JPS59103176A - On-line character recognizer - Google Patents

Abstract

PURPOSE:To recognize assuredly the continuously handwritten character by transmitting the stroke code in a segment code buffer after decomposing it into prescribed stroke codes if it contains a specific stroke code. CONSTITUTION:The coordinate point data of each sampling point delivered from a coordinate input device 1 is discriminated in terms of a prescribed basic segment code to which said data belongs together with a series of segments obtained by the 1-writing stroke through a basic segment identifying circuit 2 to which the time series information on the writing point is delivered. Then the corresponding stroke code is delivered, and a segment transmitting circuit 3 stores the input stroke code into a segment code buffer 31. A control circuit 32 receives the control signal from a deciding circuit 5 and sends the stroke codes in the buffer 31 to a segment decomposing circuit 33 when a stroke decomposition request is given. These stroke codes, however, are sent directly to a comparator 4 when no decomposition request is given. The comparator 4 performs the comparison of the code with a dictionary memory 6 and counts error. The circuit 5 delivers the result of recognition in response to the count value.

Description

[Detailed Description of the Invention] Industrial Application Field The present invention relates to an online character recognition device that recognizes written characters based on the pen point coordinates of the characters written by the user and their time elapsed information. It is.

Conventional structure and problems Traditionally, online character recognition uses a pen amplifier,
Since it is possible to relatively easily extract line segments written with a single stroke (hereinafter referred to as strokes) using down information, input strokes written with a brush can be extracted from multiple basic line segments (hereinafter referred to as strokes) selected in advance. A known method is to check which of the basic strokes (hereinafter referred to as basic strokes) matches (stroke recognition) and to recognize a character using this stroke information. An advantage of the above-described method of performing stroke recognition in advance is that, especially in handwritten kanji recognition, the number of strokes of input characters can be used to limit the range of dictionary search, thereby significantly shortening the recognition time.

However, conventional online character recognition devices have the following drawbacks. Figure 1 (a), (b)
.

(C), (d), and (e) show examples of various zokuha characters in handwritten kanji. In the case of handwritten characters, even if the scribe is instructed not to write the next character, there is a high possibility that the next character will be written as shown in Figure 1, and it is also common for the scribe to write the next character. Scribes who have a habit of writing often feel a great deal of resistance to consciously writing letters with the correct number of strokes.

In this way, when the input character data is expected to be a character written in the character ``Zoku'', a method that uses the stroke information obtained by conventional stroke recognition as is for character recognition will not misrecognize all the characters ``Zoku''. Even if a method was adopted in which all the stroke information of the expected Zokuha character pattern was stored in the dictionary, there was a practical problem.

OBJECTS OF THE INVENTION An object of the present invention is to eliminate the drawbacks of the conventional example described above, and to provide a stable recognition method even for characters written in Japanese characters, without adding a dictionary or increasing recognition time. The purpose of the present invention is to provide an online character recognition device based on

Composition of the Invention The online character recognition device of the present invention includes a coordinate input device into which the pen point coordinates of handwritten characters and time series information of each pen point are input;
a basic line segment identification circuit that determines which of a plurality of pre-selected basic line segments each line segment written with one stroke constituting a handwritten character belongs to and assigns a corresponding basic line segment code; a line segment code sending circuit that stores and sequentially sends out the basic line segment codes output from the basic line segment identification circuit; and basic line segments that correspond to the line segments that constitute each character of the character group to be recognized. a dictionary storage section that stores and sequentially sends out codes and time-series information of each writing point; and basic line segment codes of input handwritten characters that are sequentially sent out from the line segment code sending circuit and sequentially sent out from the dictionary storage section. a comparison circuit that checks for a match between basic line segment codes and outputs a line segment code match signal; and a determination circuit that identifies characters in response to the line segment code match signal supplied from the comparison circuit; ,
The line segment code sending circuit has a line segment code buffer that stores the basic line segment code sent from the basic line segment identification circuit, and a line segment code that is selected in advance from among the basic line segment codes input to the line segment code buffer. A line segment that detects a line segment that matches a basic line segment code to be decomposed, decomposes it into a plurality of predetermined basic line segment codes for each type of basic line segment code to be decomposed, and sequentially outputs the decomposed basic line segment codes. This system includes a decomposition circuit, thereby ensuring reliable recognition of handwritten characters and obtaining good recognition results.

DESCRIPTION OF EMBODIMENTS Hereinafter, embodiments of the present invention will be described with reference to the drawings.

FIG. 2 is a block diagram of essential parts showing the basic configuration of an online character recognition device according to one embodiment ψ11 of the present invention.

In FIG. 2, 1 is a coordinate input device, 2 is a basic line segment identification circuit that inputs the output of the coordinate input device 1, and 3 is a line segment code sending circuit that inputs the output of the basic segment identification circuit 2 and sequentially outputs the output. , 4 are the output of the line segment code sending circuit 3 and the dictionary storage section 6
A comparison circuit 6 compares the outputs of the comparison circuit 4, a determination circuit 6 receives the output of the comparison circuit 4 and identifies characters, and a dictionary storage section 6 sequentially sends stored dictionary data to the comparison circuit 4.

The operation of the online character recognition device of this embodiment configured as described above will be described below.

First, the coordinate input device 1 is a device that chronologically samples a sufficient number of two-dimensional coordinate points to represent the shape of a handwritten character, and may include a so-called digitizer and a pen. , a cathode ray tube (CRT) and a light pen. To facilitate the explanation below, the kanji ``1 stone''
The operation of each part will be explained using as an example.

First, the basic line segment identification circuit 2 is supplied with the coordinate point data of each sampling point output from the coordinate input device 1 and the time series information of the writing point (for example, the amplifier of the pen, the down state, the moving direction of the pen, etc.) In this case, - the stroke of the brush, that is, a series of line segments obtained by interpolating between each sampling point from the first sampling point when the pen is down to the sampling point immediately before the next pen-up, is selected in advance. It is determined which of the basic line segment codes (hereinafter referred to as basic stroke codes) it belongs to, and the corresponding stroke code is output. The following methods are available for assigning one basic stroke code to one stroke of an input character.

First, a pre-selected basic line segment that characterizes the character to be recognized is divided into several straight line elements, a certain direction code is given to each line element, and the arrangement of the direction codes is used to
Define the basic stroke code for this basic line segment.

Next, when the time series information at each sampling point of input handwritten characters is input, one of the direction codes used to define the basic loaf code above is applied to the straight line element connecting each sampling point. Each brushstroke is represented by a sequence of direction codes. Next, the direction code string of each stroke of this input character is compared with the direction code string of the basic stroke code selected in advance,
The basic stroke for which the total length of input straight line elements that do not match is the smallest is determined to be the basic stroke with the highest degree of matching with the input stroke, and is output. FIG. 3(a) shows the definition of the direction code in this embodiment, and FIG. 3(b) shows an example of the arrangement of the basic stroke shape, the stroke code, and the direction code that defines the stroke code.

Figure 4 shows the correspondence between the character shape and the stroke code of each stroke identified by the basic line segment identification circuit 2 when ``stone'' is written down by a scribe with the correct number of strokes. FIG. 5 shows the correspondence between character shapes and stroke codes when written. Note that the numbers in these figures represent stroke codes.

The stroke codes of the input characters of the glyph shape shown in Figure 4 are identified as rl +314+7+ 1 J in the order of writing, and the 5th
In the case of the character shape shown in the figure, it is [1, 3, 4, 21J.

In the line segment sending circuit 3, the stroke codes sequentially sent from the basic line segment identifying circuit 2 are stored in a line segment code buffer 31.
to be memorized. That is, the stroke code string "1, 3, 4,
7. The secondary control circuit 32 in which IJ or "1,3°4.21" is stored receives the control signal sent from the determination circuit 6, and if stroke decomposition is requested, the line segment code no. The stroke codes in the line segment code buffer 31 are sequentially sent to the line segment decomposition circuit 33, and when stroke decomposition is not requested, the stroke codes in the line segment code buffer 31 are directly sent to the comparison circuit 4. . Cases in which stroke decomposition is required include cases in which a recognized character is still not obtained even after matching with all the dictionary data to be searched;
There may be a case where it is determined that the obtained result is likely to be a misrecognition. For example, when the comparison circuit 4 compares the stroke code of an input character with the stroke code sent from the dictionary storage unit 6, if there is a stroke code that does not match, an error count is added, and the error count is determined in advance. If the characters are kept as candidate characters if they are less than a certain number, the error count will be set to "o" when the search for all dictionary data is completed.
If a candidate character exists, the character code is output, but if there are only candidate characters of "1" or more, it is determined that there is a high possibility of misrecognition, and a control signal is sent to request stroke decomposition. Perform matching with the dictionary again.

Assuming that no signal requesting stroke decomposition is being sent, the line segment code sending circuit 3 sends the stroke code [1, 3, 4, 7] of the input character in the i-minute code bath
, I J or [1,3°4.21J are sent out in sequence. In the dictionary storage unit 6, stroke codes characteristic of each character to be recognized are written in consideration of the order in which they are written. The description of the stroke code string in the one-character dictionary may allow the possibility of multiple stroke codes, or may partially include a description that does not fix the order in which the strokes are written. , there may be a description in which a group of stroke code strings is replaced with another single code, and information representing the characteristics of characters outside the stroke code (for example,
(The direction of movement force during pen-off between strokes, the direction and distance between specific sampling points, the size of coordinate values, etc.) may also be included. In this embodiment, the dictionary is divided into a plurality of blocks according to the number of strokes of the character to be recognized, and signals indicating strokes and the number of strokes are received from the line segment code sending circuit 3 to specify the dictionary block to be searched. It shall be configured as follows. The structure of the dictionary is shown in FIG.

For convenience of explanation, the following explanation will be given assuming that the dictionary for the character "stone" is written as [1, 3, 4, 7, I J] only by stroke codes. First, the characters with the shape shown in Figure 4 (stroke code string ``' + 3 + ' +
When the stroke number 6 of 7+'J) is sent from the line segment code sending circuit 3, the comparison circuit 4 stores the number of strokes in the dictionary storage section
One piece of dictionary data is taken from the beginning of the six-stroke character dictionary and checked to see if it matches the stroke code.

Assume that the description in the first dictionary is "1, 3, 1. 4. I J. 1. First, since the input stroke code and the dictionary stroke code are both 11" and match, the next stroke code is checked. The second stroke code also matches. Examining the next stroke code, the input end is "4",
The dictionary side is ``11'' and does not match, so set the error count to ``1''.
”.

Below, check the match of the stroke codes in the same way. In this example, the stroke codes that do not match are the 3rd and 4th strokes.
Since there are two strokes, a signal indicating a matching state with an error count of "2" is sent to the determination unit 6. The determining unit 6 identifies characters based on this matching signal.

That is, if it is an error callan)rOJ, the character code of the character whose correspondence is currently checked is output as the recognition result, and character recognition is ended. Further, if the error count is "11", it is registered as a candidate character, and if it is "2" or more, it is rejected as there is no possibility. The error count is now "2"
Therefore, the first character written in the dictionary is rejected, and matching with the next dictionary is repeated in the same way.

At this time, a signal requesting stroke decomposition is not sent.When matching with the dictionary for the zero character "stone" is completed,
When the error count reaches "o", a signal indicating that recognition of the character code "stone" and the input character has been completed is output, and the recognition of the input character is completed. Next, in the case of the character Zoku shown in Figure 6, the number of strokes "4" is input, so
Matching of the input stroke code "1° 3.4, 21J with the dictionary in the 4-stroke dictionary block is performed. Since the dictionary for "mojiriishi" is in the 5-stroke dictionary block, 4-stroke dictionary 1
If there is a character with an error count of "0" in
The result is output, but the result obtained is a false recognition. Suppose now that there is no character with an error count of "0" even after the search of the four-stroke dictionary block is completed. The determining unit 6 receives a signal indicating that the first dictionary search has been completed, and outputs a signal requesting stroke decomposition to the line segment code sending circuit 3.

Based on this signal, the line segment code sending circuit 3 sequentially decomposes the input stroke codes in the line segment code buffer 31 into line segment codes.
2, and first, it is checked whether the stroke code selected in advance to be decomposed is included. If it is not included, a signal indicating that it was not detected is sent to the determination circuit 6, and the determination circuit 6 either outputs the code of the candidate character obtained by the first dictionary search, or outputs the stroke that should be decomposed into the candidate character. If a code is detected, it is decomposed into a plurality of pre-selected stroke codes for each stroke. Examples of strokes to be decomposed are shown in FIGS. 7(,) to (d). In this embodiment, the stroke code "21" is included in the stroke code to be decomposed.
" is included, and the stroke code "7" is included in advance.
and "1" [Figure 8 (a)
, see (b)]. Since "21" is included in the stroke code of the input character, it is decomposed into stroke codes "7" and "1" by the line segment decomposition circuit 32, and the input stroke code string becomes Jl, 3, 4, 7.1j.

Thereafter, in the same way as the first dictionary search, matching with the six-stroke character dictionary is performed, and the character "stone" is recognized with an error count of "0". After stroke decomposition, if a character with an error count of "0" does not exist even after completing the second dictionary search, a stroke decomposition request signal is not issued, and only one of the first and second candidates is searched. Output the character with the lowest error count as the recognized character. Further, the stroke code to be decomposed can be determined using the stroke codes before and after the stroke code to be decomposed and other time-series information at each writing point as the determination conditions. Figure 9 shows the basic configuration program in this case.
Show the diagram. For example, FIGS. 10(a) and 10(b) are examples of applying the rule that only the first stroke written for stroke code 10 is decomposed, and FIG. 11(a)
), (1)) and Figure 12(a).

(b) For the u stroke code "21", an example is shown in which the stroke code to be decomposed is determined by the immediately preceding stroke code 0 Effects of the Invention As is clear from the above description, the present invention provides a coordinate input device and a basic line segment identification circuit. , an online character recognition device that includes a line segment code sending circuit, a dictionary storage unit, and a comparison circuit 1 determination circuit, the line segment code sending circuit has a line segment code buffer that stores the stroke code of an input character, and a line segment code buffer that stores the stroke code of an input character. If the stroke code in the sofa contains a specific stroke code, it is configured with a line segment decomposition circuit that decomposes it into a pre-selected stroke code and sends it out if necessary. , it is possible to recognize the ``Zoku'' character, which was considered difficult with the conventional online character recognition method using stroke extraction, without the need for an additional dictionary. An excellent effect 75 can be obtained in that the recognition speed can be made possible without significantly reducing the recognition speed. Due to this effect, scribes can input characters without being too conscious of the fact that strokes are connected in one part, and even scribes who have a habit of writing characters can use it without any hesitation. Providing an online character recognition device7
I can do 52.

Furthermore, by adopting a configuration that determines whether or not to send the stroke code in the line segment code buffer to the line segment decomposition circuit based on the control signal from the first determination circuit, there is a risk of misrecognition due to decomposition of more strokes than necessary. Sex and
Another advantage is that the reduction in recognition speed caused by checking the stroke code can be avoided.

Furthermore, by configuring the line segment decomposition circuit with a condition judgment unit that determines the strokes to be decomposed before and after the stroke code to be decomposed, the code and other time-series information, and a line segment decomposition unit, it is possible to It is possible to strictly distinguish between stroke codes and other stroke codes, and to perform more accurate stroke code decomposition, resulting in the effect of improving the recognition rate.

[Brief explanation of the drawing]

Figures 1 (a) to (,) are glyph diagrams showing examples of glyphs;
The figure is a block diagram of main parts showing the basic configuration of an online character recognition device according to an embodiment of the present invention, FIG. 3(a) is a definition diagram of a direction code, and FIG. 3(b) is a basic stroke type and stroke code. and a correspondence diagram of the arrangement of direction codes that define stroke codes, Figure 4 is a correspondence diagram of character shapes with the correct number of strokes and stroke codes, Figure 6 is a correspondence diagram of character shapes and stroke codes, and Figure 6 is a dictionary. Detailed configuration diagram of the storage unit, FIGS. 7(a) to (d) are diagrams of the shape of strokes to be decomposed, FIGS. 8(a) and (b) are diagrams of the shape showing an example of stroke decomposition, and FIG. 9 is a main part block diagram showing the basic configuration of an online character recognition device according to another embodiment of the present invention,
Each (a) of FIG. 10, FIG. 11, and FIG. 12,
(b) is a feature diagram showing an example of stroke decomposition using determination conditions. DESCRIPTION OF SYMBOLS 1... Coordinate input device, 2... Basic line segment identification circuit, 3... Line segment code sending circuit, 4... Comparison circuit, 6... Judgment circuit, 6... ...Dictionary storage section, 31.
... Line segment code buffer, 32... Line segment decomposition circuit, 33... Control circuit, 32a... Condition determination unit,
32b・Line segment decomposition part. Name of agent: Patent attorney Toshio Nakao and 1 other person No. 1
Figure-) 3R Figure 3 (C1) Figure 4 Figure 6 Figure 7 (θ) /b) Slow L stroke 3
-F 7 / -559- Moon, Fig. 10, Fig. 11), Loaf cord 2f Fig. 12, μ Totakot Do I

Claims (2)

[Claims] (1) A coordinate input device into which the pen point coordinates of handwritten characters and time-series information of each writing point are input, and a plurality of bases in which each line segment written with one stroke constituting handwritten characters is selected in advance. a basic line segment identification circuit that determines which line segment it belongs to and assigns a corresponding basic line segment code; and a line that stores and sequentially sends out the basic line segment codes output from the basic line segment identification circuit. a minute code sending circuit; a dictionary storage unit that stores and sequentially sends out basic line segment codes corresponding to line segments constituting each character of the character group to be recognized and the time-series path lines of each writing point; a comparison circuit that checks whether the basic line segment codes of input handwritten characters sequentially sent from the line segment code sending circuit match the basic line segment codes sequentially sent from the dictionary storage unit and outputs a line segment code matching signal; The line segment code sending circuit includes a determination circuit that fixes a character in response to a line segment code matching signal supplied from the comparison circuit, and the line segment code sending circuit is configured to detect the basic line segment sent from the basic line segment identification circuit. A line segment code buffer that stores codes, and a line segment code that matches a preselected basic line segment code to be decomposed from among the basic line segment codes input to the line segment code buffer, and the code to be decomposed. For each type of basic line segment code,
An online character recognition device characterized by comprising a line segment decomposition circuit that decomposes into a plurality of predetermined basic line segment codes and sequentially outputs them. (2) The determination circuit performs character identification in response to the line segment code matching signal output from the comparison circuit, and outputs a control signal indicating whether character identification has been completed or not yet completed. The line segment code output circuit is configured to either directly send the basic line segment code in the line segment code buffer to the comparison circuit in response to a control signal sent from the determination circuit, or output the basic line segment code from the line segment code to the comparison circuit. The present invention is characterized in that it includes a control circuit that sends the basic line segment code in the buffer to the line segment decomposition circuit and determines whether to send the output from the line segment decomposition circuit to the comparison circuit. An online character recognition device according to claim (1). (39 line segment decomposition circuits, when a preselected basic line segment code to be decomposed is supplied, are supplied from the basic codes before and after the basic line segment code to be decomposed and the coordinates manually set. a condition determination section that determines a basic line segment code for decomposing the basic line segment code to be decomposed using time series information of each writing point as a determination condition; Claim (1) or Claim 1, characterized in that the line segment decomposition unit that decomposes the basic line segment code to be decomposed into a plurality of pre-selected basic line segment codes is constituted by Mame. The online character recognition device described in (2).