initial value g (0, 0) = 0, g (0, J)
=■(1≦”≦Jk).

The recurrence formula operation of equation (1) is performed at the input pattern time i (1
≦i≦1), the distance between patterns Dk becomes Dk = g (I, Jk)
It can be found as (2). The category IJ l< for which the inter-pattern distance Dk obtained in this way is the minimum is set as the recognition result.

To extend the recognition of isolated characters described above to recognition of continuous characters, use a standard pattern of continuous characters that is a combination of isolated characters.
All you have to do is perform pattern matching with the input pattern of continuous characters. As standard patterns, an isolated character pattern representing the handwriting from the beginning to the end of one character, and an inter-character stroke pattern that connects the end and start points of the isolated character patterns are used. Hereinafter, the point at which writing of these character patterns begins will be referred to as the starting point, and the point at which writing will end will be referred to as the ending point. These isolated character patterns and inter-character stroke patterns are alternately connected to synthesize a standard pattern of continuous characters. In order to efficiently recognize such a plurality of standard patterns by linking them in a predetermined order, a finite state automaton described in Japanese Patent Application No. 83199/1980 is used.

In this way, continuous characters can be recognized by performing pattern matching on the basis of a standard pattern in which isolated character patterns and inter-character stroke patterns are alternately connected and synthesized.

[Problem that the invention seeks to solve]

In such a conventional continuous character recognition device, the recurrence formula calculation shown in equation (1) is performed for each standard pattern at time i of the input pattern. Therefore, as the number of manually input characters increases and the number of line segments I of the input pattern increases, the amount of calculation of the recurrence formula increases. Furthermore, for continuous characters, a standard pattern of a character string is synthesized by combining an isolated character pattern and an inter-character stroke pattern, so as the number of combinations increases, the amount of calculation further increases. In this way, as the number of characters manually input increases, the amount of calculation also increases.

Furthermore, in a human-written character string, there is a high possibility that the portion where the pen moves away from the writing surface (hereinafter referred to as pen-up stroke) corresponds to the space between characters. However, when performing pattern matching on a standard pattern synthesized with an input pattern, pen-up strokes and inter-character strokes do not necessarily correspond. Therefore, conventional continuous character recognition devices do not make full use of this character amount information.

As described above, conventional continuous character recognition devices have the drawback that the amount of calculation increases with the number of input characters, and character amount information is not fully utilized for recognition.

The present invention solves this problem, and aims to provide a continuous character recognition device with a small amount of calculation and high character division accuracy.

[Means for solving problems]

The present invention includes an input means for reading the handwriting of a written character string as a time-series input pattern, and an isolated character pattern storage unit that holds an isolated character pattern representing the handwriting from the beginning of writing to the end of the character. , a character-to-character stroke pattern storage unit that stores a character-to-character stroke pattern representing handwriting from the end of writing one character to the beginning of the next character; and one or more isolated character patterns and zero or more character-to-character stroke patterns alternately. and a recognition means for recognizing a pattern given to the means based on a pattern concatenated without contradiction, the continuous character recognition device recognizes the input pattern into a plurality of parts based on the character pitch information of the input pattern from the input means. The present invention is characterized by comprising a partial character string dividing section that divides the character string into character strings and provides the divided character strings to the recognition means.

[Principle of the invention]

In the present invention, a human-generated character string is divided into a plurality of partial character strings using information between characters, and then each partial character string is recognized by pattern matching similar to the conventional method. This character amount information refers to the spacing between characters in one character string (
(hereinafter referred to as "character pitch") are approximately equal, and it is assumed that there is a high possibility that a pen-up stroke exists between characters.

The process of dividing this human-powered character string into partial character strings will be explained below. The input character string is converted into an input pattern A, which is a time series of x and y coordinates as shown below.

A= ((xi, yi, pi) l 1≦l≦I
) −, (3) ×l is the X coordinate, yl is the y coordinate,
Further, pl represents the up and down of the pen at time 1. pl is p when the pen changes from up to down
i=1, and p1 when changing from down to up
= 2, otherwise p1 = 0. First, a pen-up stroke portion that is likely to be between characters is searched for in input pattern A. That is, at time i=m, pl=
2, and if i=m+l and pi=1, then the period from time m to m+l becomes a pen-up stroke.

Further, the amount of change in the pen-up stroke in the X direction at this time is assumed to be the character pitch. That is, the character pitch d x (u) for the U-th pen-up stroke and the time S (U) at that time are expressed as follows.

dX(u)=xs, l XIm
""'(4) S(u)=m (1≦U≦U
) (5) Next, among these U character amount candidates, only those with positive character pitch are searched, and those with negative character pitch are excluded. That is, d x (u) > 0
Only the pen-up strokes where (6) stands out are left as candidates for character spacing. Also, update the character pitch d x (u) and time s (u) that satisfy equation (6),
d x (v) = d x (u)
(7) s M = s (u) (1≦U≦T, T, 1≦V≦V, V≦U) (8). Here, when calculating the average value Pav of these V character pitches, t'av= (hax(v)) /V
, , , (9)
becomes. Character pitches dXM smaller than this average value Pav of character pitches are again excluded from candidates. That is, dx
(v)>Pay −α1
Only pen-up strokes that satisfy the following are candidates for character spacing. Also, by updating the character pitch d x (v) and time s (v) that satisfy the αC formula, d x H= d x (v)
αDS (b) = s (v) (1≦v≦V, l≦w≦W, W≦V) a! Let it be l.

Here, the input pattern A is divided by this candidate between characters. That is, if input pattern A is separated by time S(g), then Aw = ((xt, y+, pu) l s□≦l≦S
(W+1)) αJ However, it is divided into W substrings such that 1≦W≦W, W+1=I. The human-powered character string can be recognized by performing pattern matching on each of these partial character strings in the same manner as in the past.

For example, if a character string consisting of n characters is manually generated and the number of characters is equal to the number of categories of standard patterns, the number of combinations of standard patterns to be synthesized using the conventional method is K1100. However, if this is equally divided into W substrings of n/W characters, the number of combinations for each substring is K II/II α
Therefore, the entire human string is W −K″/' α
■When K”≧2 (normally, this holds true because K>2 and n>l) K”≧W−K””07)
holds true, and the number of combinations of standard patterns is reduced by dividing the human string. Therefore, the number of recurrence formula calculations of equation (1) performed for each standard pattern is also reduced, and the amount of calculations for pattern matching is reduced.

Furthermore, all pen-up strokes in the input character string are tested for the possibility of being spaced between characters, and character amount information is used.

The principle of the present invention has been explained above. In the above explanation, the inter-character pitch is defined by the amount of change in the X direction of the pen-up stroke, but there are other methods such as defining it by the interval in the X direction between partial character strings separated by the pen-up stroke. .

In addition, although we used the average value Pav of the pitch between characters to find character length candidates, it is also possible to use not only the average value (values found from statistics such as variance or the height of the character string in the X direction). .

'Furthermore, we have described the basic isolated character recognition method using directional data, but there are also other methods, such as "Kuzushi "Online handwritten kanji recognition moves toward easing restrictions on handwriting, etc."
A variety of methods can be used, as described in the literature published under the title.

In addition, regarding continuous character recognition methods, for example, the method of the paper published on page 67 or 76 of IEICE technical research report PRL84-13 entitled "Online handwritten character string recognition using candidate character lattice method". Various other items can be used.

[Effect]

In the input method, string data is converted into an input pattern,
This input pattern is divided into a plurality of partial character strings using character pitch information in a partial character string dividing section and sent to recognition means. The recognition means synthesizes the patterns held in the isolated character pattern storage section and the intercharacter stroke pattern storage section, performs pattern matching with the partial character string, and recognizes the input pattern. This allows the amount of calculation to be significantly reduced.

〔Example〕

Hereinafter, one embodiment of the present invention will be described based on the drawings. 1st
The figure is a block configuration diagram showing the configuration of this embodiment.

FIG. 2 is a block configuration diagram showing the configuration of the partial character string division section shown in FIG. 1.

This embodiment uses a tablet 10 which is an input means for reading the handwriting of a written character string as a time-series input pattern.
0, a preprocessing unit 200, an isolated character pattern storage unit 600 that holds an isolated character pattern representing the handwriting from the beginning of writing one character to the end of writing, and an isolated character pattern storage unit 600 representing the handwriting from the end of writing one character to the beginning of writing the next character. A pattern given to this means based on a character interval stroke pattern storage unit 700 that holds character interval stroke patterns, and a pattern in which one or more isolated character patterns and zero or more character interval stroke patterns are alternately and consistently connected. A data conversion unit 400 and a recognition unit 500, which are recognition means for recognizing the input pattern, divide the input pattern into a plurality of partial character strings based on character pitch information of the input pattern from the input means, and provide the partial character strings to the recognition means. This partial character string dividing section 300 is a feature of the present invention.

In FIG. 1, character string data input from a tablet 100 is converted by a preprocessing unit 200 into an input pattern A, which is a time series of the X coordinate, X coordinate, and pen data shown in equation (3), The partial string dividing section 300 uses the character pitch information to divide the input pattern into a plurality of partial character strings Aw, and sequentially transfers them to the data converting section 400.The data converting section 400 uses the X coordinate ,X
A partial character string composed of a time series of coordinates and pen data is converted into time series data of direction angles shown in document (2), and sent to the recognition unit 500. As shown in document (1), the recognition unit 500 synthesizes the standard patterns held in the isolated character pattern storage unit 600 and the intercharacter stroke pattern storage unit 700, and performs pattern matching with the partial character string. recognizes the input pattern and outputs the result R. The result R is a combination of the recognition results of the partial character string Aw.

The operation of the partial string dividing section 300 will be explained in detail using FIG. Input pattern A sent from preprocessing section 200 is temporarily stored in data buffer 301. Here, the time signal l is sent from the control unit 311 to the data buffer 3.
01 and the pen-up detection section 302. Data buffer 301 sequentially sends input pattern A=(xi, yi, pi) to pen-up detection section 302 in synchronization with this time signal l. The pen-up detection unit 302 detects a pen-up stroke where p1=2 and pi+1=1 in the input pattern A, and sends the time signal i=m to the switch 30.
4, and its X coordinates Xll and X11*1 are sent to the subtracter 303. The subtracter 303 subtracts d = Xs+
lxfi is calculated, and when d≧0 holds, the subtraction result d is sent to the D memory 305. At the same time, the subtractor 303
switches the switching signal sw to the switch 30 when d≧0 holds true.
4 to make the line conductive. Therefore, the time signal m sent from the pen-up detection section 302 is transmitted to the switch 304.
is sent to the V memory 306 and held there.

Both the D memory 305 and the V memory 306 are memories with one-dimensional arrays, and the time m when the input pattern is pen-up and the character pitch d at that time are stored in correspondence with each other. When time signal i changes from 1 to I,
In equations (7) and (8), d x (v) is stored in the D memory, and S (V) is stored in the V memory.

When the time signal i changes from 1 to I, the control unit 31
1 sends the signal t1 to the D memory 305 and the averaging section 307. In response to this signal t, the character pitch d is sequentially read out from the D memory 305 and sent to the averaging section 307.

The averaging unit 307 performs calculations such as equation (9), and sends the average value Pay of the character pitch d to the Pav memory 309, where it is stored.

When the average value Pav is determined, the control unit 311 outputs the signal t2.
D memo! 7305, V memory 306 and comparison unit 3
Send to 08. Corresponding to this signal t2, the character pitch d is output from the D memory 305, and the time m is input from the V memory 306.
are sequentially read out and sent to the comparison unit 308.

The comparison unit 308 compares P stored in the Pav memory 309.
The avO value is compared with the value of the character pitch d read from the D memory 305, and when d>Pav, that is, the α0 formula holds, the time m is sent to the W memory 310.

The W memory 310 is a one-dimensional array memory similar to the D memory 305 and the V memory 306, and stores candidate times between characters, that is, S (b) of the (sub) expression. When all the character spacing candidates are stored in the W memory 310, the control unit 31
1 sends the signal t to the W memory 310. data buffer'
7301 is the data conversion unit 40 that separates the input pattern A at this time and converts it into a partial character array Avll as shown in formula 03.
Send to 0.

As explained above, the input pattern A manually entered into the partial character string dividing unit 300 is divided into partial character strings Aw. This partial character string Aw is manually input to the recognition section 500 through the data conversion section 400, and a recognition result R is output. The configuration of this recognition unit 500 is shown in document (1).

〔Effect of the invention〕

As explained above, the present invention (1) does not require any special operation for dividing characters by the user, reduces the amount of calculations, and improves character division accuracy, so continuous characters can be recognized at high speed and with high accuracy. This has the effect of providing a recognition device.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the structure of an embodiment of the continuous character recognition device of the present invention. FIG. 2 is a detailed configuration diagram showing the configuration of the partial character string division section in FIG. 1. 100...Tablet, 200...Pretreatment section, 30
0... Partial string division unit, 301... Data buffer, 302... Pen-up detection unit, 303... Subtractor, 304... Switch, 305... D memory, 3
06...V memory, 3o7...Averaging section, 308.
...Comparison section, 309...Pav memory, 310...
W memory, 311...Control unit, 400...Data conversion unit, 500...Recognition unit, 600...Isolated character pattern storage unit, 700...Character stroke pattern storage unit.

Claims (1)

[Claims] (1) An input means (100, 200) that reads the handwriting of a written character string as a time-series input pattern, and an isolated character pattern storage unit that holds an isolated character pattern representing the handwriting of one character from the beginning to the end of writing. (600), a character-to-character stroke pattern storage unit (700) that stores a character-to-character stroke pattern representing handwriting from the end of one character to the beginning of the next character, one or more isolated character patterns and zero or more character patterns; a recognition means (500) for recognizing a pattern given to the means based on a pattern in which stroke patterns between characters are connected alternately and without contradiction; the character pitch of the input pattern from the input means; A continuous character recognition device comprising: a partial character string dividing section (300) that divides the input pattern into a plurality of partial character strings based on information and provides the divided character strings to the recognition means.