JPH07121660A - Handwritten character recognizing device - Google Patents

Abstract

PURPOSE:To recognize the complicated characters inputted by handwriting at a high speed and with high accuracy by recognizing the component parts of a complicated character after dividing them into plural character input frames. CONSTITUTION:Plural pairs of characters and character component parts are stored in a character pattern group storing part 38 and a character component table 39 opposite to each other. A single KANJI (Chinese character) is divided into simple parts (character component parts) which include at least a crown part, a left-hand radical or a right-hand radical of a Chinese character. These divided component parts are supplied into the input frames through an input part 31 and by means of a pointing device. A recognizing part 32 and a retrieving part 33 compare the pairs of character component parts supplied to the input frames with the pairs of character component parts stored in the part 38 and the table 39. Thus a character including the pair of character component parts supplied to the input frames is selected and then shown on a unified display input device via an output part 34.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a handwritten character recognition system for inputting and recognizing handwritten characters.

[0002]

2. Description of the Related Art In a conventional character input system, a method of inputting an entire desired character into a single character input box is used. As a method for recognizing the characters thus input, there is a technique described in Japanese Patent Laid-Open No. 4-303279, in which the input data is divided into a radical and other parts for recognition. Further, there is a technique described in Japanese Patent Application Laid-Open No. 3-81374 which efficiently recognizes a character used repeatedly.

[0003]

In the above-mentioned conventional technique, although it was possible to efficiently recognize alphanumeric characters and the like in recognition of characters input by handwriting, complex character recognition such as Chinese characters is performed. Requires a complex algorithm, which has put a heavy burden on the developer.

Further, even in a system that depends on such a complicated algorithm, when the user actually enters a complicated character such as a Chinese character, the performance is satisfactory from the viewpoints of recognition accuracy, recognition time and the like. It was hard to say that they had.

Therefore, in general, since a method of inputting and recognizing only kana and alphanumeric characters and then converting into kanji by kana-kanji conversion is used, efficient character input is impossible.

According to the technique described in Japanese Patent Laid-Open No. 4-303279, the user can input Chinese characters, but after inputting one Chinese character in one character frame and the radical, I had to send a "confirmation" sign to the character recognition system and inform the recognition system of the cuts between the radicals and other parts. In addition, since one character is input in one input box, there is a possibility that a small part of a kanji with a large number of strokes may be crushed. Further, since it can be recognized only by being divided into two parts, "radical" and "non-radical", there is a drawback that it is less effective for characters in which the "non radical" part is complicated.

The present invention recognizes a complex character by dividing the complex character into simpler parts (not limited to radicals such as crowns and biases) and inputting them into different character frames for recognition. It is an object of the present invention to provide a handwritten character recognition system.

[0008]

In order to solve the above-mentioned problems, according to the present invention, an integrated display / input device and a pointing device for indicating a position on the integrated display / input device are provided, In a handwriting character recognition device for inputting handwritten characters by the pointing device and recognizing the input characters in an input area defined on the integrated display / input device, for a plurality of input frames in the input area , An input mode setting means for setting an input mode for writing each constituent part of a Kanji character divided into a plurality of constituent parts as a split input mode, and a character and a set of character constituent parts of the character in association with each other. Storage means for storing a plurality of sets, a set of character constituent parts input to the plurality of input frames in the divided input mode, and a character constituent part stored in the storage means And a character candidate extracting unit for selecting a character including a set of character constituent parts input by the plurality of input frames from the storage unit and displaying the selected character on the integrated display input device. be able to.

Further, the storage means stores first each storage means in association with each of the character constituent parts and an index uniquely determined for each character constituent part, and a set of the indexes. It is also possible to provide a second storage means for storing a code of a character formed by the set in association with each other.

Further, the character candidate extraction means compares the character constituent parts input to the plurality of input boxes with the character constituent parts stored in the first storage means, and compares the character constituent parts stored in the first storage means. Recognizing means for recognizing the set of indexes of the character component input to the above, the set of indexes recognized by the recognizing means and the set of indexes stored in the second storage means are compared, Retrieval means for retrieving character codes of candidate characters input to a plurality of input boxes from the second storage means, and output means for displaying on the integrated display / input device may be provided.

Further, the storage means may additionally store, for at least one set of the character constituent parts, a set of character constituent parts including a symbol obtained by simplifying the at least one character constituent part.

[0012]

[Function] In proportion to the complexity of the recognized character pattern,
The recognition accuracy decreases and the recognition time increases. However, the complexity of the characters is due to the combination of different parts (not limited to the radical), each part itself having a simple shape. Focusing on this point, we introduced a recognition method that divides the input characters into individual parts for recognition and displays characters that are combinations of the recognized parts as candidates.

When a character is decomposed into its constituent parts and recognized, if one character is input in a single input frame, a complicated pattern is crushed and input, or the system becomes a recognition unit. The question of how to determine the breaks in the parts remains. Therefore, we also introduced a split input method in which the user splits each character input.

In the division input method, the user divides one character,
This is a method of inputting in a plurality of input boxes. In this method, the system only needs to recognize each part in units of frames (input frames), so that the break of the recognition unit can be recognized without burdening the user or the system. Further, by inputting each part in each frame, it becomes possible to input a fine pattern in a large and clear manner, and the recognition rate is improved.

Further, if only this input and recognition method is used,
Since it is not efficient at the time of inputting alphanumeric characters, the user can dynamically select the conventional method of inputting the entire character in one input box for recognition and the method of dividing and inputting the above (optional). To) You can choose.

Hereinafter, in the present specification, the conventional method of inputting one entire character into one character frame and recognizing it will be described as a collective input and a collective recognition (characters collectively input at the time of recognition are radical and non- radical). Including one that is divided into parts), one character is divided into parts (not limited to radicals) and input into multiple character boxes,
A method of recognizing an input partial unit and recognizing an entire character from a combination of parts is called divided input and divided recognition.

Most characters can be divided into parts with relatively simple shapes, but some can only be divided into complicated parts. Further, even if it can be divided, if it is divided too finely, the usability may deteriorate. Therefore, by providing a function that allows a user to register a part of a character as a simpler symbol,
Even a complicated character pattern can be recognized with a simple pattern.

A combination of a character pattern group consisting of a pattern of a portion forming a character and an index such as a numerical value unique to each pattern, a character code for each character and an index indicating the character constituting portion is stored in the internal storage device. It has a character structure table which is held in association with each other, and a recognition buffer which stores a combination of recognized character structure parts.

When it is determined that a character has been input by the division input, the character pattern group is searched, and the index indicating the character pattern (component) recognized as the same as the input pattern is stored in the recognition buffer. Next, a combination of indexes is searched for in the character composition table having the same combination as the stored constituent parts, and if there is a corresponding combination of indexes, a character having a corresponding character code is displayed as a candidate. When the user selects one of the displayed candidates, the candidate is confirmed and the recognition buffer is cleared to prepare for the next character input.
If the user does not select a character and continues to input characters, the index corresponding to the input pattern is searched in the character pattern group, and the character structure table is searched by combining the newly obtained components. The display of candidates is repeated until the user selects a character or the recognition buffer becomes full.

When batch input is selected, the recognition operation is performed with the input pattern as the entire character.

A simplified pattern of a pattern (complex pattern) already registered in the character pattern group can be registered in the character pattern group. At this time, the simplified pattern is given the same index as the complicated pattern.

Further, two or more patterns (plural patterns) already registered in the character pattern group can be registered in the character pattern group with one simplified pattern (symbol). At this time, the index of the symbol is newly provided. Then, a combination having the index of the symbol is also provided for the character code in the character configuration table having the index of the plurality of patterns.
As a result, thereafter, character recognition is enabled by the symbol registered by the user.

Further, an input mode for writing the constituent parts of one Chinese character divided into a plurality of constituent parts into a plurality of input boxes of the input area is set as a divided input mode, and the characters and the characters of the characters are set. A plurality of pairs of constituent parts are stored in association with each other, and in the divided input mode, a set of character constituent parts and a blank is input to the plurality of input boxes, and the group and the stored character constituent parts are stored. It is also possible to compare the set with a set of character constituent parts input by the plurality of input frames and a character including a character constituent part other than the set can be selected and displayed.

[0024]

DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 shows a block diagram of the configuration of a handwritten character recognition device to which the present invention is applied. In FIG. 1, 20 is
It is an integrated display and input device. Integrated display / input device 20
On the display surface of the display device 15 which is a flat display,
The transparent tablet 14 for coordinate input is superposed and integrated, and the screen display and the coordinate input can be performed on the same surface. A pen 11 is a pointing device for designating a position (coordinates) on the tablet 14. The pen tip of the pen 11 is provided with a switch that indicates by turning on / off whether or not a position is instructed.

In the tablet 14, transparent electrodes (not shown) are arranged in a grid on a transparent glass panel, and a predetermined voltage is sequentially applied to each electrode. The coordinates indicated by the pen 11 are detected based on the electrostatic induction voltage induced between the electrode embedded in the indicated position of the pen 11 (or in the vicinity thereof) and the voltage detection timing. It The tablet of the present invention is
It is not limited to this type.

Reference numeral 10 is a CPU (central processing unit) that controls the entire handwritten character recognition apparatus by program control, and reference numeral 12 is a memory for storing programs, data, etc. necessary for the control processing of the CPU 10. 16
Is a keyboard, and 13 is an external storage device which is an auxiliary storage device.

FIG. 2 shows an example of an external view of the handwritten character recognition device. The CPU and the internal storage device are stored in the main body. In FIG. 2, reference numeral 20 denotes an integrated display input device in which a tablet and a display device are integrated. Reference numeral 11 is an input pen. In this figure, a cordless pen is used, and all input data is sent to the CPU via the tablet. The pen can also be used as a mouse, and the pen tip 3 can be provided with the same function as a left click of the mouse by lightly pressing it and a right click of the mouse by pressing a button 4. Reference numeral 5 is a character input frame for inputting characters. The input frame is realized by software as a program, and can be moved to a position on the integrated display / input device where the user can easily write. Reference numeral 13 is an external storage device. FD (floppy disk), HD (hard disk) in this part
Etc. are inserted and used. A keyboard is not connected to the example of FIG. 2, but it is also possible to use a keyboard as an input device.

FIG. 3 shows a functional block diagram of the handwritten character recognition apparatus according to the present invention. The internal storage device 12 is made to correspond to a combination of a character pattern group storage unit 38 including a pattern of a portion forming a character and an index unique to each pattern, a character code for each character, and an index indicating the character constituent portion. Character composition table 3
9 and a recognition buffer 40 for storing the recognized character component combinations. The input unit 31 converts the handwritten input character pattern into bit pattern data. The recognition unit 32 collates the bit pattern data with the character patterns stored in the character pattern group storage unit 38 in the internal storage device 12, acquires the index of the matched pattern, and stores it in the recognition buffer 40. Search unit 33
Searches the character configuration table 39 for a combination of indexes of the acquired character patterns. Get the character code corresponding to the matching index combination.
The output unit 34 outputs the character recognized based on the acquired character code. The registration unit 35 adds a newly input pattern to the character pattern group storage unit 38,
A new combination is added to the character configuration table 39. The abnormality processing unit 36 performs abnormality processing via the control unit 37 when an error or the like occurs in each unit. Control unit 37
Controls all parts so that each part operates normally while interdependently. The input unit 31, the recognition unit 32, the search unit 33, the output unit 34, the registration unit 35, the abnormality processing unit 36, the control unit 37.
Is implemented as software in the CPU 10.

An explanation will be given below taking Kanji (S-JIS code system) input as an example.

FIG. 4 shows a flow chart of the division recognition method, FIG. 5 shows an input frame, FIG. 17 shows a character pattern group storage section, and FIG. 18 shows an example of a character configuration table. As shown briefly in FIG. 2, in this device in which a tablet and a display device (LCD or the like) are integrated, the input frame (character input frame) is displayed as software and is displayed on the display device. Character input is performed by inputting characters in the entered input box. This input frame will be described in more detail with reference to FIG.

Input selection frame 5 is used for selection of division and batch input.
Click on 2, 53 to input characters from the left in the upper frame (51 in Fig. 5) of the input frame, and display the recognized character candidates in the lower frame (54 in Fig. 5) from the left. To do. The user confirms the character by clicking the frame in which the displayed character candidate exists.

A method of inputting "Tetsu" using split input will be described with reference to FIGS. 3, 4, 6, and 17.

When the user inputs handwriting as "age" (hand bias) (process 401 in FIG. 4, 61 in FIG. 6A), the input unit 3
1 (see FIG. 3) passes the input character pattern to the recognition unit 32 as data. The recognition unit 32 searches the character pattern group storage unit 38 for a character pattern that matches the acquired data (process 402 in FIG. 4), and the index 172.
(See FIG. 17) “2” is obtained and stored in the recognition buffer 40 (process 403 in FIG. 4). Next, the search unit 33 (see FIG. 3) searches the character composition table 39 for a character that matches the content of the recognition buffer, but since there is no character, it waits for input (processing 404, 405, 409, 40 in FIG. 4).
1). When the user inputs “Loaf” (process 40 in FIG. 4)
1, 62 in FIG. 6B), the recognition unit 32 obtains “3” of the index 173 (see FIG. 17) from the character pattern group and stores it in the recognition buffer 40 (processes 402, 40 in FIG. 4).
3), the contents of the recognition buffer 40 are as shown in FIG. When the search unit 33 searches this value from the character configuration table, the character code “90DC” is obtained (see FIG. 18), and the output unit displays “fold” as a candidate (FIG. 6B 6).
3) Do. If the user does not confirm, the input waits again (processing 404, 405, 406, 408, 40 in FIG. 4).
9, 401). When the user inputs "mouth" (Fig. 6
(C) 64), “1” of the index 171 (see FIG. 17) is stored in the recognition buffer by the recognition unit 32 (see FIG. 3), and the contents of the recognition buffer are “2”, “3”, “1”.
(Processes 402 and 403 in FIG. 4). When the search unit 33 (see FIG. 3) searches the character configuration table 39, the character code “934E” is obtained (see FIG. 18), and the output unit displays “Tetsu” as a character candidate (65 in FIG. 6C). 4) (processing 404, 405, 406 of FIG. 4). When the user clicks the candidate character (65 in FIG. 6 (c)), the character is fixed, and the control unit clears the recognition buffer in preparation for the next character input (processing 408, 411, 412 in FIG. 4).

Here, the character composition table is searched by the combination of the indexes, but if the permutation of the index (taking into account the order) is used, the speed of the search is further improved.

FIG. 7 shows a selection method between the batch input mode and the split input mode (the two modes are collectively referred to as an input mode). In this example, the input / recognition method is determined by clicking the divided input selection frame (52 in FIG. 5) or the collective input selection frame (53 in FIG. 5) of the input frame (FIG. 7).
There is also a method of processing 71) of (1) to automatically determine division or collective input depending on the position of input to the input cell. For example,
When a character is entered in the center of the input box (Fig. 8 (a))
Is batch input, and when input is made at a position deviated from the center (FIG. 8B), it is divided input.

Further, although five input frames are prepared in this example, when division input is performed in a case where the input frames are small, it is possible to scroll and continue input when the input frames are full. FIG. 9 shows an input example in the case where there are only three input frames.

In FIG. 9A, the input box 9 is input by the user.
1, 92, 93 have all been input, and the candidate display frame 9
"Kei" is displayed in 4. Here, when the user clicks on the frame 94 in which the candidate characters are displayed, “Ken” is confirmed. When the candidate display frame 95 in which the candidate characters are not displayed is clicked, the recognition device determines that the split input is still in progress, and scrolls the input frame to the left (FIG. 9).
(B)). When the user continuously inputs "Sat" (Fig. 9
(C) 93), "pump" in the candidate display frame (Fig. 9 (c) 94)
Is displayed.

As an application of divided input, it is also possible to input a character that is not stored correctly. Generally, there are many kanji that can be read but cannot be written. When inputting such kanji, the inaccurate part is left blank (see Fig. 10).
102), input the correct part (10 in FIG. 10).
1, 103), the character including the combination of the input parts can be searched from the character composition table, displayed as a candidate (104 in FIG. 10), and the user can select from the candidates. In a device that can search by radicals, it is impractical because the search is performed by "彳", so it is necessary to select from a myriad of characters. Therefore, the correct character can be easily obtained.

In order to realize this function, FIG.
When a blank such as 0 is input as a character constituent part, a character including the input character constituent part and an unknown character constituent part is displayed as a candidate in process 404 of the flowchart of FIG. do it.

Next, a method of registering a new pattern for replacing a relatively complicated pattern with a simpler pattern will be described.

FIG. 11 shows a screen for pattern registration.
The new pattern input unit 111 is a unit for inputting an input pattern to be registered in a character pattern group referred to for recognition. The new pattern input area is divided into 16x16 pixels, and the specified pixel is highlighted by clicking the pixel with the pen. Input box 112
Is used to input and select a character component corresponding to the new input pattern. Registration is started by inputting a new pattern and character configuration portion and clicking the registration execution key 113. Before clicking the registration execution key,
The registration work can be stopped by clicking the registration stop key 114.

As long as the input of Chinese characters is limited, the character component corresponding to the new pattern is an existing character component or a combination thereof. FIG. 12 shows a flow chart of new pattern registration. A method for registering a new input pattern in an existing character component (processing 121, 122, 12
3,124,125,126) and a method of registering with a combination of existing character constituent parts (processing 121, 122, 12).
3, 124, 125, 127, 128, 129).

As a method of registering a new pattern in the existing character component, "I" is used as a simple input pattern "K".
Here is an example of registering in.

At this time, the last index in the character pattern group is Max (FIG. 19), and the index is 30.
As an example, it is assumed that “Iku” is registered (191 in FIG. 19). The user inputs a pattern of "ki" as a new input pattern in the new pattern input section by the Dodd pattern (131 in FIG. 13A, process 121 in FIG. 12). Next, when "Iku" is input in the input frame by handwriting as a character component (132 in FIG. 13B), "Iku" is displayed in the candidate frame (133 in FIG. 13B). The frame 133 in which "" is displayed is clicked to select the character component (processing 122 in FIG. 12). When the registration execution key is clicked (process 124), the registration work starts.

First, the recognition unit 32 (see FIG. 3) searches the character pattern group storage unit 38 to check whether or not the character component specified by the user already exists in the character pattern group (see FIG. 12). Process 125). Since there is a pattern of "some" as the index 30 (FIG. 19), the registration unit 3
5 (see FIG. 3) adds a combination of index 30 and "ki" to the character pattern group (193 in FIG. 19, FIG. 12).
Processing 126). After that, by inputting like 141 and 142 in FIG. 14, the character "machine" is displayed as a candidate (143).

Next, a method of registering a new pattern to a combination of existing character components will be described by registering "Δ" for "greed" which is a combination of "valley" and "missing".

Consider a case where the last index in the character pattern group is Max (204 in FIG. 20) and "valley" and "missing" are present as indexes 50 and 60 (202 and 203 in FIG. 20). The user inputs “Δ” into the new pattern input unit (111 in FIG. 11) as a new pattern (process 121 in FIG. 12, 151 in FIG. 15A) and “valley” and “missing” in the input frame. Input (15 in Fig. 15 (b))
2,153). Since “valley” and “missing” are displayed in the candidate display frame (154 and 155 in FIG. 15B), click the displayed frame (154 and 155) and select a combination of character component parts. (Process 122 of FIG. 12). here,
The registration work is started by clicking the registration execution key (process 124).

The recognition unit 32 (see FIG. 3) searches the character pattern group storage unit 38 to check whether or not the designated character constituent portion exists (process 125 in FIG. 12). If it doesn't exist,
The registration unit 35 uses the new indexes “Max + 1” and “Δ”.
The combination of is registered in the character pattern group storage unit 38 (205 in FIG. 20, processing 127 in FIG. 12). Next, the search unit 3
3 (refer to FIG. 3) searches the character configuration table 39 for a character having a combination of indexes 50 and 60, and searches for a combination of the indexes 50 and 60 (for example, 21 in FIG. 21).
All of 1) are copied and newly added to the character composition table (process 128 in FIG. 12). Finally, the combination of the character composition tables copied by the registration unit 35 is “Max +
1 ”(for example, 212 in FIG. 21, processing 129 in FIG. 12). As a result, when input is performed as indicated by 161 and 162 in FIG. 16, the character “Pi” is displayed as a candidate (163).

For a character having a completely new pattern such as an external character, it is possible to recognize it by newly registering both the character component and the input pattern using the new pattern input section (111 in FIG. 11). can do.

[0050]

As described above, according to the present invention, a complicated character is divided into simpler parts (not limited to the radicals such as crowns and biases) and each character is input into a different character frame for recognition. This makes it possible to efficiently recognize complicated characters. As a result, the recognition speed or accuracy of handwriting input of complicated characters can be improved.

[Brief description of drawings]

FIG. 1 is a block diagram of a handwritten character recognition device according to the present invention.

FIG. 2 is an external view of a handwritten character recognition device according to the present invention.

FIG. 3 is a functional block diagram of a handwritten character recognition device according to the present invention.

FIG. 4 is a flowchart showing a procedure of a divided character recognition method.

FIG. 5 is an explanatory diagram of a concept of a character input frame.

FIG. 6 is an explanatory diagram of an example of how divided characters are input and recognized.

FIG. 7 is a flowchart showing a procedure of division and batch input selection.

FIG. 8 is an explanatory diagram of an example of a division / collective input selection method.

FIG. 9 is an explanatory diagram of scrolling when the input frame is full.

FIG. 10 is an explanatory diagram of an example of inputting and recognizing a character that is not accurately stored.

FIG. 11 is an explanatory diagram of a screen example of new input pattern registration.

FIG. 12 is a flowchart showing the procedure of a new input pattern registration method.

FIG. 13 is an explanatory diagram of an input example of a user registration pattern.

FIG. 14 is an explanatory diagram of a user registration pattern input and a recognition example.

FIG. 15 is an explanatory diagram of an input example of a user registration pattern.

FIG. 16 is an explanatory diagram of a user registration pattern input and recognition example.

FIG. 17 is an explanatory diagram of a character pattern group storage unit.

FIG. 18 is an explanatory diagram of a character composition table.

FIG. 19 is an explanatory diagram of a character pattern group storage unit.

FIG. 20 is an explanatory diagram of a character pattern group storage unit.

FIG. 21 is an explanatory diagram of a character composition table.

[Explanation of symbols]

10 ... CPU, 12 ... Internal storage device, 31 ... Input unit, 3
2 ... Recognition unit, 33 ... Search unit, 334 ... Output unit, 35 ... Registration unit, 36 ... Abnormality processing unit, 37 ... Control unit, 38 ... Character pattern group storage unit, 39 ... Character configuration table, 40 ... Recognition buffer.

Claims (1)

[Claims] 1. An integrated display / input device, and a pointing device for designating a position on the integrated display / input device, wherein the pointing device is provided within an input area defined on the integrated display / input device. In a handwritten character recognition device for recognizing input characters by inputting handwritten characters with a device, for each of a plurality of input frames in an input area, each constituent part of one kanji character divided into a plurality of constituent parts An input mode setting means for setting an input mode to be written as a split input mode; a storage means for storing a plurality of sets of a character and a set of character constituent parts of the character in association with each other; Characters input by the plurality of input boxes are compared by comparing a set of character constituent parts input in a frame with a set of character constituent parts stored in the storage means. The characters, including a set of growth moiety selected from the storage unit, handwriting recognition apparatus characterized by comprising a, a character candidate extracting means for displaying on the integrated display and input device on.