JP2007220145A - Online handwriting recognition apparatus - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To automatically recognize character strings even when a user simply successively writes characters without being conscious of end of the character. <P>SOLUTION: A handwriting information acquiring means 202 takes the strokes of the characters written on a tablet 201 in a buffer 208 by unit of one stroke. A candidate character string generating means 203 generates the candidate character string to the input strokes in the buffer 208 and stores the the candidate character string in a buffer 209. A character string likelihood calculating device 204 generates a dictionary of character string structure by combining corresponding dictionary information in the dictionary of character structure 211 and corresponding dictionary information in a dictionary of biliteral structure 212 with respect to the respective candidate character strings in the buffer 209 and collates the dictionary of character string structure with the input strokes. An input string estimating means 205 estimates a character string part in which results of character segmentation from the start are common as a recognition determined character string for each recognition candidate character string in the buffer 209. The determined character string is displayed on a display 213. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to an on-line character recognition device for automatically cutting out and recognizing characters as needed from character string handwriting (stroke sequence) information input by handwriting via a coordinate input device, and displaying the result, in particular, continuous characters. The present invention relates to an on-line character recognition apparatus suitable for writing input.

2. Description of the Related Art Conventionally, an online character recognition device that automatically cuts out and recognizes characters from character handwriting information input by handwriting via a coordinate input device such as a coordinate input tablet and displays the result, the following four recognition technologies ( Recognition method), that is, (1) A technique in which a plurality of character frames are prepared and one character is sequentially written in one character frame to cut out and recognize the characters. (2) A pen is moved from the coordinate input tablet. Technology that detects the end of a certain period of time, recognizes the end of the character and cuts out and recognizes the character (3) Cuts out the character by explicitly instructing the end of one character writing with the recognition execution button, etc. Recognizing technology (4) A technology that defines a character with a special stroke, detects that a stroke has been input and the pen has left the tablet, determines that the character has been input, and cuts out and recognizes the character. It is common to apply a deviation.

  However, the conventional recognition technique described above, for example, (1) has a problem that the user must input with consciousness of the size and position of the character frame at the time of writing, and cannot perform natural input. In addition, when considering mounting on a small information device, a sufficiently large character writing area cannot be secured. Therefore, if a plurality of character frames are prepared, the size of the writing area per character is reduced, and writing is performed for the user. There was also a problem that it became difficult.

  On the other hand, in the conventional recognition techniques (2) and (3), since it is not necessary to prepare a plurality of character frames, a writing area per character can be increased, and the problem (1) can be avoided. Is possible. However, in (2), when it is desired to input a plurality of characters continuously, the pen must be separated from the tablet for a certain time or more for each character writing. Further, in (3), an operation that is essentially unrelated to character writing, that is, pressing the recognition execution button for each character input is required for each character writing. For this reason, the recognition techniques (2) and (3) have a problem that it is difficult to smoothly input a character string.

  Further, in the conventional recognition technique (4), although the problems in (1), (2), and (3) described above can be avoided, a stroke of each character is memorized in advance for the user. Force to leave. For this reason, particularly when there are a large number of characters to be recognized, it is very painful for the user.

The present invention has been made in consideration of the above circumstances, and its purpose is to allow a user to automatically recognize a character string even if the user simply writes characters continuously without being aware of character separation. Is to make it.
Another object of the present invention is to enable overwriting input of characters.

  The present invention recognizes a character on the basis of a stroke sequence expressed by a coordinate sequence of a pen tip detected by the coordinate input device from when the pen touches the coordinate input device until the pen leaves the coordinate input device. An on-line character recognition device that performs recognition and displays a recognition result on a display device, each time the stroke is detected, the shape of the stroke detected so far and the combination of each two characters that are subsequently written An optimum character string is obtained based on the positional relationship of the stroke between the two characters, and the character string is displayed on the display device. According to the present invention, the most probable character string (recognized confirmed character string) is estimated and displayed even if the user continues to write characters on the coordinate input device without being aware of character separation. Is possible. Here, since the positional relationship of strokes constituting each character between characters is also taken into consideration, it is possible to input characters by overwriting. Therefore, it is possible to provide comfortable handwritten character input means even in a small information device or the like in which only a small writing space can be prepared.

  Further, the present invention is based on a sequence of strokes expressed by a coordinate series of a pen tip detected by the coordinate input device from when the pen touches the coordinate input device until the pen leaves the coordinate input device. An on-line character recognition device that performs character recognition and displays a recognition result on a display device, wherein the coordinate input device has a character writing area for writing one character with the pen, and the writing Each time the stroke written in the area is detected, the most suitable character based on the positional relationship of the stroke between the two characters with respect to the combination of the shape of the stroke detected so far and each of the two characters that are overwritten A character string is obtained and the character string is displayed on the display device. According to the present invention, the most probable character string (recognition confirmed character string) is estimated and displayed even if the user writes the characters on the coordinate input device without being aware of the character separation. Is possible. Since the positional relationship of strokes constituting each character is also taken into consideration, it is possible to input characters by overwriting. Therefore, it is possible to provide comfortable handwritten character input means even in a small information device or the like in which only a small writing space can be prepared.

  An on-line character recognition device according to the present invention includes a character structure dictionary that registers, for each of a plurality of recognizable characters, shape information of strokes constituting the character, and dictionary information describing a structural relationship between the strokes, For each combination of two characters to be written successively among a plurality of recognizable characters, the intercharacter structure dictionary in which dictionary information describing the structural relationship between the two characters is registered, and the pen on the coordinate input device Each time a stroke is captured by the handwriting information acquisition unit, the handwriting information acquisition unit sequentially captures a stroke expressed by the coordinate series of the pen tip detected by the coordinate input device until the coordinate input device is released. In addition, a recognition candidate that generates a recognition candidate character string that can be a recognition candidate for an already-acquired stroke sequence to be recognized including the stroke. For each of the generated recognition candidate character strings, the character string generation means and the character spacing between the two candidate characters that are continuous with the dictionary information in the character structure dictionary for each candidate character constituting the recognition candidate character string. Based on the dictionary information in the structure dictionary, a character string structure dictionary corresponding to the recognition candidate character string is created, and the captured stroke string is compared with the character string structure dictionary for each recognition candidate character string. Thus, collating means (character string likelihood calculating means) for cutting out characters for determining which strokes in the stroke sequence constitute which candidate characters, and each recognition candidate character string from the top An input character string estimation means for estimating and outputting a character string portion having the same character extraction result as a recognition confirmed character string, and displaying the estimated recognition confirmed character string Characterized by comprising a recognition result displaying means.

  In such a configuration, each time a stroke (handwriting) is input, for each recognition candidate character string for the recognition target stroke string with the stroke as the last stroke, corresponding dictionary information and inter-character structure in the character structure dictionary A character string structure dictionary combining the corresponding dictionary information in the dictionary is generated, and the character string structure dictionary for each recognition candidate character string is collated with the recognition target stroke string (input handwriting). Even if characters are continuously written on the coordinate input device without being aware of the delimiter, it is possible to estimate and display the most likely input character string (recognition confirmed character string). Here, since the structural relationship between characters is also taken into consideration, it is possible to input characters by overwriting by preparing an inter-character structure dictionary on the premise of overwriting of characters. Therefore, it is possible to provide comfortable handwritten character input means even in a small information device or the like in which only a small writing space can be prepared.

  Here, a stroke buffer for storing a stroke (input stroke) captured by the handwriting information acquisition unit and a recognition candidate buffer for storing a recognition candidate character string generated by the recognition candidate character string generation unit are provided. If the collation result of the stroke string (input stroke string) by the collation means and the character string structure dictionary of each recognition candidate character string is stored in the recognition candidate buffer in association with the recognition candidate character string, It is possible to efficiently perform processing in the online character recognition device.

  In addition, when the input stroke is stored in the stroke buffer by the handwriting information acquisition unit, the feature (for example, shape feature) of the stroke and the feature (for example, the structural feature) between the stroke and the preceding stroke are acquired. In the recognition candidate character string generation means, the recognition candidate character string for the stroke string in the stroke buffer is extracted from the features of each stroke constituting the stroke string extracted by the handwriting information acquisition means and the strokes. It is good to generate based on the feature.

  Further, in the collation processing by the collation means, the probability that the input stroke string is written as a recognition candidate character string (for example, likelihood) is calculated, and the input character string estimation means Two means to be described, that is, a recognition-determined character string estimation unit that estimates and outputs the recognition-determined character string, a recognition-unrecognized character string that recognizes and outputs a recognition-unrecognized character string and recognized-unrecognized stroke string information In addition to the recognition confirmed character string output from the recognition confirmed character string estimation means, the recognition result display means estimates the recognition unconfirmed character string and the recognition unconfirmed stroke string information. The recognized unconfirmed character string and the recognized unconfirmed stroke string information output from the means may be displayed.

  In such a configuration, even if the user continuously writes characters on the coordinate input device by overwriting, each time one stroke is written, the recognized confirmed character string, the unrecognized character string and the unrecognized character string at that time are not recognized. Since the confirmed stroke string information is displayed, the recognition processing result of the handwriting can be confirmed sequentially on the screen, and comfortable character input can be performed. Here, it is preferable to add a display attribute that allows recognition-recognized character strings and unrecognized character strings to be displayed. Also, the recognition unconfirmed stroke sequence information may be displayed so that the presence or absence of the recognition unconfirmed stroke sequence can be identified.

  Further, each time the recognition confirmed character string is displayed by the recognition result display means, the input stroke string corresponding to the recognition confirmed character string is deleted from the stroke buffer, and the recognition confirmed character string is changed from the recognition candidate buffer to the recognition confirmed character string. If buffer updating means for removing the corresponding information is provided, the amount of processing can be reduced.

  In addition, after all the recognition candidate character strings in the recognition candidate buffer are collated with the input stroke string in the stroke buffer, the input stroke string obtained for each recognition candidate character string by this collation is recognized. The probability (likelihood) written as a candidate character string is compared with a predetermined probability (likelihood threshold), and all recognition candidate character strings with low probability are deleted from the recognition candidate buffer. In this way, the processing amount can be reduced.

  In addition, after the last stroke is input, detection means (pen operation determination means) for detecting that there is no next stroke input for a predetermined time or more, and according to the detection result of the detection means Provided with a recognized character string determining means for outputting the unrecognized character string as the recognized confirmed character string, the operability of the user's character input can be further improved.

  Here, the coordinate input device is arranged so as to overlap the display surface of the display device, and the character writing region and the recognition result display region are separately secured on the input surface of the coordinate input device, and You may make it provide the detection means (pen operation determination means) which detects that the pen touched this character writing area.

  In addition to this, a common area serving as both a character writing area and a recognition result display area is secured on the input surface of the coordinate input device, and instead of the detection means and the recognized character string determination means, When the tip of the pen exists within a certain coordinate range of the common area for a certain period of time, the pen input is a gesture for the purpose of moving the cursor in the recognition result display. Determining means (pen operation determining means) for determining that the input is the intended input, and when the determining means determines that the pen input is a gesture, the recognized unconfirmed character string is output as the recognized confirmed character string. And a recognized character string determining means may be provided.

  Further, when one character backward deletion instructing means for instructing one character backward deletion, and when one character backward deletion is instructed by this one character backward deletion instructing means, If the recognized unconfirmed stroke string is deleted, and there is no recognized unconfirmed stroke string and there is a recognized unconfirmed character string, the end of the recognized unconfirmed character string is deleted backward one character and the remaining recognition unconfirmed It is preferable to provide a recognition result editing unit that deletes the recognized confirmed character string backward by one character when the character string is a recognized confirmed character string and neither the recognized unconfirmed stroke string nor the recognized unconfirmed character string exists.

  In such a configuration, it is possible to smoothly perform character deletion and character writing operations.

  In addition, the recognition unconfirmed character string estimated this time is used when the probability of the recognition candidate character string including the recognition unconfirmed character string satisfies a predetermined condition, or the recognition unconfirmed character string estimated this time When the certainty of the recognition candidate character string including the condition satisfies a predetermined condition based on the certainty of the recognition candidate character string including the previously recognized unconfirmed character string, the previously unrecognized The recognized unconfirmed character string estimated this time may be displayed instead of the confirmed character string. Each time one stroke is input, the optimum recognition result including the input stroke and the input stroke input so far can be sequentially displayed. Therefore, the user can confirm the recognition result immediately after writing. It is possible to input handwritten characters efficiently.

  In addition, a character writing area for writing characters with the pen of the coordinate input device is arranged so as to overlap the handwriting display area of the display device, and this stroke is included each time the stroke is written in the character writing area. By displaying the latest predetermined N strokes (where N is an integer) in the handwriting display area, the input handwriting is appropriately displayed, so that a comfortable character writing environment can be provided to the user.

  In addition, after the pen leaves the coordinate input device, detection that the pen does not touch the coordinate input device for a predetermined time or more, or detects a pen operation other than writing of the stroke And a display of strokes displayed in the handwriting display area may be erased according to the detection result of the detection means.

  Further, when displaying the predetermined number of strokes in the handwriting display area, the stroke color and thickness and the type of line representing the stroke may be displayed differently. Good.

  Each of the above means can be realized by a hardware configuration, but can also be realized by reading and executing a program for causing the computer to function as the means. In this case, the online character recognition device of the present invention can be easily realized by an information device such as a portable information terminal.

The present invention relating to the above-described online character recognition apparatus is also established as an invention relating to a method (online character recognition method).
The present invention also provides a computer for executing a procedure corresponding to the invention (or for causing a computer to function as each means corresponding to the invention, or for causing a computer to realize a function corresponding to the invention). ) It is also established as an invention relating to a computer-readable storage medium storing a program, and further as an invention relating to the program itself.

  As described in detail above, according to the present invention, when a user writes a character string, there is no need to be aware of character separation, and even if characters are simply written continuously, the structural relationship between characters The character string can be automatically recognized by the recognition process considering the above, and therefore, smooth continuous writing input of characters can be realized.

  Furthermore, according to the present invention, since it is possible to input characters by overwriting, a comfortable handwritten character input environment can be provided even in a small information device such as a portable information terminal in which only a small writing area can be prepared.

  Embodiments of the present invention will be described below with reference to the drawings.

(Configuration and operation of online character recognition device)
FIG. 1 shows a hardware configuration of an information device that realizes an online character recognition apparatus according to an embodiment of the present invention.
1 is a small information device such as a portable information terminal, for example, and includes a CPU 1, a ROM 2, a RAM 3, a tablet (tablet device) 4, a display (display device) 5, a data reading device 6, a storage medium 7, and an interface. Part (I / F) 8, 9, 10 and the like.

The CPU 1 is the center of the information device of FIG. 1, and is a calculation / control unit that performs calculations for various data processing and controls the entire system.
A ROM 2 is a non-volatile memory (storage medium) in which basic software and the like are stored in advance, and a RAM 3 is a volatile memory used for a work area of the CPU 1.
The tablet 4 is a coordinate input device capable of inputting handwriting data, and the display 5 is a display device such as a liquid crystal display for displaying various information.

  The storage medium 7 is computer-readable storage means such as a hard disk, floppy (registered trademark) disk, CD-ROM, or optical disk. The storage medium 7 stores data and programs. The data reading device 6 is data reading means such as a hard disk drive, a floppy (registered trademark) disk drive, a CD-ROM drive, and an optical disk drive that can read data and programs stored in the storage medium 7.

  The interface unit 8, the interface unit 9, and the interface unit 10 interface with the tablet 4, the display 5, and the data reading device 6, respectively.

  The storage medium 7 stores in advance an online character recognition program for performing character recognition of the input handwriting as the program. This online character recognition program is read by the data reading device 6 and stored in the RAM 3 when the device is activated. The CPU 1 performs character recognition processing of the input handwriting according to the online character recognition program stored in the RAM 3.

  The online character recognition program may be stored in the ROM 2 in advance. In this case, the CPU 1 may perform character recognition processing of the input handwriting according to the online character recognition program in the ROM. Further, when the apparatus is activated, an online character recognition program may be transferred from the ROM 2 to the RAM 3 and the character recognition processing of the input handwriting may be performed according to the online character recognition program in the RAM 3. The online character recognition program may be downloaded via a communication line.

  The tablet 4 is transparent, for example, and is arranged on the display screen of the display 5. A character writing area 301 (see FIG. 5) is provided on the input surface of the tablet 4. When the user writes a character in the writing area using the dedicated pen P, handwriting data (time-series information of coordinate values) is detected by the tablet device 4 and transferred to the CPU 1 through the interface unit 8.

FIG. 2 is a block diagram showing a functional configuration of the online character recognition apparatus realized by the information device of FIG.
This online character recognition device includes a handwritten character input device 101 corresponding to the tablet 4 in FIG. 1, handwriting information acquisition means 102, recognition means 103, recognition result display means 104, function structure dictionary 106, and character structure dictionary 106. Each dictionary of the interstructure dictionary 107 and a display 105 corresponding to the display 5 in FIG.

  The handwriting information acquisition unit 102, the recognition unit 103, and the recognition result display unit 104 are realized by the CPU 1 in FIG. 1 reading and executing an online character recognition program stored in the RAM 3. Various data generated (generated) during the character recognition process are temporarily stored in, for example, the RAM 3.

  The character structure dictionary 106 and the intercharacter structure dictionary 107 are stored in advance in the storage medium 7 in FIG. 1, for example, together with the online character recognition program, and when the program is read by the data reading device 6 and stored in the RAM 3. The data is stored in the RAM 3. Note that the character structure dictionary 106 and the intercharacter structure dictionary 107 can be stored in the ROM 2 in advance.

  The handwritten character input device 101 is a tablet, for example, and samples two-dimensional coordinate data representing the position of the pen tip while the pen P is touching the tablet at regular time intervals. The obtained coordinate data is sent to the handwriting information acquisition means 102.

  The handwriting information acquisition means 102 treats the coordinate data string from when the pen P touches the tablet until it leaves, that is, the handwriting coordinate data string as a set of data called a stroke, and acquires it as stroke data. Each time stroke data is acquired by the handwriting information acquisition means 102, it is sent to the recognition means 103.

  Each time stroke data is input, the recognizing unit 103 recognizes an optimal character string from the stroke data input so far using the character structure dictionary 106 and the inter-character structure dictionary 107.

  The character structure dictionary 106 is data (character structure dictionary information) in which the structure of each character to be recognized is expressed, that is, for each character, the shape of the stroke constituting the character, the positional relationship (structure) between the strokes, etc. This is a dictionary in which feature information is registered.

  The inter-character structure dictionary 107 is a data (character-to-character dictionary) expressing the structural relationship between two characters for each combination of two characters written in succession among a plurality of characters registered in the character structure dictionary 106. Information).

  In some cases, two consecutive characters are written side by side on the tablet, or when they are written in layers (for example, when there is only one character writing area for writing characters with a pen). is there. The former inter-character dictionary information is characteristic information of the positional relationship (structure) between the stroke of one character and the stroke of the other character when two characters are written side by side. The inter-character dictionary information in this case is characteristic information of the positional relationship (structure) between the stroke of one character and the stroke of the other character when two characters are written in an overlapping manner.

  The recognizing unit 193 uses the character structure dictionary 106 and the inter-character structure dictionary 107 each time stroke data is input, and is most probable based on the stroke shape and the positional relationship between the strokes input so far. The optimal character string is to be calculated.

The recognition result display unit 104 outputs the optimum character string obtained by the recognition unit 103 to the display 105.
In addition, there are a tablet character writing area in which a writing area for a plurality of characters is secured and a writing area for only one character is secured in either case. Is applicable. The difference between the two is only the inter-character dictionary information.

  Next, the procedure of online character recognition processing in the online character recognition apparatus configured as shown in FIG. 2 will be described with reference to the flowchart of FIG.

  In step S <b> 1, when one stroke is written in the character writing area, a coordinate data string of the stroke, that is, stroke data is captured by the handwriting information acquisition unit 102 and sent to the recognition unit 103.

  In step S2, the recognition unit 103 recognizes an optimum character string from the stroke data input so far, using the character structure dictionary 106 and the inter-character structure dictionary 107. For example, with this recognition process, of the strokes that have been input so far, it is possible to conclude that the character string has been written, the most likely character (unrecognized character), and the character in the middle of writing. Stroke (recognition unconfirmed stroke) can be estimated.

  In step S <b> 3, the recognition result display unit 104 displays the optimum character string recognized by the character recognition unit 103 (for example, a recognized confirmed character string and a recognized unrecognized character string) on the display 105.

  The above steps S1 to S3 are repeated until the input of the stroke is completed or until all the input strokes input so far are estimated as the recognition confirmed character strings (step S4). As a result, the on-line character recognition device in FIG. 2 automatically recognizes the handwriting as a character string simply by continuously writing characters on the handwritten character input device 101 without being aware of character separation. can do.

FIG. 4 is a block diagram showing in more detail the functional configuration of the online character recognition apparatus realized by the information device of FIG.
This online character recognition device includes a tablet 201 corresponding to the tablet 4 in FIG. 1, handwriting information acquisition means 202, recognition candidate character string generation means 203, character string likelihood calculation means 204, input character string estimation means 205, buffer Updating means 206 and recognition result display means 207, each function of stroke buffer 208, recognition candidate buffer 209 and recognition result buffer 210, each of character structure dictionary 211 and inter-character structure dictionary 212, 1 is composed of a display 213 corresponding to the display 5 in 1.

  The CPU 1 in FIG. 1 stores the handwriting information acquisition unit 202, the recognition candidate character string generation unit 203, the character string likelihood calculation unit 204, the input character string estimation unit 205, the buffer update unit 206, and the recognition result display unit 207. This is realized by reading and executing the stored online character recognition program.

  The stroke buffer 208, the recognition candidate buffer 209, and the recognition result buffer 210 are secured on the RAM 3, for example.

  The character structure dictionary 211 and the character structure dictionary 212 are stored in advance in, for example, the storage medium 7 in FIG. 1 together with the online character recognition program, and when the program is read by the data reading device 6 and stored in the RAM 3. Are stored in the RAM 3. The character structure dictionary 211 and the inter-character structure dictionary 212 can be stored in the ROM 2 in advance.

  The handwriting information acquisition unit 202 in FIG. 4 corresponds to the handwriting information acquisition unit 102 in FIG. 2, and the recognition candidate character string generation unit 203, the character string likelihood calculation unit 204, the input character string estimation unit 205, and the buffer update in FIG. The means 206 corresponds to the recognition means 103 in FIG. 3, and the recognition result display means 207 in FIG. 4 corresponds to the recognition result display means 104 in FIG. 4 corresponds to the character structure dictionary 106 in FIG. 2, and the inter-character structure dictionary 212 in FIG. 4 corresponds to the inter-character structure dictionary 107 in FIG.

The tablet 201 samples two-dimensional coordinate data representing the position of the pen tip while the pen P is touching the tablet 201 at regular time intervals. The obtained coordinate data is sent to the handwriting information acquisition means 202.
The handwriting information acquisition unit 202 treats the coordinate data string from when the pen P touches the tablet 201 until it leaves, that is, the handwriting coordinate data string as a set of data called a stroke, and stores it in the stroke buffer 208.

The recognition candidate character string generation unit 203 generates a recognition candidate character string group that becomes a recognition candidate for a sequence of strokes (stroke data) in the stroke buffer 208 and stores the recognition candidate character string group in the recognition candidate buffer 209.
The character string likelihood calculating means 204 creates a collation dictionary (character string structure dictionary) for each recognition candidate character string in the recognition candidate buffer 209 from the character structure dictionary 211 and the intercharacter structure dictionary 212, and the stroke buffer. Matching with the stroke sequence in 208 is performed. For each recognition candidate character string, the character string likelihood calculating unit 204 obtains a probability (likelihood) that the stroke string is the character string as a matching result between the stroke string and the recognition candidate character string, and the stroke string. A character cutout result indicating which stroke in the map corresponds to which character is acquired. The character string likelihood calculating unit 204 stores the acquired collation result for each recognition candidate character string in the recognition candidate buffer 209 in association with the recognition candidate character string.

  The input character string estimation unit 205 obtains a recognition confirmed character string, a recognition unconfirmed character string, and a recognition unconfirmed stroke string information from the recognition candidate character string in the recognition candidate buffer 209 and its collation result, and stores it in the recognition result buffer 210. Store. The recognition-determined character string is a character string portion that can be determined that this character string has been input, among the character strings configured by the stroke string (input stroke string). The unrecognized character string is a character string portion that cannot be determined but is estimated to have been input from among the character strings formed from the input stroke string. The unrecognized stroke sequence information is information indicating whether or not a stroke estimated to be a stroke in the middle of writing one character exists in the input stroke sequence.

The recognition result display means 206 displays the estimated input character string (recognized confirmed character string, recognized unconfirmed character string, and recognized unconfirmed stroke string information) in the recognition result buffer 210 estimated by the input character string estimating means 205 on the display 213. Output to.
Based on the information in the recognition result buffer 210, the buffer update unit 207 deletes information on a portion corresponding to the recognized character string from the stroke buffer 208 and the recognition candidate buffer 209 and updates the content.

FIG. 5 shows the external appearance of the information device of FIG.
As shown in the figure, on the main surface of the information device of FIG. 1, that is, the surface on which the transparent tablet 4 (201) is laminated on the display screen of the display 5 (213), 201) a character writing area 301 for writing a character, a recognition result display area 302 for displaying a result of recognizing a handwriting written in the character writing area 301 as a character string and a cursor C indicating a character insertion position; A one-character backward deletion button 303 for instructing deletion of the character immediately before the cursor position is secured.

FIG. 6 shows an example of the data structure of the stroke buffer 208 in FIG.
The stroke sequence information stored in the stroke buffer 208 is composed of NSTRK indicating the number of strokes (number of strokes) in the buffer 208 and NSTRK pieces of stroke data. The I-th stroke data (I = 1 to NSTRK) includes the number of coordinate points NPOINT [I] (indicating the number of coordinate points) and NPOINT [I] x, y coordinate data. Here, the x and y coordinate data of the Jth point (J = 1 to NPOINT [I]) of the Ith stroke data is expressed as x [I] [J], y [I] [J]. .

FIG. 7 shows an example of the data structure of the recognition candidate buffer 209.
The recognition candidate buffer 209 is used to store a character string that is a recognition candidate for the stroke string in the stroke buffer 208 and a matching result. In this embodiment, information (recognition candidate information) stored in the recognition candidate buffer 209 includes NCAND indicating the number of candidates (recognition candidate character strings) and NCAND candidates (candidate data).

  The data of candidate #I, that is, the I-th candidate (I = 1 to NCAND) includes a character segmentation result. This character segmentation result includes a rejection flag CNAD_REJFLAG [I] indicating whether or not the I candidate is a recognition rejection target, the number of characters CAND_NCODE [I] of the character string (recognition candidate character string) constituting the candidate, and CAND_NCODE [ I] entry numbers CAND_REFID [I] [J] in each character structure dictionary 211 of J characters (Jth character (J = 1 to CAND_NCODE [I])) and the total number of strokes CAND_NSTRK [ I], the number of input strokes CAND_STRKCTR [I] in the stroke buffer 208 corresponding to the last character of the candidate character string, and the stroke buffer 208 corresponding to each of CAND_NCODE [I] characters (Jth character) The input stroke sequence includes a start stroke number CAND_BS [I] [J] and an end stroke number CAND_ES [I] [J].

  The data of the I-th candidate (I = 1 to NCAND) is the likelihood of the result of collating the above-described character cutout result with the input stroke sequence corresponding to the CAND_NCODE [I] character (Jth character) character structure dictionary 211. Logarithmic value of logarithm (log likelihood) CAND_L1 [I] [J], inter-character structure dictionary 212 between character J and character J + 1, and the corresponding inter-stroke feature (the last character of character J) Logarithmic likelihood CAND_L2 [I] [J] as a result of collating the stroke and the first stroke of the next J + 1 character) and the sum of the log likelihoods, that is, the character string of the I-th candidate And log likelihood CAND_L [I] as a result of matching the entire input stroke sequence.

  In the example of FIG. 7, there are ten candidates (recognition candidate character strings) in the recognition candidate buffer 209, the first candidate (candidate # 1) to the tenth candidate (candidate # 10). For the first candidate, the rejection flag is set to 0, and the recognition candidate character string is a two-character string “Ai” consisting of the first entry “A” and the second entry “I” in the character structure dictionary 211. It is. The total number of strokes of this recognition candidate character string is five. Here, the result of collating by inputting up to the first stroke of the last character “I” of the recognition candidate character string “Ai” is stored. The input strokes corresponding to the first character “A” in the recognition candidate character string “Ai” are three strokes from the first stroke to the third stroke. The input stroke corresponding to the second character “I” of the recognition candidate character string “Ai” is only one of the fourth strokes. In other words, up to one screen on the left side of “I” is input.

  The logarithm value (log likelihood) of the likelihood as a result of collating the character structure dictionary 211 of the first character “A” of the recognition candidate character string “Ai” with the input stroke sequence from the first stroke to the third stroke is -0.70. Matching between the inter-character structure dictionary 212 between “A” and “I” and the inter-stroke structure between the third stroke (the last stroke of “A”) and the fourth stroke (the first stroke of “I”) The logarithmic value of the likelihood of the result of the (log likelihood between characters) is -0.36. The logarithmic value (log likelihood) of the likelihood as a result of collating the head stroke portion in the character structure dictionary 211 of “I” with the fourth stroke of the input stroke sequence is −0.22. The log likelihood of “a”, the log likelihood of letters “a” and “i”, and the log likelihood of the first stroke of “i” are added together to give a total log likelihood of −1.28.

FIG. 8 shows an example of the data structure of the character structure dictionary 211.
The character structure dictionary 211 is a dictionary in which data (character structure dictionary information) representing the structure of characters to be recognized is registered. The character structure dictionary 211 includes an NREF indicating the number of characters to be recognized and a structure dictionary for each of the NREF characters.

  The I-th (I = 1 to NREF) character structure dictionary (dictionary #I) has REF_CODE [I] that represents the character in code (for example, SHIFT-JIS code) and the total number of strokes that make up the character Average vector s [I] [J] [1-6] of shape features as features of REF_NSTRK [I] and REF_NSTRK [I] strokes (Jth stroke (J = 1 to REF_NSTRK [I])) and A covariance vector σ [I] [J] [1-6] and an average vector s2 [I] [J] [1-2] of the inter-stroke structural features between the J-1 and Jth strokes; Covariance vector σ2 [I] [J] [1-2]. Where s [I] [J] [1-6] consists of s [I] [J] [1], s [I] [J] [2], ... s [I] [J] [6] Assume that a 6-dimensional vector is represented. σ [I] [J] [1-6], s2 [I] [J] [1-2], and σ2 [I] [J] [1-2] are similar vector expressions. The stroke shape feature and the inter-stroke structure feature will be described later.

FIG. 9 shows an example of the data structure of the intercharacter structure dictionary 212.
The inter-character structure dictionary 212 expresses the structural relationship between the two characters for each combination of two characters written successively among the NREF characters (recognizable characters) registered in the character structure dictionary 211. This is a dictionary in which data (character dictionary information) is registered. FIG. 9 shows the data structure of the inter-character structure dictionary information representing the structural relationship between the characters for a set of two characters. This inter-character structure information is composed of an inter-character structure feature average vector z [1-2] and a covariance vector θ [1-2]. As the inter-character structure feature, a stroke structure feature between the last stroke of the previous character and the first stroke of the subsequent character is used.

  Next, the procedure of online character recognition processing in the online character recognition apparatus configured as described above will be described with reference to the flowcharts of FIGS.

First, the entire process will be described with reference to the flowchart of FIG.
Step 801 is a step of initializing each buffer in the online character recognition apparatus.

  In step 802, when one stroke is written in the character writing area 301, a coordinate data string of the stroke, that is, stroke data is taken into the stroke buffer 208 by the handwriting information acquisition means 202, and feature extraction of the stroke data is performed. Is made.

  In step 803, candidate character strings (recognition candidate character strings) that can be recognition candidates for the stroke strings captured in the stroke buffer 208 are generated by the recognition candidate character string generation unit 203 and stored in the recognition candidate buffer 209.

  In step 804, the recognition candidate character string in the recognition candidate buffer 209 generated in step 803 and the stroke string captured in step 802 are collated by the character string likelihood calculating unit 204, and the stroke string is used as the recognition candidate character string. The certainty that is written is calculated.

  In step 805, based on the result of collation with each recognition candidate character string in step 804, the recognition confirmed character string portion that can be determined that the recognition candidate character string has been written in the input stroke string, and the most likely character string. The input character string estimation means 205 estimates a (recognized unconfirmed character string) portion and a stroke string (recognized unconfirmed stroke sequence) portion in the middle of writing.

In step 806, the estimation result in step 805 is displayed in the recognition result display area 302 by the recognition result display unit 207.
In steps 807 and 808, the stroke buffer 208 and the recognition candidate buffer 209 are updated by the buffer update unit 206.

  Steps 802 to 808 are repeated until all input stroke sequences are estimated as recognition confirmed character strings (until the recognition candidate buffer 209 is empty) (step 809). As a result, the online character recognition device in FIG. 4 automatically recognizes the handwriting as a character string simply by continuously writing characters in the character writing area 301 without the user being aware of character separation. be able to.

Hereinafter, the processing content of each step in the flowchart of FIG. 10 will be described in detail.
First, in step 801, the stroke buffer 208 and the recognition candidate buffer 209 are initialized. Here, the number of strokes NSTRK in the stroke buffer 208 having the data structure shown in FIG. 6 and the number of recognition candidate character strings NCAND in the recognition candidate buffer 209 having the data structure shown in FIG. Is set, both buffers 208 and 209 are initialized.

  The next step 802 is executed each time a stroke is written by the user with the pen P in the character writing area 301 secured on the tablet 201 (arranged on the display screen of the display 5). Written stroke data (coordinate data string) is acquired by the tablet 201. In step 802, stroke data acquired by the tablet 201 is captured by the handwriting information acquisition unit 202 and stored in the stroke buffer 208. In step 802, a process for extracting features of the stroke data (the shape of the stroke indicated) stored in the stroke buffer 208 is performed.

Details of the processing of step 802 by the handwriting information acquisition means 202 will be described with reference to the flowchart of FIG.
First, in step 901, the number of strokes NSTRK in the stroke buffer 208 having the data structure shown in FIG.

In the next step 902 to step 905, the x-coordinate and y-coordinate data of the pen tip acquired by the tablet 201 before the pen P leaves the tablet 201 is sequentially taken into the stroke buffer 208. Here, the x-coordinate and y-coordinate of the J-th point of the I-th stroke are set to x [I] [J] and y [I] [J] in the stroke buffer 208 shown in FIG.
In step 906, the number of coordinate points captured (composing one stroke) until the pen P leaves the tablet 201 is set to NPOINT [I].

  In step 907, a stroke shape feature vector u [1-6] representing the stroke shape is extracted from the latest stroke taken into the stroke buffer 208, that is, the Nth stroke data. Here, u [1 to 6] represent a six-dimensional vector composed of u [1], u [2],... U [6]. As the shape feature, for example, six coefficients that express low-frequency components up to the second order of a P-type Fourier descriptor obtained by Fourier expansion of a complex value function having the total curvature function of stroke data in the exponent part of the exponential function. Shall be used. For the calculation procedure of the P-type Fourier descriptor, refer to the document “Online Handwritten Kanji Recognition Characterized by Curved Lines” (Journal 1990, Vol.J73-D-II No.4 pp.519 -525) is used.

If the input stroke is after the second stroke, in step 909, a two-dimensional inter-stroke structure representing the structural relationship between the NSTRK stroke and the previous input NSTRK-1 stroke. Feature vectors u2 [1-2] are extracted. As the structural feature, for example, a vector obtained by normalizing the norm of the vector connecting the end point of the NTRK-1 stroke to the start point of the NSTRK stroke to 1 is used.
The detailed processing procedure in step 802 has been described above.

  Next, in step 803, the recognition candidate character string generation unit 203 updates the recognition candidate character string in the recognition candidate buffer 209. Step 803 is a process of generating a character string that can be a recognition candidate for the input stroke string stored in the stroke buffer 208 at this time.

Details of the processing in step 803 will be described with reference to the flowchart of FIG.
First, in step 1002, it is determined whether or not a recognition candidate character string exists based on NCAND indicating the current number of recognition candidate character strings in the recognition candidate buffer 209.

  If it does not exist (NCAND = 0), the process jumps to step 1010, and each of the NREF characters registered in the character structure dictionary 211 having the data structure shown in FIG. 8 is replaced with CAND_NCODE [I] (I = 1 ˜NREF) is stored (generated) in the recognition candidate buffer 209 as an I-th recognition candidate character string (new recognition candidate character string) with “1”. A detailed processing procedure in step 1010 is shown in the flowchart of FIG.

On the other hand, when a recognition candidate character string already exists in the recognition candidate buffer 209 (NCAND> 0), the following processing is performed for each I-th recognition candidate character string.
First, at step 1005, whether or not the stroke input now is the first stroke of the next new one character following the current Ith recognition candidate character string is determined by the number of strokes NSTRK in the stroke buffer 208 as the first recognition. This is determined by whether the total number of strokes CAND_NSTRK [I] of the candidate character string has been exceeded.

  If it is determined that the stroke is not the first stroke of a new character, in step 1006, the counter CAND_STRKCTR [I] indicating the number of strokes constituting the final character of the I-th recognition candidate character string is incremented by one.

On the other hand, if it is determined that the stroke is the first stroke of a new character, in step 1008, a new recognition candidate character string in which one character is added to the current I-th recognition candidate character string is generated and a recognition candidate is created. Processing for registration in the buffer 209 is performed. One character to be added is all of the NREF characters included in the character structure dictionary 211, and as many new Kth recognition candidate character strings (K = 1 to NREF) are registered in the recognition candidate buffer 209 as the number of characters. . The detailed processing procedure in step 1008 is shown in the flowchart of FIG.
For the current I-th recognition candidate character string, in step 1009, a rejection flag is set as the old recognition candidate character string.

When NREF new recognition candidate character strings are generated and registered in the recognition candidate buffer 209 for all NCAND recognition candidate character strings in this way (step 1004), the old flag with a rejection flag set is set. In step 1011, a recognition candidate character string update process for removing the recognition candidate character string from the recognition candidate buffer 209 and aligning the new recognition candidate character string is performed. The detailed processing procedure in step 1011 is shown in the flowchart of FIG.
The detailed processing procedure in step 803 has been described above.

  Next, in step 804, each recognition candidate character string in the recognition candidate buffer 209 is checked against the input stroke string in the stroke buffer 208 by the character string likelihood calculating means 204, and the result of the matching (likelihood calculation). Result) is stored in the recognition candidate buffer 209.

Details of the processing of step 804 will be described with reference to the flowchart of FIG.
First, in step 1403, a character string structure dictionary used for collating input stroke strings is created for each I-th recognition candidate character string in the recognition candidate buffer 209. In the next step 1404, the input stroke string is collated with the created character string structure dictionary.

  The creation process of the character string structure dictionary of the I-th recognition candidate character string in step 1403 is performed as follows according to the flowchart of FIG. First, in step 1502, whether or not there are two or more strokes in the stroke buffer 208 is determined by whether or not the number of strokes NSTRK in the buffer 208 is greater than one. If there is only one stroke, the process jumps to step 1506. If there are two or more, it is determined in step 1503 whether or not the latest stroke is the first stroke of the last character of the I-th recognition candidate character string.

  In the case of the first stroke, the interval between the stroke and the previous stroke is considered to be between characters. In this case, in step 1504, the inter-character structure dictionary information between the corresponding characters is extracted from the inter-character structure dictionary 212, and the average vector z [1-2] and the covariance vector θ [1- 2] is set to v2 [1-2] and φ2 [1-2], respectively, and then the process proceeds to step 1506.

  On the other hand, if it is not the first stroke, it is considered that the stroke and the previous stroke are between strokes in the last character of the I-th recognition candidate character string. After the average vector and covariance vector of the corresponding inter-stroke structure features described in the dictionary information in the structure dictionary 211 are set to v2 [1-2] and φ2 [1-2], respectively, step 1506 follows. move on.

  In step 1506, the average vectors and covariance vectors of the stroke shape features described in the dictionary information in the character structure dictionary 211 for the last character corresponding to the latest input stroke are v [1-6] and φ [ 1 ~ 6].

  Next, the matching process between the character string structure dictionary of the I-th recognition candidate character string and the input stroke string in step 1404 is performed as follows according to the flowchart of FIG.

  First, in step 1602, whether or not there are two or more strokes in the stroke buffer 208 is determined in the same manner as in step 1502. If there is only one stroke, the process jumps to step 1607. If there are two or more, it is determined in step 1603 whether or not the latest stroke is the first stroke of the last character of the current I-th recognition candidate character string.

  In the case of the first stroke, the interval between the stroke and the previous stroke is considered to be between characters. In this case, in step 1604, the input stroke structure feature vector u2 [1-2] extracted in the previous step 909, the intercharacter structure feature average vector v2 [1-2] set in step 1504, and the covariance Likelihood calculation is performed with respect to the vector φ2 [1-2], and the logarithmic value log f (u2 | v2, φ2) is the corresponding inter-character structure portion of the I-th recognition candidate character string, that is, the latest stroke. Is set as the log likelihood of the inter-character structure portion between the character having the previous stroke as the last stroke and the next character (the last character of the current I-th recognition candidate character string), and then proceeds to step 1606 .

Here, the likelihood is the input vector u2 [1 to 2 when the multidimensional uncorrelated normal distribution with the mean vector v2 [1 to 2] and the covariance vector φ2 [1 to 2] is a probability density function. 2] is calculated by the following equation (1).

  On the other hand, if it is determined in step 1603 that the latest stroke is not the first stroke of the last character of the current I-th recognition candidate character string, the I-th recognition candidate character string is between the stroke and the previous stroke. Between strokes in the last character of. In this case, in step 1605, the inter-stroke structural feature vector u2 [1-2] extracted in step 909, the inter-stroke structural feature average vector v2 [1-2] set in step 1505, and the covariance vector φ2 Likelihood calculation is performed between [1 and 2], and the logarithmic value log f (u2 | v2, φ2) is the corresponding character structure portion of the I-th recognition candidate character string, that is, the current I-th recognition candidate. After accumulatively setting the log likelihood of the character structure part of the last character of the character string, the process proceeds to step 1606. For the likelihood calculation, a probability density function having the same format as in step 1604 is used.

  In step 1606, the logarithm value log f (u2 | v2, φ2) of the likelihood calculated in step 1604 or 1605 preceding step 1606 is input as the I-th recognition candidate character string obtained up to the present time. It is accumulated and set in the log likelihood CAND_L [I] as a result of matching the entire stroke sequence.

  In step 1607, the input stroke shape feature vector u [1-6] extracted in step 907 and the corresponding stroke shape feature average vector v [1˜ in the character structure dictionary of the last character of the I-th recognition candidate character string. 6] and the covariance vector φ [1-6], the likelihood is calculated, and the logarithmic value log f (u | v, φ) is obtained.

Here, the likelihood is the input vector u [1 to when the multidimensional uncorrelated normal distribution with the mean vector v [1 to 6] and the covariance vector φ [1 to 6] is a probability density function. 6] is calculated by the following equation (2).

  In step 1608, the logarithmic value log f (u | v, φ) obtained in step 1607, that is, the logarithm value log f (u | v, φ) of likelihood obtained by collating the stroke shape feature is It is accumulated and set to the log likelihood of the character structure portion corresponding to the I recognition candidate character string, that is, the character structure portion of the last character of the I recognition candidate character string.

  In step 1609, the logarithmic value log f (u | v, φ) obtained in step 1607 is a logarithmic likelihood as a result of collating the I-th recognition candidate character string and the entire input stroke string obtained so far. Accumulated and set in degree CAND_L [I].

  When all the recognition candidate character strings in the recognition candidate buffer 209 are collated with the input stroke string in the stroke buffer 208 (step 1405), the recognition candidate character strings are narrowed down in step 1406.

  The narrowing process in step 1406 is executed according to the flowchart of FIG. Here, the sum of log likelihood obtained for each I-th recognition candidate character string (I = 1 to NCAND) in the recognition candidate buffer 209, that is, the result of collating the I-th recognition candidate character string with the entire input stroke string If the log likelihood CAND_L [I] is less than the predetermined threshold value α, it is determined that the possibility that the recognition candidate character string has been input is low (step 1703). In this case, the recognition candidate character string is deleted from the recognition candidate buffer 209.

  On the other hand, a recognition candidate character string having a log likelihood CAND_L [I] greater than or equal to the threshold value α is determined to have a high possibility that the recognition candidate character string has been input, and is stored in the recognition candidate buffer 209 as the Jth recognition candidate character string. It is left (step 1704).

  The process of steps 1403 and 1404 in the flowchart of FIG. 16 described above (detailed procedure of the likelihood calculation process for the recognition candidate character string in step 804) is overwritten with the character “ai” in the character writing area 301. A specific example of likelihood calculation between the written handwriting and the recognition candidate character string “Ai” will be described in the order of writing with reference to FIG.

  First, when the first stroke is written, the shape feature u [1 to 6] extracted from the stroke (first input stroke) and the shape feature s [1] of the first stroke in the character structure dictionary of “A” [1] [1-6] and σ [1] [1] [1-6] are collated.

  When the second input stroke is written, the inter-stroke structure feature u2 [1-2] between the previous input stroke (first input stroke) and the first stroke of the character structure dictionary of “A” Inter-stroke structural features between the second stroke and s2 [1] [1] [1-2], σ2 [1] [1] [1-2], and the second input stroke Shape feature u [1-6] and shape feature s [1] [2] [1-6], σ [1] [2] [1-6] of the second stroke of the character structure dictionary of “A” Are also collated.

  Likelihood calculation is performed for the third input stroke as well, but the fourth input stroke and the fourth input stroke are determined for the fourth input stroke because the stroke is determined to be the first stroke of “yes”. The inter-input stroke structural features u2 [1-2] between the two are collated between the inter-character structural features z [1-2], θ [1-2] of “A” and “I”. The cumulative value of the logarithmic values of the likelihoods calculated by these collations is the log likelihood between the input stroke string and the recognition candidate character string.

  In step 805 following step 804, the input character string estimation unit 205 uses the recognition candidate character string stored in the recognition candidate buffer 209 and the matching result for each recognition candidate character string to recognize the recognized character string. The undetermined character string and the recognized undetermined stroke string information are determined.

  FIG. 20 is a flowchart for explaining the processing procedure in step 805. As shown in this flowchart, the input character string estimation process in step 805 includes a recognition confirmed character string estimation step 1801 and a recognition unconfirmed character string and recognition unconfirmed stroke information estimation step 1802.

  FIG. 21 is a flowchart for explaining a detailed processing procedure in step 1801. Here, all the first recognition candidate character strings (I = 1 to NCAND) in the recognition candidate buffer 209 are common from the top with reference to the first recognition candidate character string (steps 1902, 1906, 1908). A string (J = 1 to NSTRING1) of the character code STRING1 [J] of the character string portion made up of NSTRING1 characters is extracted as a recognition confirmed character string STRING1 (step 1907) and set in the recognition result buffer 210. In step 1907, the total number of strokes NSTRK1 of NSTRING1 characters is also obtained and set in the recognition result buffer 210 in association with the recognition confirmed character string STRING1.

  FIG. 22 is a flowchart for explaining the processing procedure in step 1802. Here, the first MI recognition candidate character string (maximum likelihood recognition candidate character) having the largest log likelihood CAND_L [I] among all the first I recognition candidate character strings (I = 1 to NCAND) in the recognition candidate buffer 209. Column) (MI is any one of 1 to NCAND) (steps 2001 to 2005).

  Next, for the maximum likelihood recognition candidate character string obtained, whether or not the final character of the candidate character string is in the middle of writing is determined by the end stroke number CAND_ES [MI] [CAND_NCODE [MI of the input stroke sequence corresponding to the final character. ]] Is smaller than the value of CAND_NSTRK [MI] representing the total number of strokes of the candidate character string (step 2006).

  If the last character of the maximum likelihood recognition candidate character string is in the middle of writing, the recognition unconfirmed stroke string information USTRK_FLAG is set to “1”, and the recognition confirmed character string and the final character are excluded from the maximum likelihood recognition candidate character string. The character string portion consisting of NSTRING2 characters is taken out, and the character code string STRING2 [J] (J = 1 to NSTRING2) of the character string portion is set in the recognition result buffer 210 as a recognized unconfirmed character string STRING2 ( Step 2007, 2009-2011). At this time, the log likelihood of the maximum likelihood recognition candidate character string including the recognized unconfirmed character string may be stored in the recognition result buffer 210 together with the recognized unconfirmed character string STRING2. The log likelihood of the maximum likelihood recognition candidate character string stored in the recognition result buffer 210 can be used when the recognition result display unit 207 displays the recognition unconfirmed character string STRING2 later.

  On the other hand, if all strokes of the last character of the maximum likelihood recognition candidate character string are written, the recognition unconfirmed stroke information USTRK_FLAG is set to “0”, and the recognition confirmed character string is further excluded from the maximum likelihood recognition candidate character string. A character string portion consisting of NSTRING2 characters is extracted, and a character code string STRING2 [J] string (J = 1 to NSTRING2) of the character string portion is set in the recognition result buffer 210 as a recognized unconfirmed character string STRING2 (step 2008-2011).

  As is clear from the above description, the recognition-determined character string STRING1 is a character string of a portion determined in the sense that the estimation result does not change due to future writing in the input stroke string. Similarly, the unrecognized character string STRING2 is a character string of a portion that is estimated to be most likely in the meaning of the maximum likelihood at the present time, although the estimation result may change depending on future writing. The unrecognized stroke sequence information USTRK_FLAG represents the presence or absence of a stroke sequence for which characters have not yet been written.

  In step 806 following step 805, the recognition result display means 207 converts the recognized confirmed character string, recognized unconfirmed character string, and recognized unconfirmed stroke string information in the recognition result buffer 210 into a display pattern, and It is displayed in the recognition result display area 302 in the display screen.

  An example of the display of the recognized confirmed character string, the recognized unconfirmed character string, and the recognized unconfirmed stroke string information that is estimated each time a stroke is written is overwritten with the character “ai” in the character writing area 301. FIG. 26 shows the third column in association with each stroke (input stroke) and the stroke number.

  In the figure, a black square symbol is a cursor indicating a character insertion position and corresponds to the cursor C in FIG. A character string portion without an underline represents a recognized confirmed character string, and an underlined character string portion represents a recognized unconfirmed character string. The symbol “⇒” is displayed at the character position next to the recognized unconfirmed character string when the recognized unconfirmed stroke string information USTRK_FLAG is “1”, and indicates that a recognized unconfirmed stroke string exists.

  Thus, in this embodiment, a display attribute that allows the user to easily identify (view) the recognized confirmed character string, the recognized unconfirmed character string, and the recognized unconfirmed stroke string information is added each time a stroke is input. Since it is displayed on the screen, the user can check the handwriting recognition processing result sequentially, and comfortable character input is possible.

  When step 806 is completed, the buffer update unit 206 updates the recognition candidate buffer 209 in step 807, and the stroke buffer 208 is updated in step 808.

  The detailed processing procedure in step 807 is shown in the flowchart of FIG. Here, when the recognition confirmed character string NSTRING1 exists, for each I-th recognition candidate character string (I = 1 to NCAND) in the recognition candidate buffer 209, a portion corresponding to the recognition confirmed character string NSTRING1 from each candidate character string Information is removed.

  Next, the detailed processing procedure in step 808 is shown in the flowchart of FIG. Here, of the NSTRK number I stroke data (I = 1 to NSTRK) in the stroke buffer 208, I = NSTRK1 + 1 to NSTRK I stroke data, that is, NSTRK1 + 1 stroke data to NSTRK stroke data. Is the new NSTRK (new NSTRK = old NSTRK-NSTRK1) J stroke data, so that the input stroke string data (consisting of the first stroke data to the NSTRK1 stroke data) corresponding to the recognized character string NSTRING1 Is updated from the stroke buffer 208.

  This buffer update process is roughly divided into a loop of steps 2107 and 2108 and steps 2109 to 2111. In the loop of steps 2107 and 2108, the input stroke string data corresponding to the character string excluding the final character in the recognition confirmed character string NSTRING1 is deleted from the stroke buffer 208. In steps 2109 to 2111, the final stroke in the recognition confirmed character string NSTRING1 is deleted. The input stroke string data corresponding to the character is deleted from the stroke buffer 208.

  In step 809 following step 808, whether or not the recognition candidate buffer 209 is empty is determined by, for example, the buffer update unit 206. If there is a recognition candidate character string, the process returns to step 802 to return to the handwriting information acquisition unit 202. Control is passed to and the next stroke is captured.

  On the other hand, when the recognition candidate buffer 209 is empty, the recognition process is terminated, assuming that a character string whose recognition result has been determined for all the input stroke strings is displayed and output.

  Therefore, the on-line character recognition apparatus according to the present embodiment can recognize and input a character string continuously written by the user without being aware of character separation by the processing procedure described above.

(Recognition result display method)
As described above, the recognition result display unit 207 in FIG. 4 converts the recognized confirmed character string, the recognized unconfirmed character string, and the recognized unconfirmed stroke string information in the recognition result buffer 210 into a display pattern, and displays on the display 213. It is displayed in the recognition result display area 302 in the screen.

  Next, an example of a recognition result display procedure in the recognition result display means 207 in step 806 of FIG. 10 will be described with reference to the flowchart shown in FIG. The flowchart shown in FIG. 27 shows the procedure for displaying the recognized confirmed character string and the recognized unconfirmed character string among the recognition results. In particular, when displaying the recognized unconfirmed character string, the likelihood is displayed. This is characterized in that the display is updated by comparing the degree with a predetermined threshold. The likelihood of the unrecognized character string is the log likelihood of the maximum likelihood recognition candidate character string including the unrecognized character string when the unrecognized character string is estimated by the input character string estimating unit 205. is there. This log likelihood is assumed to be stored in the recognition result buffer 210 together with the unrecognized character string.

  First, the recognition result display unit 207 extracts the recognition confirmed character string from the recognition result buffer 210 and displays it. That is, the recognition confirmed character string is converted into a display pattern and displayed in the recognition result display area 302 in the display screen of the display 213 (step S11).

  Next, the unrecognized character string is displayed. Here, first, the unrecognized character string and the log likelihood stored together with the unrecognized character string are extracted from the recognition result buffer 210. . When this log likelihood value is larger than a predetermined threshold (or larger than the threshold), the current unrecognized character string is displayed. That is, the unrecognized character string is converted into a display pattern and displayed in the recognition result display area 302 in the display screen of the display 213 (steps S12 and S13).

  On the other hand, when the value of the log likelihood stored together with the recognized unconfirmed character string is equal to or smaller than a predetermined threshold (or smaller than the threshold), the current unconfirmed character string is not displayed. Then, the currently unrecognized character string that is currently displayed is displayed as it is (step S12, step S14).

  Next, another example of the recognition result display procedure by the recognition result display means 207 in step 806 of FIG. 10 will be described with reference to the flowchart shown in FIG. The flowchart shown in FIG. 28 also shows the procedure for displaying the recognized confirmed character string and the recognized unconfirmed character string in the recognition result. In particular, when displaying the recognized unconfirmed character string, the likelihood is displayed. It is characterized in that the display is updated by comparing the degree and the likelihood of the unrecognized character string currently displayed. The likelihood of the unrecognized character string is the log likelihood of the maximum likelihood recognition candidate character string including the unrecognized character string when the input unrecognized character string is estimated by the input character string estimating unit 205. Degree. This log likelihood is assumed to be stored in the recognition result buffer 210 together with the unrecognized character string.

  First, the recognition result display unit 207 extracts the recognition confirmed character string from the recognition result buffer 210 and displays it. That is, the recognition confirmed character string is converted into a display pattern and displayed in the recognition result display area 302 in the display screen of the display 213 (step S21).

Next, the unrecognized character string is displayed. Here, first, the unrecognized character string and the log likelihood stored together with the unrecognized character string are extracted from the recognition result buffer 210. . When the log likelihood value is greater than (or greater than) the likelihood of the currently unrecognized character string that is currently displayed (held in the recognition result display means 207), The current unconfirmed character string is displayed. That is, the unrecognized character string is converted into a display pattern and displayed in the recognition result display area 302 in the display screen of the display 213 (steps S22 and S23). Then, the likelihood of the recognized unconfirmed character string displayed this time is held (step S24).
On the other hand, the value of the log likelihood stored together with the recognized unconfirmed character string is equal to or less than the likelihood of the currently recognized unconfirmed character string (held in the recognition result display means 207). If it is (or smaller), the currently unrecognized character string currently displayed is displayed as it is without displaying the current unrecognized character string (steps S22 and S25).

  Examples of the display of the recognized confirmed character string and the unrecognized character string that are estimated each time a stroke is written are shown in each stroke of the handwriting in which the character “word” is overwritten in the character writing area 301 ( The first drawing of “Te”, the first drawing of “Ga”, the second drawing,...), And the recognized confirmed character string and the recognized unconfirmed character string estimated each time the stroke is written, 29 in the fourth column.

  In the fourth column of the figure, the underlined character string portion is a recognized confirmed character string, and the underlined character string portion represents a recognized unconfirmed character string.

  As described above, in the present embodiment, a display attribute that allows the user to easily identify (recognize) a recognized confirmed character string and a recognized unconfirmed character string is added and displayed on the screen every time a stroke is input. The user can check the handwriting recognition processing result sequentially, and can input handwritten characters comfortably and efficiently.

(Handwriting display method)
When a character is written on the character writing area 301, if the handwriting is not displayed at all, the position of the previous stroke cannot be confirmed. Not only can not be entered, it also causes misrecognition. In particular, when characters are overwritten on the character writing area 301 for one character, when all the input strokes are displayed, strokes other than the character currently being input are displayed, which is difficult to understand.

  Then, next, the display method of the stroke written in the character writing area 301 of the tablet 201 is demonstrated.

  FIG. 30 shows the configuration of an on-line character recognition device having handwriting display means for displaying strokes written in the character writing area 301 of the tablet 201. In FIG. 30, the same parts as those in FIG. 4 are denoted by the same reference numerals, and only different parts will be described. That is, the handwriting display means 220 and the pen operation determination means 221 are newly added. The tablet 201 is transparent, and this tablet is provided on the handwriting display area of the display 213 for displaying the handwriting written on the character writing area. When a stroke written by the user using P is displayed, the user can see the displayed stroke via the tablet 201.

  The handwriting display means 220 displays the handwriting in the handwriting display area on the display 213 using the stroke data acquired by the handwriting information acquisition means 202.

  Since the coordinate data representing the position of the pen tip when the pen P is touching the tablet is acquired by the handwriting information acquisition unit 202, the pen operation determination unit 221 determines the type of pen operation based on the coordinate data. It is determined, and a predetermined instruction is given to the handwriting display means 220 or the like according to the determination result.

  The handwriting display means 220 displays the latest N input strokes including the currently input stroke. The number N of strokes to be displayed is set in advance.

  Next, the handwriting display processing operation of the handwriting display means 220 will be described with reference to the flowchart shown in FIG. The process shown in FIG. 31 is executed in step S1 of FIG.

  For display control of the latest N input strokes, the handwriting display means 220 includes a stroke counter M that counts the number of strokes input on the tablet 201 until it reaches N, and a stroke buffer that stores input stroke data. B and a ring buffer pointer P indicating the storage position on the stroke buffer. The stroke buffer B has an area for storing the first to Nth input stroke data, and the ring buffer pointer P points in order from the first storage area of the stroke buffer B to the second, third,. When it reaches the second position, the storage area of the stroke buffer B is pointed endlessly by returning to the first position again. Here, for simplicity of explanation, the possible values of the ring buffer pointer P are values indicating the first to Nth storage areas of the stroke data of the stroke buffer B, that is, 1 to N, and the stroke buffer The Pth storage area of B is represented as B (P).

  First, with the start of handwritten character input by the pen P, the handwriting display means 220 is initialized. In this initialization, for example, if there is a handwriting currently displayed in the handwriting display area, it is deleted and the stroke counter M and the ring buffer pointer P are set to “0” (step S101). .

  When one stroke is input (step S102), the value of the ring buffer pointer P at that time is compared with N. If P is not equal to N (step S103), the process proceeds to step S104, and the ring buffer pointer P is set to 1. Increment by one. On the other hand, if P is equal to N in step S103, the process proceeds to step S105, and the value of the ring buffer pointer P is returned to “1” so that the ring buffer pointer P points to the first storage area of the stroke buffer B.

  Next, the process proceeds to step S106. In step S106, the stroke data input in step S102 is stored in the Pth storage area of the stroke buffer B (step S106). When the number M of strokes input so far is smaller than N (or less than N) (step S107), the process proceeds to step S108, and after incrementing the stroke counter M by one, in step S109, The latest P strokes including the stroke data stored in the Pth storage area this time are extracted from the stroke buffer B, and the handwriting is displayed. In this case, the stroke data stored in the stroke buffer B are B (P), B (P-1),... B (1) when arranged in order from the newest.

  On the other hand, when the number M of strokes input so far is N or more (or smaller than N) in step S107, the process proceeds to step S110 without updating the stroke counter M, and the stroke buffer B is updated. The latest N strokes including the stroke data stored in the Pth storage area this time are taken out and the handwriting is displayed. In this case, when the stroke data stored in the stroke buffer B are arranged in order from the newest, B (p), B (P-1),... B (1), B (N), B (1), B (2),... B (P + 1).

  Here, a display method of stroke data will be described. As described above, each stroke data is a two-dimensional coordinate data string representing the position of the pen tip while the pen P is touching the tablet 201. Assume that one stroke is composed of J coordinate data. Each coordinate data is represented as (x [j], y [j]). Here, j = 1 to J. For example, FIG. 32 shows a handwriting display example when the user has written a character “no” (character composed of one stroke). In this case, since there are twelve coordinate points constituting the input stroke, the twelve coordinate data strings (x [j], y [j]), j = 1 to J are connected in order, and this is indicated by a broken line. An input stroke can be expressed.

  When the input stroke handwriting is displayed in the procedure shown in FIG. 31, for example, when N is set to “2”, the handwriting display area of the display 213 has a handwriting as shown in FIG. Is displayed. FIG. 33A shows a handwriting display example when the second stroke of the character “I” has been input in the character writing area 301 of the tablet 201. FIG. 33B shows an example of a handwriting display when the third stroke of the character “A” has been entered in the character writing area 301 of the tablet 201. As is apparent from the figure, the first stroke is not displayed. Similarly, FIG. 33 (c) shows a handwriting display example when the fourth stroke of the character “TA” has been entered in the character writing area 301 of the tablet 201. As is clear from the figure, the first and second strokes are not displayed.

  In the handwriting display area of the display 213, it is desirable that the stroke written by the user using the pen P on the tablet 201 is displayed in real time simultaneously with the writing. The processing operation of the handwriting display means 221 for that purpose will be described with reference to the flowchart shown in FIG. Note that the process of displaying the stroke being input shown in FIG. 34 is executed in step S102 of FIG.

  As described above, the handwriting information acquisition unit 202 acquires coordinate data representing the position of the pen tip while the pen P is touching the tablet. That is, the coordinate data acquired while the user is writing for one stroke is input to the handwriting display means 221. At that time, after the pen tip leaves the tablet last time (that is, after the end of one stroke writing). When the pen tip touches the tablet for the first time, the handwriting display starts when the coordinate data is first input. First, a variable K for counting the number of coordinate points is set to “0” (step S201). Then, the variable K is incremented by 1 (step S202), and the coordinate data (x [K], y [K]) at that time is acquired (step S203). When the coordinate data acquired from the start of the handwriting display including the coordinate data acquired this time is one, that is, when K = 1 (step S204), the process proceeds to step S206 to display the coordinate point. . On the other hand, when the coordinate data acquired after starting the handwriting display is the second or more, that is, when K> 1, the process proceeds to step S205, and the current coordinate data is (x [K], y [K ]), A line segment connecting the current coordinate point and the previously displayed coordinate point (x [K-1], y [K-1]) is displayed. Steps S202 to S206 are repeated until it is detected that the pen tip has left the tablet (end of one stroke writing) (step S207).

  The end of writing of one stroke may be determined by the pen operation determining means 220 detecting that the pen tip has left the tablet and notifying the handwriting display means 221 of this, or the handwriting display means. 221 may determine the end of one stroke of writing when the input of coordinate data from the handwriting information acquisition unit 202 is temporarily interrupted.

  Further, when displaying N strokes, the strokes may be displayed so that each stroke can be distinguished. For example, the strokes may be displayed with different colors and thicknesses, or as shown in FIG. 35, the strokes may be displayed with different line types such as solid lines, dotted lines, and wavy lines. Good.

  Note that the value of N may be set to a desired value by the user, or whether or not to display N input strokes may be set by the user. By doing in this way, the input environment of the handwritten character optimal for the user can be provided for every user.

  As described above, the pen operation determination unit 220 determines which position on the tablet the pen P is touching to determine the type of pen operation. “Types of pen operations include, for example, operations for writing (characters) input and other operations (for example, cursor movement).

  When the pen P touches a predetermined character writing area on the tablet, it may be determined that writing (character) input is started. If it is determined that character input has been started, the processing operations shown in FIG. 3 and FIG. 31 are performed.

  For example, when the pen P touches a predetermined area other than a predetermined character writing area on the tablet, or for a predetermined time after the pen P touches the tablet, the predetermined predetermined with reference to the contact point When the pen tip remains within the range, it may be determined that the operation is other than writing input.

  In addition, the time after the pen tip leaves the tablet is measured, and when the predetermined time is exceeded, the handwriting display means 221 is initialized and displayed at the time in the handwriting display area of the display 213. You may make it erase the handwriting which was done.

  In this way, even when the character cutout position is not explicit as in the case of overwritten character input, the input handwriting can be appropriately displayed, and an environment in which handwritten characters can be easily input is provided. be able to.

(Additional function of online character recognition device)
Hereinafter, the additional function will be described with reference to a configuration example of the online character recognition apparatus shown in FIG.

  In the embodiment described so far, as shown in FIG. 5, the character writing area 301 and the recognition result display area 302 are separated. However, also in the recognition result display area 302, it is possible to realize pen input for instructing movement of the cursor C indicating the character insertion position. That is, in this embodiment, when the display 213 includes a handwriting display area and a recognition result display area, a transparent tablet 201 is provided so as to cover them, and the user can display the display via the tablet 201. The handwriting displayed on 213 can be seen, and by specifying an arbitrary position in the recognition result display area 302 with the pen P, an editing operation can be performed on the character string obtained as the recognition result. An instruction for this editing operation may be called a “gesture”.

  For example, in the configuration shown in FIG. 30, when the pen operation determination unit 220 detects that an arbitrary position in the recognition result display area 302 is designated (touched) by the pen P, the cursor is moved. An instruction is given and it is determined that writing has been completed for the character string currently being written. Then, if there is an unrecognized character string that is recognized at that time, the input character string estimating unit 205 determines that the recognized character string is displayed in the recognition result display area 302 by the recognition result display unit 207. An instruction (confirmation instruction) to perform is performed. At this time, if there is a recognized undetermined stroke sequence, the input character string estimating unit 205 deletes the recognized undetermined stroke sequence from the stroke buffer 208. This deletion process can also be performed by the buffer update unit 206.

  By doing so, it becomes possible to smoothly repeat the instruction of the character insertion position and the character string writing operation, and comfortable character input becomes possible. The pen operation determination unit 220 can be realized by the CPU 1 in the information device of FIG. 1 executing an online character recognition program, like the units 202 to 207.

  In the present embodiment, after the last stroke is input, the pen operation determination unit 220 detects that there is no next stroke input for a predetermined time or more. Even when it is detected that there is no next stroke input for a certain time or longer, the pen operation determination unit 220 regards the character string currently being written as having been written, and performs a determination process similar to the above. .

  The third and fifth columns in FIG. 26 are associated with the strokes (input strokes) and stroke numbers of the handwriting in which the characters “Ashita” are overwritten. Shown in

  Further, the character writing area 301 and the recognition result display area 302 shown in FIG. 5 can be made common as shown in FIG. In such a configuration, when the coordinates of the pen tip do not move from within a certain coordinate range for a certain period of time or longer, the pen operation determination unit 220 performs a gesture for instructing movement of the cursor C whose pen input indicates the recognized character insertion position. It is determined that When the pen operation determining unit 220 determines that the pen input is a gesture, the input character string estimating unit 205 performs a determination process similar to the above in which the unrecognized character string is output as a recognized fixed character string. do it.

  Now, the online character recognition device of this embodiment has a one-character backward deletion button 303 shown in FIG. 5 as an editing operation means (one-character backward deletion instruction means) for deleting an erroneously input character. Yes. When the user touches this one-character backward deletion button 303 with the pen P, the pen operation determining means 220 is instructed to delete one character backward because the coordinate data is within the area of the one-character backward deletion button 303. And deletion of the character immediately before the character position pointed to by the cursor C can be instructed.

  In the present embodiment, as shown in the third column of FIG. 26, a one-character backward deletion instruction has been issued in a state where the symbol “⇒” indicating the presence of an unrecognized stroke sequence is displayed in the recognition result display area 302. In this case, a recognition result editing means (not shown) for deleting “⇒” is provided.

  This recognition result editing means, when there is no recognized unconfirmed stroke string but a recognition unconfirmed character string exists and one character backward deletion instruction is given, the last of the recognized unconfirmed character string is moved backward by one character. At the same time, the remaining unrecognized character string is confirmed and displayed as a recognized character string. The recognition result editing means deletes the confirmed character string backward by one character when an instruction for backward deletion of one character is given in a state where neither the recognized unconfirmed stroke string nor the recognized unconfirmed character string exists.

  By executing such processing, it is possible to smoothly and continuously delete characters and write characters, and a comfortable character editing environment is realized. The example of changing the display screen before and after the one character backward deletion instruction is associated with each stroke (input stroke) and stroke number of the handwriting overwritten with the character “Ashita”, and the third and fourth columns in FIG. Shown in

  The instruction for deleting one character backward is not limited to the button (one character backward deletion button 303). For example, the pen operation determination unit 220 can also determine that a stroke having a specific shape written in the character writing area 301 is a gesture for instructing the backward deletion of one character. In addition to this, for example, a straight stroke input from the right to the left, which is not input by normal character writing, can be determined as a gesture for deleting one character backward. The pen operation determination unit 220 can be easily realized by collating stroke shape features used in character recognition.

  In addition, this invention is not limited to the said embodiment, In the implementation stage, it can change variously in the range which does not deviate from the summary. Further, the above embodiments include inventions at various stages, and various inventions can be extracted by appropriately combining a plurality of disclosed constituent elements. For example, even if some constituent elements are deleted from all the constituent elements shown in the embodiment, at least one of the problems described in the column of the problem to be solved by the invention can be solved, and is described in the column of the effect of the invention. When at least one of the effects is obtained, a configuration in which this configuration requirement is deleted can be extracted as an invention.

The block diagram which shows the hardware constitutions of the information equipment which implement | achieves the online character recognition apparatus concerning one Embodiment of this invention. The block diagram which shows the function structure of the online character recognition apparatus concerning one Embodiment of this invention. The flowchart for demonstrating the procedure of the whole online character recognition process in the online character recognition apparatus of FIG. The block diagram which shows the more detailed function structure of the online character recognition apparatus implement | achieved by the information equipment of FIG. The figure which shows the external appearance of the information equipment of FIG. The figure which shows the example of a data structure of the stroke buffer 208 in FIG. The figure which shows the example of a data structure of the recognition candidate buffer 209 in FIG. The figure which shows the example of a data structure of the character structure dictionary 211 in FIG. The figure which shows the example of a data structure of the structure structure dictionary 212 in FIG. The flowchart for demonstrating the procedure of the whole online character recognition process in the online character recognition apparatus of FIG. 4 implement | achieved by the information equipment of FIG. 11 is a flowchart for explaining a detailed processing procedure in step 802 in FIG. 10. 11 is a flowchart for explaining a detailed processing procedure in step 803 in FIG. 10. The flowchart for demonstrating the detailed process sequence in step 1010 in FIG. The flowchart for demonstrating the detailed process sequence in step 1008 in FIG. 14 is a flowchart for explaining a detailed processing procedure in step 1011 in FIG. 13. 11 is a flowchart for explaining a detailed processing procedure in step 804 in FIG. 10. 17 is a flowchart for explaining a detailed processing procedure in step 1403 in FIG. 16. 17 is a flowchart for explaining a detailed processing procedure in step 1404 in FIG. 16. 17 is a flowchart for explaining a detailed processing procedure in step 1406 in FIG. 16. 11 is a flowchart for explaining a detailed processing procedure in step 805 in FIG. 10. The flowchart for demonstrating the detailed process sequence in step 1801 in FIG. The flowchart for demonstrating the detailed process sequence in step 1802 in FIG. 11 is a flowchart for explaining a detailed processing procedure in step 807 in FIG. 10. 11 is a flowchart for explaining a detailed processing procedure in step 808 in FIG. 10. The figure for demonstrating the collation relationship between an input feature and a dictionary feature. The figure which shows the example of a recognition result display. The flowchart for demonstrating the display processing operation | movement of a recognition fixed character string and a recognition unconfirmed character string. The flowchart for demonstrating other display processing operation | movement of a recognition definite character string and a recognition undecided character string. The figure which showed the example of a display of the recognition confirmation character string estimated every time a stroke is written, and the recognition unconfirmed character string. FIG. 5 is a block diagram showing a more detailed functional configuration of the on-line character recognition device realized by the information device of FIG. 1, in which handwriting display means and pen operation determination means are added to the configuration shown in FIG. 4. The flowchart for demonstrating the handwriting display process operation | movement of a handwriting display means. The figure which showed the example of a display of handwriting. The figure which showed the example of a handwriting display when N is set to "2". The flowchart for demonstrating the processing operation for displaying the stroke in the middle of writing on a tablet simultaneously with writing in real time. The figure which showed the example of a display of the latest N strokes. The figure which shows the modification of the external appearance of the information equipment of FIG.

Explanation of symbols

1 ... CPU
2 ... ROM
3 ... RAM
4,201 ... Tablet (coordinate input device)
5,213 ... Display (display device)
6 ... Data reading device 7 ... Storage medium 202 ... Handwriting information acquisition means 203 ... Recognition candidate character string generation means 204 ... Character string likelihood calculation means (collation means)
205: Input character string estimation means 206 ... Buffer update means 207 ... Recognition result display means 208 ... Stroke buffer 209 ... Recognition candidate buffer 210 ... Recognition result buffer 211 ... Character structure dictionary 212 ... Inter-character structure dictionary 220 ... Pen operation display means 221 ... handwriting display means 301 ... character writing area 302 ... recognition result display area 303 ... one character backward deletion button P ... pen C ... cursor

Claims (4)

Character recognition is performed based on the stroke sequence expressed by the coordinate sequence of the pen tip detected by the coordinate input device before the pen touches the coordinate input device on the coordinate input device, and then displayed. In the online character recognition device that displays the recognition result on the device,
For each of a plurality of recognizable characters, a character structure dictionary that registers the shape information of the strokes constituting the character and dictionary information describing the structural relationship between the strokes;
An inter-character structure dictionary in which dictionary information describing the structural relationship between the two characters is registered for each combination of two characters written successively among the plurality of recognizable characters;
Handwriting information acquisition means for sequentially capturing strokes detected by the coordinate input device;
A recognition candidate character string generation unit that generates a recognition candidate character string that can be a recognition candidate for an already-acquired stroke sequence that should be recognized each time a stroke is captured by the handwriting information acquisition unit;
For each recognition candidate character string, dictionary information in the character structure dictionary for each candidate character constituting the recognition candidate character string and dictionary information in the intercharacter structure dictionary between two consecutive candidate characters; and An input character string estimation means for outputting a character string in a recognition candidate character string that is most similar to the stroke string as a recognized character string by collating with an already captured stroke string;
Recognition result display means for displaying the recognition character string output by the input character string estimation means;
An on-line character recognition device comprising: A character writing area for writing a character with the pen of the coordinate input device is arranged to overlap the handwriting display area of the display device,
5. Each time the stroke is written in the character writing area, the latest predetermined N strokes (where N is an integer) including the latest stroke are displayed in the handwriting display area. The on-line character recognition device according to 1. After detecting the pen from the coordinate input device, the detecting means for detecting a pen operation other than writing the stroke,
3. The on-line character recognition apparatus according to claim 2, wherein the display of the stroke displayed in the handwriting display area is erased according to the detection result of the detecting means.   When displaying the predetermined number of strokes in the handwriting display area, one of the color and thickness of each stroke and the type of line representing the stroke based on the input order of the strokes The on-line character recognition apparatus according to claim 2, wherein the characters are displayed differently.

Images