JP2007072691A - Handwritten character recognition device and handwritten character recognition method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To achieve a handwritten character recognition device for recognizing a character string written with various inclinations and a handwritten character string written with fixed inclination. <P>SOLUTION: The input trajectory coordinate rotation of each character and the matching of the rotation coordinate data and erect handwritten trajectory character dictionary data is operated, and a character code and scores and the angle of rotation whose matching with the dictionary data is the highest are output as recognition results. The character font of the recognition result is rotated and displayed according to the angle of rotation of the recognition result, and when the recognition result is different from the intended character, and the result is different from the input angle, a user inputs the instruction of correction of the angle of rotation of the character on the character font. Furthermore, the input trajectory coordinate rotated trajectory coordinate data of each character are selected, and matched with the selected trajectory data. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention recognizes handwritten characters written online, and particularly recognizes a memo or the like in which characters entered at different angles are mixed.

  Conventionally, in a character string reading device such as Patent Document 1, a photoelectric conversion unit that reads a character string on a form, a character detection and clipping unit that sequentially extracts character images from the read character string image one character at a time, and a character reading direction Is rotated in units of 90 °. Character rotation recognition means for obtaining a character recognition result for each rotation angle for each character image and a character string for each angle by extracting the recognition character of the same rotation angle from the character of the recognition result obtained by this character rotation recognition means The configuration is such that the means for outputting as a candidate, the character string dictionary for storing the character string to be read in advance, and each character string candidate output from the output means are compared with the character string dictionary to obtain the recognition result of the character string. .

  Conventionally, a handwriting input data processing apparatus such as Patent Document 2 determines whether or not a character has a horizontal / vertical stroke, detects the inclination of the character based on the horizontal / vertical stroke, and determines the detected inclination. Accordingly, the character is rotated by the character rotating means to correct the inclination of the character.

  In addition, the rotationally misaligned character recognition apparatus such as Patent Document 4 has a configuration in which a mark indicating a direction is provided in the symbol information.

In addition, in the character inclination angle detection correction method described in Patent Document 3, etc., a character string inclination extracting means for generating a line segment connecting the centers of a plurality of character frames and obtaining the inclination of the line segment, and rotating the character with the inclination. It is comprised with the character rotation means to make it rotate.
Japanese Patent Publication No. 07-95332 Japanese Patent Registration No. 3299614 JP 63-308689 A JP 62-127984 A

  In the conventional character string reading device of Patent Document 1 or the like, since the document processing is the main focus, character rotation is processed in character string units, and the angle in the rotation direction is limited to 90 ° units. In the above configuration, it is not considered that a character string whose angle with respect to the reading direction is different for each character unit. For example, a handwritten memo input in which 90 ° characters and 45 ° characters are mixed is used. In the above configuration, in the above configuration, a 90 ° character becomes 0 ° by a rotation process in units of 90 °, but when 90 ° is 0 °, a 45 ° character has a slope of 315 °. For this reason, the recognition result of 45 ° characters tends to be unrecognizable or different from the input characters. Even if the character string dictionary is used for collation based on the character unit result, a correct result cannot be obtained.

  In addition, since the handwriting input data processing apparatus such as the conventional patent document 2 detects the inclination of the character based on the horizontal / vertical stroke of the character, the inclination is detected in the case of a character that originally does not have the horizontal / vertical stroke. In addition, in the case of handwritten characters, it is difficult to detect the inclination because strokes, which are essentially horizontal straight lines, are also curved or slanted compared to printed characters.

  In addition, in the rotationally misaligned character recognition apparatus such as Patent Document 4, the direction of the character is determined by a mark indicating the direction of the character. Therefore, if the technique is applied to an on-line character recognition apparatus for handwritten characters, Since a special stroke indicating the direction of the mark must be added as a mark, the operation becomes complicated.

  In addition, in the character inclination angle detection correction method disclosed in Patent Document 3, etc., the inclination is obtained by the character string inclination extracting means for generating a line segment connecting the centers of the frames of a plurality of characters and obtaining the inclination of the line segment. Although the inclination of the character string unit can be obtained, it cannot be applied well when the character in the character string is different from the inclination of other characters or when the inclination of the character string is different from the inclination of the character.

  The present invention has been made in view of such problems.

  By providing input trajectory coordinate rotation means for each character, rotational coordinate conversion trajectory data having a plurality of angles (20 °, 45 °, 70 °, 90 °, etc.) for each character is obtained. In the conventional handwritten character recognizing means, the rotation coordinate conversion trajectory data is matched with the upright handwritten trajectory character dictionary data, and the result of recognizing the character code, the score and the rotation angle with the highest degree of coincidence with the dictionary data Output as.

  Furthermore, display means for rotating and displaying the character font of the recognition result according to the rotation angle of the recognition result is provided, and when the recognition result is different from the intended character and the result is different from the input angle, the user can display the character font on the character font. Thus, the user's correction means is provided by providing correction instruction input means for the rotation angle of the character.

  Also, a recognition means for selecting the locus coordinate data rotated by the input locus coordinate rotation means for each character is provided, and the recognition time is shortened by matching the selected locus data.

  Further, the handwritten character string input means, the character cutting means for dividing the inputted handwritten character string trajectory data for each character, the input trajectory coordinate rotating means for each character, and the coordinate rotating means rotated at a plurality of angles. A selection means for selecting locus coordinate data of an upright image of a character from a plurality of input locus coordinate data, a recognition means for recognizing the selected locus coordinate data, and angle information as a result of the recognition means is determined as a character string. If it is determined that the angle is wrong, a re-recognition means is provided to recognize it again according to the angle of another character. Perform appropriate processing. Provided is a handwritten character recognition device capable of obtaining a correct recognition result by performing processing corresponding to the inclination when the inclination is constant in the character string.

  By providing input trajectory coordinate rotation means for each character, rotational coordinate conversion trajectory data of a plurality of angles is obtained. As a result, the upright data is obtained from the tilted input character, matched with the upright handwriting trajectory character dictionary data, and the character code, score and rotation angle with the highest degree of coincidence with the dictionary data are output as the recognition result. .

  Furthermore, display means for rotating and displaying the character font of the recognition result according to the rotation angle of the recognition result is provided, and when the recognition result is different from the intended character and the result is different from the input angle, the user can display the character font on the character font. Thus, the user's correction means is provided by providing correction instruction input means for the rotation angle of the character.

  Further, by providing means for selecting trajectory coordinate data rotated by the input trajectory coordinate rotating means for each character and matching the selected trajectory data, the recognition time can be shortened and the recognition rate can be improved.

  Further, a handwriting character string input means, a character cutting means for dividing the inputted handwriting character string trajectory data for each character, an input trajectory coordinate rotation means for each character, and a plurality of pieces rotated at a plurality of angles by the coordinate rotation means. A selection means for selecting the locus coordinate data of the upright image of the character from the input locus coordinate data, a recognition means for recognizing the selected locus coordinate data, and angle information as a result of the recognition means is determined as a character string. By providing a re-recognition means for re-recognizing according to the angle of other characters when it is determined that the angle is wrong, if there are multiple characters with different inclinations in the character string, it corresponds to the inclination. Process. Provided is a handwritten character recognition device capable of obtaining a correct recognition result by performing processing corresponding to the inclination when the inclination is constant in the character string.

  By providing the above means, it is possible to realize a device that can naturally input characters with different inclinations and can also input character strings with the same inclination as in handwritten notepads.

[First embodiment]
FIG. 1 shows an actual usage pattern of the present invention. Reference numeral 1 denotes an information device, a liquid crystal display unit, a display-integrated coordinate input unit, a CPU, ROM, RAM, HD, a slot into which a recording medium is inserted, an interface such as a USB for connecting to other information devices, and other communication The wireless communication unit communicates with a personal computer having a function. The handwritten character trajectory input by the coordinate input unit is recognized by the processing of the present invention, and the recognition result is displayed.

  Reference numeral 2 denotes a general liquid crystal display unit which displays a handwriting trajectory input to the liquid crystal display unit, a recognition result character font, and the like. Reference numeral 3 denotes a general local area network, which communicates with other personal computers. Reference numeral 4 denotes a wireless communication antenna for sharing a screen displayed on one information device and exchanging input text information.

  Reference numeral 5 denotes a general personal computer, which is connected to the information device of the actual usage pattern of the present invention through a local area network.

  FIG. 2 is a block diagram of the portable information device according to the embodiment of the present invention.

  Reference numeral 2-1 denotes a general coordinate input unit, which is composed of a transparent resistive film digitizer or the like and is connected to 6 CPUs via 11 system buses. A general liquid crystal display unit 2-2 includes a liquid crystal display element, a liquid crystal control circuit, and a display memory, and is connected to 6 CPUs via 11 system buses. In response to an instruction from the CPU 6, the handwriting trajectory display and the character font display are executed. Reference numeral 6 denotes a general CPU, which is connected to a RAM 7, a ROM 8, a liquid crystal display unit 2-1, etc. via a system bus 11, and is processed by a program stored in the ROM 8. Perform the action. Reference numeral 7 denotes a general RAM, which is used as a work area for storing image data and for taking a photographed image processing program. Reference numeral 8 denotes a general ROM, which stores a captured image processing procedure of this patent, an application processing procedure on another portable terminal, and the like. In addition, a CCD control program, a liquid crystal display control program, and other processing procedures are also stored. Reference numeral 9 denotes a small hard disk used in a general portable device, which is connected to 6 CPUs via 11 system buses. The input text file is stored. Reference numeral 10 denotes a general local area network interface card, which communicates with other personal computers. A general system bus 11 exchanges data between the CPU 2-6 and the ROM / RAM or other devices. A general wireless communication card 12 has a wireless LAN controller and antenna built therein. By passing a prescribed command to the controller in the communication card, it is possible to connect to the wireless local area network and execute transmission / reception of digital images and transmission / reception of control commands. The transmission / reception protocol itself is the same as the conventional method.

  FIG. 3 is a block diagram of the handwritten character recognition apparatus of the embodiment of the present invention.

  Reference numeral 2-1 denotes a general coordinate input unit, which is composed of a transparent resistance film digitizer or the like, and can read XY single table data of a pen-written locus. Reference numeral 3-1 denotes a trajectory position coordinate rotating unit that performs coordinate conversion so that the position coordinate data representing the trajectory of the handwritten character input in 2-1 is rotated at a specified angle. 3-2 is a handwritten character recognition dictionary in which normalized data of general handwritten trajectory data of characters to be recognized is stored. This normalized data is obtained by normalizing handwritten character data in an upright state. 3-3 is a handwritten character recognition means, which extracts a feature point of a handwritten character trajectory written by a general online handwritten character recognition algorithm, and a matching process with a handwritten character dictionary of 3-2 And the most similar dictionary data is set as the first candidate.

  3-4 is a recognition result selection means, and the handwritten character recognition means in 3-3 makes a judgment for each character on the basis of the degree of coincidence with the dictionary and information on the rotation angle of the coordinates, and is the most evaluated. Candidates with high points are selected and sent to the display means 2-2. Reference numeral 3-5 denotes a rotation angle change instructing unit for instructing correction of the result when the rotation angle of the recognition result character displayed on the display unit 2-2 is different from the user's desired angle. Means. With this means, the user presses the pen on a character with a different result and inputs correction angle information, so that the means 3-5 detects the angle information and passes it to the locus position coordinate rotation means 3-1. The trajectory coordinate data is rotated at the designated angle, and re-recognition is performed using the data. A general liquid crystal display unit 2-2 displays a handwriting trajectory and a character font.

  FIG. 4 is a screen for inputting handwritten characters and displaying results. The “empty” character of the handwritten text displayed in FIG. 4 is recognized, and the “empty” character of the font is displayed as shown in FIG.

  FIG. 6 is a flowchart.

  FIG. 7 is an example of an instruction operation screen when FIG. 8 shows that the desired result of handwriting recognition is not obtained.

  FIG. 9 is a flowchart of the rotation angle change instruction process.

  FIG. 10 shows another example of a candidate change instruction operation screen.

  First, the recognition process for one character will be described with reference to the flowchart of FIG.

  In step S6-1, a recognition process for one character is started. The work area and the like used by the recognition process for one character are initialized.

  In step S6-2, a process of reading one-character handwriting locus position coordinate data into the buffer area is performed. One-character handwriting trajectory position coordinate data is input by coordinate input means constituted by a general transparent resistive film digitizer represented by 2-1, in FIG. When a user uses a pen to input a one-character trajectory on the transparent resistive film digitizer and the pen-up time exceeds the character cut-off time, a one-character trajectory input process is determined, and the position coordinate data string of the trajectory is shown in FIG. Processed by the CPU -6, stored in a specific area in the RAM 7 in FIG. 2, and the program shown in this flowchart is started.

  In step S6-3, a process for converting the position coordinate data of the trajectory of one character into a rotational coordinate is performed. In this embodiment, conversion data of position coordinates rotated by 30 ° is created. Unrotated 0 ° data, 30 ° rotated data, 60 ° rotated data, 90 ° rotated data, −30 °, −60 °, and −90 ° rotated data are created.

If you want to perform rotational coordinate transformation around the origin,
New X = (x * cos (r))-(y * sin (r))
New Y = (x * sin (r)) + (y * cos (r))
It is only necessary to process the XY data sequence of the position coordinates of the trajectory according to the above formula.

  In this embodiment, in order to simplify the explanation, the increment is 30 °, but it is needless to say that the increment may be 15 ° or 5 °.

  In step S6-4, a normalized feature point extraction process is performed for each rotation data. The start point midpoint end point of one trajectory of the data for each rotation angle transformed in step 6-3 is extracted and normalized to a range from 0 to 100. The minimum XY coordinate of the character trajectory is normalized to 0 and the maximum XY coordinate is normalized to 100. In the handwritten character recognition dictionary of 3-2, data normalized in the same way from 0 to 100 of the start point, the middle point, and the end point of the trajectory data of the handwritten character in an upright state written in a standard way is written. It is remembered. The recognition algorithm itself for matching the input data with the dictionary may be the same as the conventional recognition algorithm.

  In step S6-5, matching processing with dictionary data is performed. A matching process is performed between the start point midpoint end point of the input character trajectory normalized in step S6-4 and the start point midpoint end point of the dictionary data trajectory in the 3-2 handwritten character recognition dictionary having the same number of strokes. If handwritten characters with exactly the same shape as the dictionary are input, the distance difference will be 0 because the start point, midpoint and end point of the matching strokes are the same. Arise.

In this way, matching processing is first performed between the input handwriting trajectory data as it is and all data of the trajectory of the dictionary in the 3-2 handwriting character recognition dictionary with the same number of strokes. As a result, dictionary data having a small distance difference between the feature points of the handwriting is output. The output includes a score that is the sum of the distance differences, a character code corresponding to the data, and angle data. (In this case, the input coordinates remain 0 °)
Next, the handwriting trajectory data rotated at 30 ° is matched with all the data of the trajectory of the dictionary in the 3-2 handwritten character recognition dictionary with the same number of strokes. Dictionary data with a small distance difference between points is output as a result. The output is a score that is the sum of the distance differences, a character code corresponding to the data, and an angle of 30 °.

  Similarly, matching processing with data of 60 °, 90 °, −30 °, −60 °, and −90 ° is performed.

  In step S6-6, a process of storing the degree of coincidence with the dictionary data, the character code, and the rotation angle information is performed. Information on the character code of the most similar dictionary data obtained at step S6-5 and the degree of coincidence of the dictionary data and the rotation angle of the input data are stored in a specific area of the RAM 2-7.

  In step S6-7, a process for determining a candidate having the highest degree of matching is performed by comparing the maximum degree of matching of each angle. Data having the highest degree of coincidence is extracted from the data of each angle stored in step S6-6. For example, “sky” in the example of FIG. 4 is input in a state tilted at −30 ° and written with eight strokes. Matching with a dictionary of 8 brushes and extracting the data with the highest degree of coincidence, but as shown in the figure, the degree of coincidence of the input trajectory data rotated 30 ° from the degree of coincidence with the 0 ° data The result is chosen.

  In step S6-8, processing is performed for outputting candidate character codes together with the degree of coincidence and angle data in descending order of the degree of coincidence of the determined angle. If the degree of coincidence of the data of −30 ° is high in the step S6-7, the second candidate with the next highest degree of coincidence at the same angle is output in order of the degree of coincidence.

  In step S6-9, the one character recognition process is terminated. Processing for releasing the work area and buffer area secured in step S6-1 is performed.

  By performing the processing as described above, the recognition processing for the input characters can be performed well with different angles for each character as shown in FIG. As shown in FIG. 4, the characters of “blue sky” are input at different inclinations, but the recognition result of the first character is a “blue” angle of 30 ° and the recognition result of the second character by the above recognition processing. If the “i” angle is 0 ° and the third character recognition result is “empty” angle −30 °, the display is as shown in FIG.

  The above example is an example in which recognition is successful, but depending on the written handwriting, a recognition result different from the result intended by the user may be displayed. A recognition result correction instruction process in such a case will be described.

  FIG. 7 is a display example showing a state in which the first recognition result is displayed by the processing described in FIG. In this case, if the result is the result intended by the user, the next character input is performed as is. If the result is not the intended result, the specified angle of the character is input by operating the pen with the pen. The operation is performed. The locus 7 in FIG. 7 represents the input position coordinates of the rotation instruction for the character input on the character “three”.

  FIG. 8 is a display example of the result of re-recognition based on the result instructed in FIG.

  FIG. 9 is a flowchart showing the rotation angle change instruction process.

  The processing will be described along the flowchart of FIG.

  In step S9-1, the rotation angle change instruction process is started. Initialize the work area and buffer area.

  In step S9-2, it is checked whether pen-down has been input. If pen-down is input, the process proceeds to step S9-3. If pen-down is not input, the process proceeds to step S9-11, and another process is performed.

  In step S9-3, it is checked whether or not there is a character on the coordinates of the position coordinates that are pen-down and input. If there is a character on the coordinate, the process proceeds to step S9-4. If no character exists on the coordinate, the process proceeds to step S9-11, and the process proceeds to a normal character input process or the like.

  To check whether or not there is a character on the coordinates, if the coordinates are within the circumscribed rectangular frame of the displayed character, processing for determining that the character is on the character is performed.

  By checking with all circumscribed rectangle data of the character in the screen shown in FIG. 7, it is checked whether it is on the character or not.

  In step S9-4, a process of reading the trajectory data of the characters on the coordinates is performed. As shown in FIG. 7, the information on the currently displayed character of the handwriting recognition result is stored in a specific area of the RAM 2-7. The information includes coordinate information of a circumscribed rectangular frame of the displayed character, a character code, and coordinate data of a handwriting locus used for recognition. Since the circumscribed rectangle in which a certain character is displayed does not display other characters, the coordinate data of the handwriting trajectory of the character retrieved with the coordinate information of the circumscribed rectangle in the check in step S9-3. Read the file. In step S9-5, processing for storing the position coordinate data of the locus is performed until pen-up is performed. The position coordinate data sent from the coordinate input means 2-1 in FIG. 3 is stored in a specific area of the RAM 2-7.

  In step S9-6, a process for obtaining an angle between the start point and the end point is performed. The first pen-down position of the trajectory input in the rotation change instruction operation is used as the start point, and the position immediately before pen-up is the end point. Ask.

  In step S9-7, a process of rotating and converting the trajectory data read at the designated angle is performed. As shown in FIG. 7, when the character “three” is tilted by −30 ° and the user rotates the character 30 ° and desires the recognition result in the state of 0 °, the user moves the character on the character with the pen. If a straight line having an inclination of 30 ° with respect to the X axis is entered so as to rotate 30 ° along the locus, 30 ° is obtained in step S9-6, and the following coordinate conversion is performed based on the angle.

New X = (x * cos (r))-(y * sin (r))
New Y = (x * sin (r)) + (y * cos (r))
The character in FIG. 7 is rotated to create 0 ° handwritten character trajectory data.

  In step S9-8, matching processing between the converted trajectory data and dictionary data is performed. Matching processing is performed on the handwritten character trajectory data converted in step S9-7, and the matching processing itself is the same as that described in step S6-5, and thus description thereof is omitted.

  If the handwritten trajectory character data to which the recognition result of “three” tilt of −3 ° in FIG. 7 is input is the result of tilting by −30 °, the data tilted by −30 ° is rotated by 30 ° to 0 °. The recognition result is “mi” and the character inclination is 0 ° as shown in FIG.

  If it is different from the previous recognition result in step S9-9, the recognition result is displayed. If the recognition result obtained in step S9-8 is given an angle of 0 ° during the angle indicating operation and the handwriting trajectory data is the same as the previous one, the result is the same, so the result is not displayed again and is left as it is. As shown in FIG. 7 and FIG. 8, if the first result is a “three” slope of −30 ° and the next result is a “mi” slope of 0, the characters of the “three” slope of −30 ° are erased first. Then, a character “mi” with a tilt of 0 ° is displayed at the same position.

  In step S9-10, the rotation angle change instruction process is terminated. Perform processing to release the secured work area and buffer area.

  By adding the processing described above, if the character and the inclination of the result that were automatically recognized first are different from the desired case, direct the inclination to the displayed character with a pen. The desired recognition result can be obtained by rotating the character at a specified angle.

  Other than the above method, a user interface may be considered in which candidates for recognition results are displayed and when the angle needs to be changed, the angle is directly input.

  FIG. 10 is a display example showing an example of candidate display.

When the result of “three” is not a desired one, when the pen is stopped on the character “three” and pressed for 3 seconds or longer, the candidate selection screen of FIG. 10 is displayed. (This candidate display selection process itself is a general process and will not be described.)
Each recognition candidate is displayed in 10-1, 10-2, and 10-3. When there are many candidate results and it cannot be displayed on one screen, it is displayed on a plurality of screens by a general scroll bar or the like.

  If the currently selected candidate “three” is not the desired result and “river” is the desired result, the display area of “river” 10-3 can be pressed with a pen.

  If there is no candidate of 10-1, 10-2, 10-3, and there is no desired object in the other candidates, and there is no candidate of the desired angle, an operation to directly input a different angle is performed. Do. When the angle addition area 10-4 in FIG. 10 is pressed with a pen, the angle in the 10-6 area is added, and when the angle subtraction area 10-5 is pressed with the pen, the angle in the 10-6 area is changed. Subtracted. By this operation, a desired angle is displayed in the angle display area of 10-6. If the display angle in the angle display area 10-6 is a desired angle, the angle determination area 10-7 is pressed with a pen.

  When the 10-7 angle determination area is pressed with a pen, the processing from step 9-7 is performed at the angle displayed at 10-6. For example, if the angle displayed at 10-6 is 180 °, the handwriting trajectory data corresponding to the displayed character is converted into position coordinates rotated by 180 °, and the hand 3-2 in FIG. Matches with the letter dictionary and displays the result.

  By performing processing in this way, it is possible to provide a handwritten character recognition apparatus that instructs angle change consistently with candidate selection.

  By configuring the processing in this way, it is possible to provide a handwritten character recognition device such as a handwritten memo that can obtain a desired recognition result even when written with a different inclination for each character. Further, it is possible to provide a handwritten character recognition device that can easily indicate a desired angle and perform re-recognition even if the initial recognition result is not a desired one.

[Second Embodiment]
In the first embodiment, handwriting trajectory coordinate data obtained by converting a plurality of rotation angles at a constant angle is created by the trajectory coordinate rotating means of 3-1, and dictionary matching is performed on the input handwriting trajectory. The recognition result and angle were output based on the degree of coincidence of the result with the dictionary.

  However, since a plurality of handwriting trajectory coordinate data is created, there is a drawback that the recognition time increases depending on the number of handwritten locus data. In view of this, there may be a configuration in which the rotation handwriting trajectory data created by the trajectory coordinate rotation means is checked to check whether the rotation handwriting trajectory data is valid, thereby narrowing down and recognizing the rotation handwriting trajectory data considered to be effective. It is done.

  FIG. 11 is a processing configuration diagram of the second embodiment. Parts having the same part numbers as those in the process configuration diagram of FIG. 3 of the first embodiment are the same and will not be described.

  Reference numeral 11-1 denotes a rotation data selection means for selecting data in an upright state of characters from the shape information of the rotated handwriting trajectory data. It is ideal if it is uniquely determined, but it is only necessary to select a small number such as 2 to 3 from a lot of data.

  FIG. 12 is a conceptual diagram of handwriting trajectory data rotation processing. 12-1 is an input character image, and the result of rotating the handwriting trajectory data at a certain angle is the rotation result A of 12-2, the rotation result B of 12-3, and the rotation result C of 12-4. .

  FIG. 13 is a flowchart showing the selection process of the input trajectory rotation data.

  Processing will be described with reference to the flowchart of FIG.

  In step S13-1, the selection process of the input trajectory rotation data is started. Work area and buffer area are secured and initialized.

  In step S13-2, a process for reading the handwritten position coordinate data of one character to be processed into the buffer is performed. Since this process itself is the same as the process of step S6-2, description thereof will be omitted.

  In step S13-3, rotation coordinate conversion is performed on the handwritten position coordinate data of one character read in step S13-2. In the second embodiment, input handwriting position coordinate data rotated by 15 ° is created. In Example 2, when writing horizontally, it is not assumed, so input handwriting position coordinate data of seven angles of −45 °, −30 °, −15 °, 0 °, 15 °, 30 °, and 45 ° are obtained. Created. If horizontal characters are also input, the interval between −90 ° and 90 ° may be set in increments of 15 °.

  In step S13-4, a circumscribed rectangle of the position coordinate data of the handwritten trajectory data subjected to rotation conversion is obtained. Imageally, as shown in 12-3 of FIG. 12, when the minimum and maximum XY coordinates of the handwritten character trajectory are obtained and the circumscribed rectangle that apex is obtained, the circumscribed rectangle enclosing the character like 12-3 It becomes. In terms of processing, the maximum and minimum XY coordinate values of the position coordinate data of the handwriting trajectory data are obtained for each angle.

  In step S13-5, the area of the circumscribed rectangle is obtained for each angle. The area is obtained based on the maximum and minimum XY coordinate values for each angle in step S13-4.

DX = maximum X coordinate value-minimum X coordinate value DY = maximum Y coordinate value-minimum Y coordinate value circumscribed rectangle area at a certain angle = DX x DY
It is.

  In step S13-6, the minimum area data is selected from the circumscribed rectangle for each angle. The smallest angle is obtained from the circumscribed rectangular areas of the angles of −45 °, −30 °, −15 °, 0 °, 15 °, 30 °, and 45 ° obtained in step S13-5. If it cannot be uniquely determined by the shape of the input handwriting trajectory data (if the difference between the area of the A angle and the area of the B angle is a small difference of 1% or less of the total area), the two angles are regarded as correct. To do.

  As can be seen from the example of FIG. 12, the area of the circumscribed rectangle of the rotation result A of 12-2 in which the input character is rotated and the oblique character is made upright is the area of the rotation result B and the area of the rotation result C. When compared with, the data is minimized and the data is selected.

  In step S13-7, the selected position coordinate data is sent to the recognition processing module. In step S13-6, the selected position coordinate data (rotated angle value and XY position coordinate data string of the trajectory of one character converted by the angle) is sent to the recognition processing module shown in the flowchart of FIG. The recognition result is obtained and displayed.

  In step S13-8, the input trajectory data rotation selection process is terminated. Release the secured work area and buffer area.

  By adding the processing means described above, handwriting data with an inclination close to an upright handwritten character stored in the dictionary is selected from handwriting trajectory data that has undergone rotational coordinate conversion at multiple angles. Can be realized, thereby shortening the recognition time and improving the recognition rate.

  In the above description, the upright state of the character is examined with the circumscribed rectangle of the trajectory data as a result of the rotation. However, another method is conceivable and will be described. It is another Example of the rotation data selection means of 11-1.

  FIG. 14 is a conceptual diagram showing the area of the surrounding space of the character trajectory in the circumscribed rectangle. 14A is an example of an upright state of rotated character trajectory data, and 14B is an example of tilted rotated character trajectory data. As can be seen from the example of FIG. 14, the area of the peripheral space of the character locus in the circumscribed rectangle is smaller in 14A than in 14B. As a result, the data in the upright state is selected by selecting data with a small area in the space around the character locus in the circumscribed rectangle.

  Referring to the example 14A in FIG. 14, a portion that touches the handwriting trajectory from the top left vertex of the circumscribed rectangle is detected, and a polygon composed of a plurality of contact points and sides of the circumscribed rectangle is represented by a diagram. -1 is obtained, and the area can be calculated by a general algorithm.

  Similarly, if a portion in contact with the handwriting trajectory is detected from the upper right vertex, 14A-2 is obtained. In this example, since the polygon from the lower left vertex is very small, the processing is omitted. Similarly, the processing at the lower right is omitted. The area of 14A becomes = area of 14A-1 + area of 14A-2.

  Next, an example written with an inclination of “three” will be described. 14B is the example, 14B-1 which detected the part which contact | connects a handwriting locus | trajectory from the upper left vertex of a circumscribed rectangle, and represented the polygon which consists of the some contact point and the circumscribed rectangle side in figure. And the area can be calculated by a general algorithm.

  A part that touches the handwriting trajectory is detected from the upper right vertex of the circumscribed rectangle, and 14B-2 that graphically represents a polygon composed of a plurality of contact points and sides of the circumscribed rectangle is obtained, and the area is calculated. .

  The part that touches the handwriting trajectory from the lower left vertex of the circumscribed rectangle is detected, and 14B-3 that graphically represents the polygon composed of the contact points and the sides of the circumscribed rectangle is obtained, and the area is calculated. .

  Detects the part that touches the handwriting trajectory from the lower right vertex of the circumscribed rectangle, and 14B-4 is obtained, which represents the polygon composed of the contact points and the sides of the circumscribed rectangle. To do.

  The area of 14B becomes = area of 14B-1 + area of 14B-2 + area of 14B-3 + area of 14B-4.

  Compare the area of 14A and the area of 14B, and select the smaller one.

  By processing in this way, more detailed rotation handwriting trajectory data can be selected.

[Third embodiment]
In the first and second embodiments, processing in units of one character has been described. However, there are actually implementation situations that may be better handled in units of character strings.

  A configuration is possible in which a recognized character string result selecting means to be processed in character string units is added.

  FIG. 15 is a processing block diagram of the third embodiment. Parts having the same part numbers as those in the process configuration diagram of FIG. 3 of the first embodiment are the same and will not be described.

  Reference numeral 15-1 denotes a character cutting unit, which cuts characters into a trajectory group for each character according to a character distribution by a general character frame or a trajectory sequence data group of a written character string according to a distribution status of the trajectory or a separation status in character units. Perform the separation process.

  15-2 is a recognition character string result selection means for correcting the result in character units by evaluating the recognition result for each character and the recognition angle in character string units.

  FIG. 16 shows an example of handwriting input in character string units.

  FIG. 17 is a recognition result example of the input example of FIG.

  FIG. 18 is a flowchart showing an input character string recognition process.

  Processing will be described with reference to the flowchart of FIG.

  The example in FIG. 16 assumes an example like a handwritten memo input application. In the case of a paper memo pad, it is natural to write characters with various inclinations as shown in FIG. However, in the conventional handwritten character input device, it is stipulated to write in a character frame such as manuscript paper or to write upright. In this embodiment, it is possible to enter characters at different angles, such as “Notepad” represented by 16-0, or 16-1, 16-2, 16-3, 16-4. , 16-5 may be written in one column with the same inclination, such as “Blue Crescent Moon”.

  In step S18-1, an input character string recognition process is started. Work area and buffer area are secured and initialized.

  In step S18-2, the trajectory data of the input character string is read. The handwritten character strings 16-1, 16-2, 16-3, 16-4, and 16-5 shown in FIG. 16 by the coordinate input unit 2-1 in FIG. The data string is read into the buffer area.

  In step S18-3, the character string data is divided into characters. This character cutting process itself may be a conventional algorithm described in JP-A-5-197840. A virtual frame for one character is determined from the XY coordinates of the data strings 16-1, 16-2, 16-3, 16-4, and 16-5, and the first character is 16- based on the virtual frame. 1 is cut out, and 16-2 is cut out from the next trajectory with a virtual frame. By doing so, the first character is 16-1, the first character is 16-2, the third character is 16-3, the fourth character is 16-4, and the fifth character is 16-5. become.

  In step S18-4, the character counter is initialized. Processing is performed from the first character with i = 1.

  In step S18-5, i-th multiple rotation angle conversion is performed. Since this process itself is the same as the process described in step S6-3, detailed description thereof is omitted. For simplicity, the rotation angle is set to 45 °, and data rotated to −45 °, 0 °, and 45 ° is created.

  In step S18-6, recognition processing is performed for each i-th character rotation angle conversion data. Since the matching process itself is the same as that described in steps S6-4 and S6-5, a detailed description thereof will be omitted. The feature point of each coordinate data of the handwritten character rotated at each angle in the above step is extracted, and the feature point stored in the dictionary is subjected to the matching process, and the dictionary data having the most similar shape is recognized as the recognition result. Output.

  In step S18-7, a process of storing recognition results in descending order of matching degree of the i-th character is performed. The recognition results are stored in a specific area of the RAM 2-7 in descending order of coincidence with the dictionary shape for each input character string order.

  In step S18-8, 1 is added to i to check the next character.

  In step S18-9, it is checked whether there is a next character. If there is a next character, the process proceeds to step S18-5, and if there is no next character, the process proceeds to step S18-10. If the i-th data does not exist in the character without the character string stored in step S18-3, all data has been processed, and the process proceeds to the next step. If there is unprocessed data, process that data.

  In step S18-10, it is checked whether the angle of the result of all characters is the same. If all characters are recognized at the same angle, the process proceeds to step S18-13, and the process is terminated. Otherwise, the process proceeds to step S18-11.

  In step S18-11, it is checked whether only one character has a different angle. If only one character has a different angle, the process proceeds to step S18-12. Otherwise, the process proceeds to step S18-13 and the process is terminated.

  If it is determined in steps S18-10 and S18-11 that three characters such as 16-0 are written at different angles, the input angle is used as it is. If the input data of “blue crescent moon” represented by 16-1 to 16-5 is blue: angle −45 °, yes: angle −45 ° mi: 0 °, day: −45 °, month: −45 ° In such a case, “Mi: 0 °” is detected and corrected. In the present embodiment, an example is described in which only one character in the character string is corrected when the angle is different. However, when the characters within 20% in the character string are different, the different characters may be corrected. good.

  In step S18-12, the result of the character at a different angle is corrected. For example, when the input data of “blue crescent moon” is blue: angle −45 °, yes: angle −45 ° mi: 0 °, day: −45 °, month: −45 °, mi: 0 ° is the other Because it is an angle different from the character, it becomes a correction target. The recognition result of the third character is changed to the recognition result having the highest similarity among the results of −45 ° among the recognition results stored in step S18-7.

  In step S18-13, the input character string recognition process is terminated. Release the secured work area and buffer area.

  By adding the processing procedure described above, a handwritten character recognition device that automatically corrects when a misrecognition result with a different inclination is mixed in the result of a handwritten letter written with a certain inclination. Can be provided.

It is a figure showing the actual usage pattern of this invention. It is a block diagram of the portable information device of the present invention. It is a process block diagram of an Example of this invention. It is an example of the input screen of the character of an Example of this invention. It is an example of the input screen of the character of an Example of this invention. It is a flowchart of 1 character recognition. It is an example of correction operation screen of a recognition result. It is an example of correction operation screen of a recognition result. It is a flowchart of the rotation angle correction process of a recognition result. It is an example of correction operation screen of a recognition result. It is a processing block diagram of 2nd Example. It is a conceptual diagram of the rotation data of the input character of 2nd Example. It is a flowchart of 2nd Example. It is a conceptual diagram of the input character rotation data of 2nd Example. It is a processing block diagram of 3rd Example. It is an example of the display screen of the input character string of 3rd Example. It is an example of the display screen of the input character string of 3rd Example. It is a flowchart of 3rd Example.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Information terminal 2 Image display part 2-1 Coordinate input part 2-2 Image display part 3 LAN cable 4 Wireless antenna 5 Personal computer 6 CPU
7 RAM
8 ROM
9 HD
10 LAN interface 11 System bus 12 Communication card

Claims (7)

  Handwritten character input means, input trajectory coordinate rotation means for each character, recognition means for recognizing a plurality of input trajectory coordinate data rotated at a plurality of angles by the coordinate rotation means, and selecting an optimum candidate from the result of the recognition means A handwritten character recognition apparatus characterized by comprising selection means for performing   A handwriting character input means, an input locus coordinate rotation means, a selection means for selecting locus coordinate data of an upright image of a character from a plurality of input locus coordinate data rotated at a plurality of angles by the coordinate rotation means; A handwritten character recognizing device provided with a recognizing means for recognizing locus coordinate data.   Means for rotating and displaying the character font of the recognition result according to the angle of the recognition result recognized by the character recognition device; user instruction changing means when the display recognition angle is different from the user intention; and an input locus of the indicated angle 3. A handwritten character recognition apparatus according to claim 1, further comprising a conversion means for converting the coordinates of the character and a re-recognition means for re-recognizing the locus data after the coordinate conversion.   A handwriting character string input means, a character cutting means for dividing the inputted handwriting character string locus data into characters, a plurality of input locus coordinate rotation means for each character and a plurality of rotations rotated at a plurality of angles by the coordinate rotation means. The recognition means for recognizing input trajectory coordinate data, and the angle of the character string when it is determined that the character recognition result angle information of a part of the recognition result character string of the recognition means is an angle different from the angle of the entire character string A handwritten character recognizing device, characterized in that a correcting means for correcting in combination is provided.   A handwritten character input step, an input trajectory coordinate rotation step for each character, a recognition step for recognizing a plurality of input trajectory coordinate data rotated at a plurality of angles by the coordinate rotation step, and selecting an optimum candidate from the result of the recognition step A method for recognizing handwritten characters, characterized in that a selection step is provided.   A selection step for selecting trajectory coordinate data of an upright image of a character from a plurality of input trajectory coordinate data rotated at a plurality of angles by the handwriting character input step, the input trajectory coordinate rotation step, and the coordinate rotation step; A handwritten character recognition method comprising a recognition step for recognizing locus coordinate data.   A handwriting character string input step, a character cutting step for dividing the input handwriting character string trajectory data for each character, an input trajectory coordinate rotation step for each character, and a plurality of rotations rotated at a plurality of angles by the coordinate rotation step; The recognition step for recognizing the input trajectory coordinate data, and the angle of the character string when it is determined that a part of the character recognition result angle information in the recognition result character string of the recognition means is different from the angle of the entire character string A method for recognizing handwritten characters, characterized by comprising a step of correcting the combined information.

Images