JP3925247B2 - Online handwritten character recognition device without real time frame and program for realizing the same - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a handwritten character recognition device that recognizes a handwritten character written with a stylus pen or the like on a handwritten character input device such as a tablet, converts it into a character code, and uses it in a program on a computer.
[0002]
In particular, it is possible to write a character string at any position on the writing surface without using a character area (for example, a character frame) for writing each character. About.
[0003]
[Prior art]
The handwritten character recognition device is roughly classified into a handwritten character recognition device with a frame having a character input area for each character and a handwritten character recognition device without a frame without a character input region for each character. In this specification, the “character input area for each character” is an area in which handwritten characters are input for each character, separated by a delimiter such as a character frame, vertical line, underline or comma, The color may be changed every time, and there is no particular limitation on how to divide the area for each character. Many conventional on-line handwritten character recognition devices use a character input area for each character that does not require the determination of character separation. The reason is that if there is no character input area for each character, it is necessary to separate the handwritten character string input before character recognition into character units, but it is difficult to separate characters accurately and quickly. is there. However, if there is a character input area for each character, it is necessary to write characters in the character input area by separating each character, and this will impair natural handwriting input. In addition, the development of a frameless handwritten character recognition technology that allows a character string to be freely written as naturally as possible without using a character input area for each character even on the screen.
[0004]
When there is no handwritten character input area for each character, for example, as shown in FIG. 17, there is almost no character spacing between “woman” and “child” even though the intention is to input handwritten as “female college student”. If the character separation position is not determined correctly, the character recognition itself will be incorrect. Cutting out is a big problem.
[0005]
As an example of the frameless character recognition method currently proposed, a method for collectively recognizing the entire input character string pattern will be described below.
[0006]
Senda et al.'S paper “Unframed online string recognition method that integrates segmentation, recognition, and language certainty” (Science Technical Report, PRMU98-138, 1998.12) ”and Fukushima et al. "None handwritten character string recognition (Science Technical Report, PRMU98-139, 1999.12)", the entire input character string pattern is collectively divided into individual segments that make up each character, and these are combined. A method has been proposed in which character recognition is performed for each character region, and the combination having the highest likelihood among the possible combinations of character regions is used as a character string recognition result (hereinafter referred to as Conventional Example 1). A segment is composed of one or a plurality of strokes and is a constituent element of a character, and one character is composed of one or a plurality of segments.
[0007]
The method of determining the character area by combining the segments used in Conventional Example 1 is equivalent to the method of configuring the individual segments in a network shape and obtaining the optimum path from this network, as shown in FIG. A square circumscribing the handwritten character in FIG. 18 is a portion determined as a segment, a dotted arrow indicates a path between segments, and a thick arrow indicates an optimal path, that is, the start of the next character.
[0008]
FIG. 19 shows a configuration diagram of a frameless character recognition method for collectively recognizing the entire input character string pattern in this way. First, a character string pattern is input from the character string pattern input unit 30, the character string pattern input by the segment determination unit 31 is divided into segments, and then the character region candidates are combined by the character region determination unit 32. By creating a link between adjacent character areas, a network structure as shown in FIG. 18 can be generated. Finally, the individual character recognition unit 33 performs character recognition on each character region to create recognition candidates, and the optimum path generation unit 34 determines the character region by obtaining the optimum path from the network, and determines the determined character. Based on the area, a character recognition result character string of each character area is obtained.
[0009]
When the above method is used, the character is cut out and recognized after all the character strings have been written. Therefore, when the user writes a long character string, there is a time lag that cannot be ignored from the end of writing until the recognition result is obtained. Resulting in reduced usability for users
Therefore, as a method to reduce the time lag at the time of recognition, the input character string is divided into line units, or whether the pen movement amount (off stroke) at the time of pen-up is longer than a certain amount in the direction opposite to the writing direction By detecting the end of the line, the method of starting the frameless character recognition process at that time or when the pen-up time exceeds a certain value or the aspect ratio of the written text area exceeds a certain value There has been proposed a method of starting recognition when it becomes. According to these methods, even when a long character string is written, the recognition process up to that point is performed when a certain length is written, so the time for which the user waits for the recognition result (time lag) should be kept within a certain range. Can do. Recognize character strings written so far.
[0010]
As described above, when a certain amount is written, a recognition process is performed for each character pattern string input so far, and a character recognition result as an intermediate result is obtained during writing. However, since these methods start recognition processing in units of lines or several characters, the time lag until a recognition result is still obtained is much longer than character recognition with a frame. The impression of frameless recognition is very bad for users who have used, and this is the main cause of the lack of widespread use of frameless character recognition devices.
[0011]
Therefore, realization of on-line handwritten character recognition without a real time frame is required in which written characters are recognized with almost no waiting time and the recognition result is immediately presented to the user. Next, a conventional example for realizing online handwritten character recognition without a real time frame with a small time lag will be described.
[0012]
In order to realize real time frameless recognition, it is necessary to cut out characters simultaneously with writing. For a method of extracting unframed input characters in real time, see Aizawa et al.'S paper, “Real-time character extraction from handwritten character strings using multiple stroke features (IEICE Technical Report, PRU95-185, pp57-62). 1995-12) ”(hereinafter referred to as Conventional Example 2). In this method, a plurality of features such as stroke intervals and off-stroke directions are used in combination, and each time one stroke is input, it is determined whether or not it is the first stroke of the character.
[0013]
FIG. 20 shows a configuration diagram of real time frameless recognition using this method. The operation flowchart is shown in FIG. Hereinafter, the operation of Conventional Example 2 will be described with reference to these drawings.
[0014]
First, handwriting data of characters written by the user is input for each stroke from the stroke input unit 40 (step S40). In the input stroke, the character boundary determination unit 41 determines whether or not the stroke is the head stroke of the character (step S41). Specifically, the character boundary determination unit 41 determines any one of the three strokes “the beginning of the character”, “not the beginning of the character”, “possibly the beginning of the character but not fixed” To do. Here, even if the input stroke is the beginning of a line, it is regarded as the beginning of a character. Here, if the stroke is not the beginning of the character (step S41: no), or there is a possibility that it is the beginning of the character but it is not fixed (step S41: undecided), the stroke data is added to the temporary stroke storage unit 44. If the stroke is the head of the character (step S41: yes), the character string pattern generation unit 42 extracts the stroke data group already stored in the stroke temporary storage unit 44, A character string pattern is generated (step S43), and the frameless character recognition unit 43 performs a frameless handwritten character recognition process on the character string pattern. (Step S44). The recognized stroke data is cleared from the temporary stroke storage unit 44 (step S45), and the stroke data regarded as the head of the character is stored in the temporary stroke storage unit 44 (step S42). Then, when a predetermined time t1 (normally 0.5 to 2.0 seconds) elapses or a time-out occurs or a recognition button provided on the handwritten character input device is pressed (step S45: YES), a character string pattern is generated. The unit 42 extracts the stroke data group already stored in the temporary stroke storage unit 44, generates a character string pattern (step S46), and the frameless character recognition unit 43 recognizes the character string pattern without framed handwritten characters. Processing is performed (step S47), and the recognized stroke data is cleared from the temporary stroke storage unit 44 (step S48). When a predetermined time t2 (usually 3 to 5 seconds) elapses and a time-out occurs or a confirmation button provided on the handwritten character input device is pressed (step S49: YES), the recognition result is sent to an application or the like (step S49). S50), the process is terminated. In other cases (step S49: no), the process returns to step S40. The recognition method in the frameless character recognition unit 43 can clearly determine whether or not the input stroke is the beginning of the character, and there must be a stroke that may be the beginning of the character but is not fixed. For example, since stroke data for one character is stored in the stroke data stored in the temporary stroke storage unit 44, it is possible to use the frame recognition method. Since it may not be possible to clearly determine whether the character is at the beginning, stroke data for a plurality of characters may be stored. Therefore, it is necessary to use a frameless recognition method so that it can cope with either case.
[0015]
FIG. 22 shows a success example and a failure example of character segmentation according to Conventional Example 2. In the figure, a bold line is a stroke determined to be the beginning of a character, a dotted line is a stroke that is not the beginning of the character, and a thin solid line is a stroke that can be the beginning of the character but is not definite. In the example of (a), all bold strokes are at the beginning of the character, and the determination result is correct. However, in the example of (b), there are two thick lines in the character “sen”. This is because at the time when the vertical line of the 3rd stroke of "Sen" was written, this stroke was determined to be at the beginning of the character with a sufficient distance from the previous 2 strokes. When a stroke is written, the interval between the previous two strokes is narrowed, and the condition that can be determined at the beginning of the character is lost. In the conventional example, this cannot be avoided.
[0016]
The above errors are caused by trying to determine the boundary position of the character, although the boundary position of the character cannot be determined at the time of writing the stroke. That is, the method of determining whether or not the beginning of the character is input every time one stroke is input as in the conventional example cannot accurately determine the character boundary, so that accurate character recognition without a real time frame cannot be realized. That's it.
[0017]
[Problems to be solved by the invention]
As described above, in the batch type frameless character recognition as in the conventional example 1, since character recognition is performed collectively, there is no real time frame in which recognition takes time and recognition processing is performed in parallel with writing. Character recognition cannot be realized.
[0018]
On the other hand, although the real-time recognition is realized in Conventional Example 2, the character boundary position of the input pattern is determined every time a stroke is input. May be changed to the determination that the stroke is not the beginning of the character, and in that case, if frameless character recognition is performed with the stroke group before the stroke determined to be the beginning of the character, Unrecognizable character without frame. That is, since it cannot be accurately determined whether or not it is the beginning of a character with only one stroke, there is a problem in that the accuracy of character extraction is reduced and accurate character recognition cannot be performed.
[0019]
[Means for Solving the Problems]
Briefly, the present invention performs character recognition processing in parallel with writing in order to minimize the time lag as in the conventional example 1, and at the same time, causes a character cut-out error as in the conventional example 2. It is difficult to achieve accurate real-time frameless character recognition that is easy to correct even if an error occurs.
[0020]
That is, the online handwritten character recognition device without a real time frame according to the present invention inputs a stroke of a handwritten character and converts it into stroke data; Said Based on stroke data and existing indoubt segments, Said A segment determination unit that determines whether stroke data is a new unconfirmed segment or an addition to an existing unconfirmed segment, and based on the segment, determines whether the segment is a deterministic segment or an unconfirmed segment, and confirms the segment. A segment determination unit to perform, a character region determination unit for determining a character region for each character composed of a combination of determined segments based on the determined segment, and for each character region based on the stroke data included in each character region An individual character recognition unit that recognizes the handwritten character online and generates a character candidate, and an optimal path generation unit that generates an optimal path of a combination of character candidates for each character region and outputs an optimal character combination as a recognition result When the stroke data is input to a confirmed segment, it is determined whether or not the stroke data is a deletion instruction. In the case of a deletion instruction, the determined segment to which the stroke data is input is deleted and thereafter If the confirmed segment is in an unconfirmed state and there is no deletion instruction, a segment confirmed canceling unit that places the confirmed segment after the confirmed segment in which the stroke data is input in an unconfirmed state, and a segment processed by the segment confirmed canceling unit On the other hand, when the character area canceling unit cancels the determination of the character region, the optimum path canceling unit canceling the optimal path for the segment processed by the segment confirmation canceling unit, and when the stroke data is a deletion instruction, Returned to the unconfirmed state by the segment confirmation cancellation unit for the segment determination unit A segment confirmation instructing section for instructing to redo the determination of the segment, and a correction start of the recognition result character string after the character modified by the cancellation of the segment confirmation and the number of characters in the output recognition result character string. Recognition result information output that outputs correction information consisting of position information Comprising When the stroke data is input to a confirmed segment, the segment confirmation cancellation unit, character area cancellation unit, optimum path cancellation unit, and segment determination instruction unit are processed. If the stroke data is not input to the confirmed segment, the correction information for the output recognition result character string is output to the linked application. Every time one stroke is input, Execute the process of the segment determination unit, segment determination unit, character region determination unit, individual character recognition unit and optimal path generation unit, Character recognition is performed on a segment portion for which input has been confirmed.
[0021]
With this configuration, the real-time frameless online recognition apparatus according to the present invention can obtain a recognition result up to a portion that is a predetermined distance or more away from the writing position in real time with high accuracy, and the user can perform frameless character recognition. It is possible to eliminate the need to worry about the waiting time.
[0022]
Also, With this configuration, The online recognition device without a real time frame according to the present invention is: Handwritten characters can be corrected, and even if the recognition / confirmation process is wrong, it is possible to redo only the wrong part, and even if the deviation between the writing position and the recognition position is reduced (= threshold value d described later is reduced). No problem will occur, and it will be possible to avoid performance degradation due to errors in segment confirmation processing. The
[0023]
Also, With such a configuration, the online recognition device without a real time frame according to the present invention is Even if a handwritten character that has already been confirmed is deleted or a stroke is added to the handwritten character, the segment after the corrected character becomes unconfirmed, without waiting for the next stroke to be input. Perform recognition processing for definite segments It becomes possible.
[0024]
Also, With this configuration, An online handwritten character recognition device without a real time frame according to the present invention is: Even if the recognition result character string is modified, it is possible to notify the linked application of the modified content. The "
[0025]
Also , Online handwritten character recognition without real time frame according to the present invention The program recognizes online handwritten characters without real time frame, means for inputting strokes of handwritten characters and converting them into stroke data, the stroke data based on the stroke data and existing undefined segments, Segment means for determining whether a new unconfirmed segment or an addition to an existing unconfirmed segment, segment means for determining whether or not the segment is a definite segment or an indeterminate segment based on the segment, and confirming the segment Based on the segment, character area determination means for determining a character area for each character consisting of a combination of confirmed segments, on the basis of the stroke data included in each character area, the handwritten characters are recognized online for each character area, Individual generation of character candidates Character recognition means, optimal path generation means for generating an optimum path of a combination of character candidates for each character area, and outputting an optimum character combination as a recognition result, and when the stroke data is input to a definite segment, It is determined whether or not the stroke data is a delete instruction, and if it is a delete instruction, the determined segment to which the stroke data has been input is deleted and the subsequent determined segment is set to an unconfirmed state. Segment finalization canceling means for setting the final segment after the finalized segment to which data is input to an indeterminate state, character area canceling means for canceling the determination of the character area for the segment processed by the segment finalization canceling means, and the segment finalization For the segment processed by the canceling means, the best path to cancel A path cancellation means, a segment confirmation instruction means for instructing the segment determination means to redo the determination of the segment returned to the unconfirmed state by the segment confirmation cancellation means when the stroke data is a deletion instruction; and A recognition result information output means for outputting correction information including a recognition result character string after the character corrected by canceling the segment confirmation and correction start position information indicating from which character of the output recognition result character string has been corrected; It is characterized by functioning as The
[0026]
By reading the program having such a configuration into a computer and executing it, Online handwritten character recognition device without real time frame In real-time online frames, the recognition result up to a part more than a predetermined distance from the writing position can be obtained with high accuracy in real time, and the user does not have to worry about the waiting time for frameless character recognition None Handwritten character input device can be realized The
[0027]
Also, Such configuration By loading and running the program on your computer, Online recognition apparatus without real time frame according to the present invention In this case, it is possible to correct handwritten characters, and even if the recognition / confirmation process is wrong, it is possible to redo only the wrong part, and even if the deviation between the writing position and the recognition position is reduced (= threshold d described later). This will not cause any problems, and a real-time online frameless handwritten character input device that can avoid performance degradation due to an error in segment determination processing can be realized. The
[0028]
Moreover, the program according to the present invention is executed by causing the computer to read the program having such a configuration and execute the program. Online recognition device without real time frame Even in the case of deleting the already determined handwritten character or adding a stroke to the handwritten character, even if the segment after the corrected character is unconfirmed, without waiting for the next stroke to be input, Real-time online frameless handwritten character input device capable of instant recognition processing for unconfirmed segments can be realized. The
[0029]
Moreover, the program according to the present invention is executed by causing the computer to read the program having such a configuration and execute the program. Online handwritten character recognition device without real time frame Even if the recognition result character string is modified, it is possible to notify the linked application of the modified content. Real-time online frameless handwritten character input device can be realized. "
[0030]
DETAILED DESCRIPTION OF THE INVENTION
The principle of the basic embodiment of the present invention will be described below.
[0031]
First, it will be described how the present invention solves the decrease in accuracy in Conventional Example 2.
[0032]
In Conventional Example 2, the time lag in Conventional Example 1 is eliminated, but the character boundary cannot be accurately determined because the character boundary is determined for each stroke. In order to accurately determine whether or not each stroke is at the boundary position of the character, it is necessary to make a determination including a stroke existing around the input stroke. For example, as shown in FIG. 23, the character “sen” is a combination of “people” and “mountain”, and the strokes constituting each part are located in the vicinity and can be grouped. This group is a part that constitutes a character and does not extend over a plurality of characters. Such a group that forms a part of a character is called a “segment”. In batch type frameless character recognition as in Conventional Example 1, the input character string pattern is divided into segments and then the character boundary is determined, so the time lag until recognition is long, but the character boundary is It does not fluctuate. In contrast, Conventional Example 2 determines the character boundaries in units of strokes, so the time lag until recognition is short. There was no problem.
[0033]
In order to solve this problem, in the present invention, instead of determining a character boundary every time a stroke is input, each time a stroke is input, first, a group of strokes input before that is also included. The segment is determined and segmented, and then the character boundary is determined. Here, the problem is that at the time when the input stroke group is divided into segments, it is undecided whether or not each segment is the head of a character. That is, there is no fixed character boundary position by simply dividing the segment. For this reason, when segments are extracted, it is necessary to determine character boundaries by combining the segments. Here, it is referred to as character area determination in the sense of determining an area constituting one character.
[0034]
In order to determine the character area, it is necessary to determine the segments constituting the character area. For example, in the conventional example 2, when the character area or segment is determined, the first stroke constituting each area is used. However, in order to determine the character area, all strokes constituting the segment are necessary. It is not enough to know the leading stroke. Since the segment is extracted while the user is writing characters, even if one segment is extracted, there is a possibility that the segment may change when the user adds a stroke to the segment. Therefore, when the character area is determined, it is necessary that the segment constituting the character area is fixed, that is, the segment is not changed by a user input. That is, a segment determination determination unit that determines whether or not a segment has been determined is required in addition to a segment determination unit.
[0035]
When the character area is determined, a network structure (see FIG. 18) using only the currently determined segment can be created. Therefore, individual character recognition and optimum path calculation can be performed using this network, and a character recognition result as an intermediate result can be obtained while writing a character string.
[0036]
The above is the principle of the present invention, which is a method for accurately realizing character recognition without a real time frame. The points of the present invention can be roughly summarized as follows.
[0037]
While writing a character string, since the character recognition result including the latest stroke is unconfirmed, erroneous recognition will inevitably occur when the recognition result using all the written strokes is obtained. Therefore, in the present invention, the recognition result up to the part where the input is confirmed, slightly before the stroke being written, is calculated and output as an intermediate result. As a result, a recognition result up to a portion not far from the writing location is obtained in real time, and the user does not need to worry about the waiting time for frameless character recognition.
[0038]
(Embodiment 1)
An embodiment of the present invention will be described based on the configuration diagram of FIG.
[0039]
Normally, handwritten characters are input using a stylus pen on a tablet in a personal computer and on a transparent tablet integrated with a display screen in a PDA. In addition, input may be performed using a pointing device such as a mouse.
[0040]
First, the handwriting input by the user is input as continuous coordinate point sequence data every time one stroke is input by using a function of a general-purpose OS such as Windows (trademark of Microsoft Corporation). Thus, stroke data is generated by extracting coordinate points at predetermined time intervals from continuous coordinate point sequence data. The generated stroke data is normally non-continuous coordinate point sequence data. However, when echo back display is performed, these coordinate point sequences are connected by lines.
[0041]
Taking an example of character string pattern data of input handwritten characters expressed in C language, the data structure shown in FIG. 2 is formed as an array of stroke data and stored in the memory. nPoint indicates the data of the coordinate point sequence constituting the stroke data, pPoint indicates the coordinate point of the coordinate point sequence constituting the stroke data, and x and y are the x coordinates of the coordinate points of pPoint, respectively. , Y coordinate. nStroke indicates the number of stroke data of the character string pattern data, and pStroke indicates the stroke data.
[0042]
The input stroke data is transferred to the segment determination unit 2, and a segment including the stroke data is extracted. The segment extracted here is stored in the segment temporary storage unit 3 as an undetermined segment. The segment temporary storage unit 3 can be provided on a storage device such as a hard disk, but it is preferably provided on a memory because the capacity is small and speed is required.
[0043]
What is important in the segment determination is that an undefined segment already stored in the segment temporary storage unit 3 is affected by a new stroke input. Specifically, when a new stroke is input, the unconfirmed segment before the stroke input becomes a definite segment, the newly input stroke forms a new unconfirmed segment, or the newly input stroke is added to the unconfirmed segment. It is a point to do. In this way, each time a stroke is input, the input character string pattern data is divided into a definite segment that is used for character recognition at that time and an undefined segment that is not used for character recognition. It is.
[0044]
Taking the character “sen” as an example, FIG. 3 shows how the unconfirmed segment (segment surrounded by a bold circumscribing frame) changes when the fifth to fifth drawings of “sen” are input. . First, assume that two strokes have been input. When the third image is input, a new unconfirmed segment is created with the stroke of the third image, as shown in FIG. When the fourth image is input, the 2-stroke segment of the third image and the fourth image becomes a new unconfirmed segment. Similarly, when the fifth image is input, three strokes from the third image to the fifth image become a new unconfirmed segment. As described above, the unconfirmed segment changes each time a stroke is input.
[0045]
In the description of FIG. 3, it has been described how the unconfirmed segment changes when the character “sen” is written. When there are a plurality of unconfirmed segments, the segment determination unit 3 selects a plurality of unconfirmed segments. How to determine whether or not to combine will be described.
[0046]
The segment determination unit 2 creates an undetermined segment according to the procedure shown in the flowchart of FIG. FIG. 5 shows how the segment determination unit 1 actually registers an undetermined segment according to FIG.
[0047]
First, the operation of the segment determination unit 2 will be described using the flowchart of FIG.
[0048]
First, a stroke is input from the stroke input unit 1 (step S1). Next, only one input stroke is registered as a new segment (step S2), and an index is set for the final unconfirmed segment of the unconfirmed segment group composed of already input strokes (step S3). ). If the index position is not empty, that is, if an undefined segment exists before the new segment (step S4: no), information on the undefined segment at the position where the index is set is acquired (step S6). . Then, the position information (x coordinate in the present embodiment) of the new segment and the acquired uncertain segment is compared (step S7), and it is determined whether or not both segments can be connected (step S8). If both segments are not connectable (step S8: no), the index is moved forward (step S9), and the process returns to step S4. If both segments can be connected (step S8: yes), both segments are connected to form a new segment (step S10), the connected unconfirmed segment is deleted (step S11), and the index is determined as an unconfirmed segment. It resets to the end (step S12) and returns to step S4. When these processes are repeated and the index position becomes empty, that is, when there is no unconfirmed segment before the new segment (step S4: yes), the new segment is registered in the segment temporary storage unit 3 as an unconfirmed segment. (Step S5).
[0049]
A specific example of registering a new unconfirmed segment by connecting a new segment and an unconfirmed segment will be described with reference to FIG. In FIG. 5, it is assumed that a dotted line frame represents an undetermined segment, a solid line frame represents a new segment, and a thick arrow represents an index indicating the previous undetermined segment.
[0050]
In this embodiment, a segment that is at least two character sizes to the left of the current input position is set as a confirmed segment.
[0051]
First, (a) of FIG. 5 is a segment division state at the time of writing up to the third stroke of “No” of “No frame”, and each segment has an interval in the x-coordinate direction. Divided into unconfirmed segments.
[0052]
When the next stroke (the fourth stroke of “NA”) is input in the initial state of (a), the fourth stroke of “NA” becomes a new segment, as shown in (b), as shown in (c). Thus, the index is set at the end of the unconfirmed segment (here, the unconfirmed segment of the third stroke of “NA”). Subsequently, the x-coordinate data of the indeterminate segment at the index position and the new segment are compared. If the x-coordinate data is at a position that satisfies the predetermined criteria, both segments are combined to create a new new segment as shown in (d). It is done. Next, as shown in (e), the index is reset to the previous unconfirmed segment. Similarly, as shown in (f) and (g), segment combination and index resetting are performed. In the state of (f), since there is a clear interval between the new segment (“NA”) and the unconfirmed segment (“）”), as shown in FIG. Therefore, the index moves further forward. There is a clear interval between the segment pointed to by the index (“卆”) and the previous segment (“Tree”), and as shown in (h), the index is undecided by the previous one. Reset to segment ("tree"). Since there is no unconfirmed segment in front of the “tree” segment, as shown in (i), only the index is moved forward and finally it reaches before the first segment and points to nothing ( = Index position is empty), a new segment is registered as a new unconfirmed segment. Further, when there is a segment before, the determination of the unconfirmed segment is performed up to the confirmed segment (in this example, the segment within two characters from the latest segment).
[0053]
Here, an example of a criterion for determining whether or not to connect an indeterminate segment and a new segment is given. As one example of implementation, here, “(A) two segments in the x-direction region overlap each other by a certain value” or “(B) one segment in the x-direction is the other in the x-direction. The segment is connected when either of the following is true. Specifically, the maximum value and minimum value of the x coordinate of the coordinate point of one or more stroke data included in the segment are the maximum value and minimum value of the x coordinate of the segment, and the x coordinate of the new segment and the undefined segment It is recommended to compare the maximum value and the minimum value of these to check whether the above-mentioned conditions are satisfied. This is a criterion for horizontal writing. In the case of vertical writing, the determination is made in the y-direction area. Note that the criterion for determining segment connection is not particularly limited to this.
[0054]
In the segment temporary storage unit 3, an undefined segment array and a confirmed segment array are stored separately. The new undefined segment and the definite segment are added to the end of the undefined segment array and the definite segment array, respectively. FIG. 6 shows an example of the data structure of the segment information using C language. nStroke is the number of strokes constituting the segment, and pStrIdx is an array of indexes indicating the strokes constituting the segment. The index is an integer starting from 1, indicating the number of strokes in the input pattern. top and btm indicate the coordinates of the upper left and lower right vertices of the circumscribed frame of the segment. These coordinates are used to check the positional relationship of the segments, such as when the segment determination unit determines the overlap of the x region of the undefined segment and the new segment. The segment temporary storage unit 3 may store the unconfirmed segment and the deterministic segment in different areas or files on the memory, or the undetermined segment and the deterministic segment can be identified so that the same file The storage method is not particularly limited.
[0055]
After performing the segment determination, the segment determination determination unit 5 determines which of the unconfirmed segments stored in the segment temporary storage unit 3 can be determined. Segment determination is a segment that can be regarded as not deformed even when a new stroke is input. For example, in the case of a horizontally written character string, characters are written in order from left to right (from top to bottom in vertical writing), so new strokes are written to the left (up in vertical writing) beyond the input strokes. It can be said that this is a phenomenon only within one character. Therefore, even if the size of the handwritten character varies to some extent and the character spacing is taken into account, the segment that is to the left of the character width maximum value w or more than the stroke currently being written (above for vertical writing) is fixed. You can say that. Therefore, the segment determination determination unit 5 uses such a determination criterion to set a segment that is left to the left by a certain threshold d or more than the segment including the stroke being written. Good. In other words, in the case of horizontal writing, “determine the segment to the left of one character size from the current input position”, in the case of vertical writing, “from one character size from the current input position. In other words, the determination is made based on a predetermined determination criterion such as “determine the upper segment”.
[0056]
An example of the operation of the segment determination determination unit 5 by the above method is shown in FIG. In FIG. 7 (a), up to “frame” has been written, and the dotted character after that is an unfilled character. Here, all the segments indicated by dotted lines indicate that they are all undetermined segments. (B) shows a state where the following “shi” is entered. Here, the segment “tree” located at the left of the threshold “d” or more from the left end of the “shi” segment is in a definite state. Similarly, at the time when the “verification” bias of “approval” is input, the “卆” segment is determined because it is separated by more than the threshold value d, and the “na” segment is not determined because it is not separated by more than the threshold value d, At the time when “Ninja” of “approval” is inputted, the “na” segment is determined because it is more than the threshold value d. In this way, as writing progresses, the segment located at the left of the latest stroke by more than the threshold value d is determined. For the information of the segment that has become a new confirmed segment from the unconfirmed segment, copy the information of the corresponding unconfirmed segment in the segment temporary storage unit 3, add it to the list of confirmed segments, and delete the information of the corresponding unconfirmed segment To do so. Here, the threshold value d may have an appropriate value in advance, for example, the maximum value w of the character width or the character size as a predetermined value, and if the handwritten character input is a single line input, the height of the frame of the single line input There is no particular limitation. Further, the threshold value d may be obtained as, for example, the maximum value w × n of the character width, instead of using the maximum value w of the character width, the character size, or the height of the frame for one line input. Usually, n is in the range of 1.0 to 2.0, but is not particularly limited.
[0057]
The character area determination unit 4 creates a set of confirmed segments that can be one character from a series of confirmed segments as a character area candidate, and creates a network in which individual character areas are connected by links. Then, a character region is determined by combining a plurality of adjacent segments. FIG. 8 is an example of a combination of segments of the “frame” portion when handwritten as “frameless recognition”. Individual character areas are linked between those that may be adjacent. FIG. 9 shows, by arrows, links that can be combined when handwritten as “no frame”. For example, since the first “tree” and “卆” can be connected, a link from “卆” to “tree” is set, but “tree” and “na” have “卆” between them. Therefore, since the character area is not adjacent, it cannot be connected and the link is not established.
[0058]
A lot of combinations are possible only by combining adjacent segments, but the combination is quite limited because it must be a plausible combination as one character. In FIG. 9, the criterion of combining only segments whose width in the x direction of the character area is 1.5 times or less of the standard character size w is used, but it is not particularly limited to this number.
[0059]
An example in which the data structure of the character area is defined is shown in FIG. Let nSegment and pSegIdx be the number and index of the definite segments that make up the character area, respectively. A serial number is assigned to the character area in order from 1, and the linked character area is indicated by nLink and pLink. In nCand and pCand, the result of the character individually recognized for each character area is stored. This is used in the individual character recognition unit 6 described below.
[0060]
The individual character recognition unit 6 performs online handwritten character recognition for each character area based on the stroke data included in the segments constituting each character area, and obtains one or more character candidates as a character recognition result. A character code / recognition score pair list is obtained for each character candidate for each character region, and stored in code and score, which are character region structures in FIG. The method for recognizing individual characters is not particularly limited as long as it is online recognition, but may be hybrid recognition combining on-line recognition and offline recognition.
[0061]
The optimal path generation unit 7 obtains a transition path with the highest cumulative score from the network using DP matching, and obtains a character recognition result from recognition candidate characters on the path. The character recognition result is normally notified to an application such as a word processor, and the application displays the recognition result on the screen or uses it for some other processing. DP matching is a method used in the literature of Conventional Example 1, and since it is a known method, description thereof is omitted.
[0062]
With such a configuration, in the frameless online recognition device, recognition results up to a portion that is more than a predetermined distance from the writing position can be obtained with high accuracy in real time, and the user cares about the waiting time for frameless character recognition. It is possible to eliminate the need to do this.
[0063]
(Embodiment 2)
Next, based on the configuration diagram of FIG. 11 and the flowchart of FIG. 12, an embodiment will be described in which a confirmed segment is once returned to an unconfirmed state by deleting or adding handwritten characters.
[0064]
In this embodiment, in the case where the confirmed segment changes due to the deletion or addition of handwritten characters, the configuration corresponding to the first embodiment deletes the segment corresponding to the deleted character portion. When the confirmed segment after that is returned to the unconfirmed state and added, the segment after the confirmed confirmed segment is returned to the unconfirmed state, and the character area and optimum path of the segment returned to the unconfirmed segment are returned. Segment cancellation / cancellation means for canceling, and segment determination processing means for determining whether or not a segment returned to an indeterminate state can be determined and determining a determinable segment are provided.
[0065]
This is mainly aimed at two effects. First, when there is an error in the determination result of the segment determination determination unit 5, for example, when a newly input stroke reaches one fixed segment that exceeds the threshold value d, the determination is determined. This is to cancel and start over. As described in the first embodiment, the segment once determined is used to execute the character recognition process. For this reason, when the determined segment is simply re-registered in the segment temporary storage unit 3, the processing is not restarted. In order to cancel the confirmation of the segment, the character recognition result obtained by the erroneously determined segment must be recovered. This embodiment provides means for that purpose.
[0066]
The second case is a case where a stroke already input by the user is canceled. If the user accidentally writes a character and deletes it including the part where the recognition result has already been obtained, or forgets to write a part of the character (for example, a small dot) and there is a confirmed segment later For example, when a region is written, it is necessary to correct the confirmed segment, that is, move the confirmed segment to the unconfirmed state and transform the unconfirmed segment.
[0067]
In FIG. 11, a portion indicated by a dotted line is a portion added to the first embodiment. Configuration of Embodiment 1 Compared to FIG. 1, as a segment canceling means, a segment fixed cancellation unit 8 that cancels a confirmed segment stored in the segment temporary storage unit 3 and sets it as an unconfirmed segment, and the confirmed segment becomes an unconfirmed segment. Thus, the character area canceling unit 9 for canceling the information related to the character area using the segment (held in the character area determining unit 4), and the information regarding the optimal path using the segment (optimal path generating unit) 7 is added to the optimum path canceling unit 10. Where to determine whether or not a stroke has been input to the confirmed segment is determined by the stroke input unit 1. When a stroke is input to the confirmed segment, the process is transferred to the segment finalization canceling unit 8. The control may be transferred to the segment determination unit 2, or the determination may be performed by the segment determination unit 2, and the process may be transferred to the segment confirmation cancellation unit 8 when a stroke is input to the determined segment. Immediately after the processing of the stroke input unit 1, the control is transferred to the segment confirmation / cancellation unit 8, and the segment confirmation / cancellation unit 8 determines whether or not a stroke is input to the confirmed segment, and the stroke is input to the confirmed segment. Then, the segment confirmation cancellation process is performed, and in other cases, the process may be moved to the segment determination unit 2. The character area canceling unit 9 and the optimum path canceling unit 10 are activated in response to the processing of the segment confirmation canceling unit 8, respectively.
[0068]
The flow of the segment confirmation / cancellation process when the handwritten character is deleted will be described with reference to the flowchart of FIG. The processing is started in response to an instruction to delete handwritten characters. The method for instructing deletion is a case where a specific gesture is written on a handwritten character (for example, a horizontal line is drawn on a confirmed handwritten character), but the method is not particularly limited.
[0069]
First, it is determined whether or not the input stroke is a deletion instruction (step S20). In the case of a deletion instruction (step S20: yes), the confirmed segment corresponding to the handwritten character to be deleted is deleted from the segment temporary storage unit 3 (step S21). The confirmed segment after the deleted segment needs to be returned to the unconfirmed segment in order to review the character area and the optimum path. Even if it is not a deletion instruction (step S20: no), it is necessary to return the confirmed segment after the confirmed segment to which the stroke is input to the unconfirmed segment in order to review the character area and the optimum path. Therefore, the segment to be returned from the confirmed segment to the unconfirmed segment is copied from the confirmed segment list in the segment temporary storage unit 3 and added to the new unconfirmed segment list (step S22), and the confirmed segment in the segment temporary storage unit 3 Are deleted from the list (step S23). By doing so, the confirmed segment to be processed returns to the unconfirmed segment. Next, since it is necessary to review the character area and the optimum path, the character area of the deleted confirmed segment (including the segment returned to the unconfirmed segment) is deleted from the character area held in the character area determination unit 4. (Step S24), the link to the deleted character area is deleted from the link held in the optimum path generation unit 7 (Step S25).
[0070]
Next, FIG. 13 shows a specific operation example when the recognized part is deleted. FIG. 13A shows a state immediately after writing “frameless recognition”. As for the character of “K”, it is just divided into segments because it is immediately after input, and the character area is not determined. That is, the last two segments are in an indeterminate state, which is indicated by a dotted frame. Moreover, the recognition result character string output to the user is shown on the right side of the network diagram. At the time point (a), segments up to “no frame recognition” are determined, character area candidates up to that are determined, an optimum path is calculated, and a recognition result character string is obtained. In the network, characters of individual character recognition results for each character area are displayed, and characters on the optimum path are surrounded by squares.
[0071]
Here, consider a case where the handwriting of the recognized character “shi” is deleted. The state immediately after being deleted is (b). Here, the segment part that constitutes the “recognition” after the deleted “Shi” has already been determined including the positional relationship with the “Shi” pattern, so “Shi” is deleted. Once done, it is necessary to return to the indeterminate state. Therefore, (c) is a state in which “SHI” and after are returned to undetermined. Here, not only “word” and “shinobi”, which are segments constituting “approval”, are unconfirmed, but also the network path and the recognition result character string are deleted. This is because the character area / optimum path after “shi” is deleted by the character area cancellation unit 9 and the optimum path cancellation unit 10. As a result, the recognition result character string is left with only “frame”.
[0072]
As described above, when the handwriting is deleted, the character recognition result related to the handwriting and the confirmed segment, the character area, and the link of the recognized pattern behind it are restored to the original state.
[0073]
The only problem in realizing this embodiment is the process of adding the unconfirmed segment created by copying the confirmed segment in step S22 of the flowchart to the list. This is a process of adding an undetermined segment to the segment temporary storage unit 3. In the description of the first embodiment, when adding an indeterminate segment, it was written that it should be added to the end of the list, but this is because handwritten characters are written in order from right to left (from top to bottom in the case of vertical writing). In this embodiment, when a confirmed character is corrected, unconfirmed segments are not necessarily registered in the order of writing, and may have to be added in the middle of the list. Therefore, when an unconfirmed segment is added to the list, the unconfirmed segments in the list may be arranged so as to be sorted in the ascending order of the x coordinate at the right end of the segment (in the descending order of the y coordinate in the case of vertical writing). As a result, the same processing can be performed when an unconfirmed segment newly generated from a stroke is added or when an unconfirmed segment is added in association with the cancellation of the confirmed segment.
[0074]
The handwritten character recognition apparatus without a real time frame according to the first embodiment recognizes up to a portion slightly before the currently written portion and outputs the result. Therefore, in order not to determine the segment by mistake, it is necessary to determine the segment with a sufficient margin (= threshold value d larger) than the writing position. That is, it is necessary to make a large difference between the writing position and the recognition position.
[0075]
However, by adding the recognition / determination process redo function according to the second embodiment, it is not fatal if the mistaken decision is made, and only the wrong part can be redone. Even if the deviation is reduced (= threshold d is reduced), no problem occurs. That is, in the first embodiment, the threshold value d is normally set to the maximum character width w × n (n is 1.0 to 2.0). However, n can be less than 1.0, and the time lag is further reduced. It can be reduced. Further, according to the second embodiment, performance degradation due to an error in segment determination processing can be avoided.
[0076]
(Embodiment 3)
In the third embodiment, immediately after the confirmed segment is returned to the undetermined state according to the second embodiment, the segment deciding process is performed again for a newly decidable segment. In the second embodiment, the segment reconfirmation process is not performed until the next stroke is input. However, since the number of unconfirmed segments increases when a segment is confirmed and canceled, it is desirable to perform the reconfirmation process immediately after canceling the confirmation in order to minimize the time lag.
[0077]
The configuration of the third embodiment is shown in FIG. In this configuration diagram, the segment confirmation cancellation unit 8, the character area cancellation unit 9, and the optimum path cancellation unit 10, which are the segment determination cancellation unit of the second embodiment, are configured by four processing blocks including a segment determination instruction unit 11. The segment reconfirmation unit 12 is realized as a segment reconfirmation unit. In this case, by issuing an instruction to re-establish the segment after the confirmed segment is canceled, the latest segment confirmed state is realized even if a new stroke is not input. The instruction for canceling the confirmed segment and re-establishing the segment is not particularly required by the user, and it is preferable to automatically move when a character correction instruction is issued.
[0078]
Normally, when a stroke is input, an undefined segment is created by the segment determination unit 2 and stored in the segment temporary storage unit 3. Thereafter, the segment confirmation determination unit 5 determines a definable segment among the unconfirmed segments stored in the segment temporary storage unit 3, and uses the determinable segment as a definite segment. When a character is deleted or a handwritten character is added, segment confirmation cancellation processing and segment confirmation processing are subsequently performed without performing stroke input.
[0079]
The segment confirmation instruction unit 11 instructs the segment determination unit 2 to start the confirmation process after the processing of the segment confirmation cancellation unit 8, the character area cancellation unit 9, and the optimum path cancellation unit 10. The segment determination unit 2 redoes the segment determination, and moves the segment that has returned from the confirmed segment to the unconfirmed segment from the confirmed segment list in the segment temporary storage unit 3 to the unconfirmed segment list. When the segment determination unit 2 finishes storing the undetermined segment, the segment determination determination unit 5 starts the segment determination process. As a result, of the unconfirmed segments that can be confirmed are stored in the segment temporary storage unit 3 as confirmed segments.
[0080]
Specifically, in the example of FIG. 13, when “shi” of the handwritten character “frameless recognition” is deleted, the confirmed segment “shi” is deleted, and the confirmed segments “word” and “shinobi” of “acknowledgment” are deleted. Is returned to the indoubt segment. In the case of the second embodiment, segment determination / segment confirmation determination, character area determination / character recognition / optimum path generation is not performed until the next stroke is input, but in this embodiment, the input of the next stroke is awaited. There is no need to perform segment determination / segment determination / character area determination / character recognition / optimum path generation for “acknowledgement” and “shinobi” of “acknowledgement” that were unconfirmed segments in (c) Then, as shown in (d), the recognition result changes from “no frame” to “no frame”.
[0081]
With this configuration, even if the segment after the corrected character becomes unconfirmed due to deletion or addition of handwritten characters, recognition processing for the unconfirmed segment can be performed immediately without waiting for the next stroke to be input. It becomes.
[0082]
(Embodiment 4)
The configuration of the fourth embodiment is shown in FIG. In the second and third embodiments, the function of canceling a confirmed segment and performing the segment confirmation process again has been described. As a result, for example, as shown in FIG. 13, the recognition result character string can also be corrected. In this embodiment, in the second to fourth embodiments, when correction of a handwritten character occurs and the recognized character string changes, an example of notifying how the recognized result character string already notified to the application has changed is notified. An example of an application that displays on a screen handwritten character recognition characters included in software such as schedule management using a handwritten character input or word processor will be described.
[0083]
First, how the character recognized by handwriting is displayed in the present invention will be described. In the present invention, every time one stroke is input, the input handwritten character is divided into a confirmed segment and an unconfirmed segment, and character recognition is performed only for the confirmed segment. This process is repeated every time one stroke is input. However, since the number of newly recognized characters does not always increase even if one stroke is input, the recognition result character to the application is input once or a plurality of times. Only once when the number of newly recognized characters increases. For this reason, when strokes are input in order from the front, the display application is usually notified one character at a time as the recognition result, and the display application additionally displays the notified recognition result character on the screen. If a delete instruction is issued or a stroke is added to an already recognized handwritten character (consisting of a confirmed segment), all segments after that segment become unconfirmed segments. The confirmed segment is deleted, and the subsequent segments become undefined segments. The process from segment determination to optimal path generation is executed again, and only when a new recognition result is generated, a new recognition result string ( The number of characters is 1) or more is notified to the display application. In this case, if the display application receives the new recognition result character string as it is and additionally displays it on the screen, the display application will additionally display the handwritten character that has been displayed once without correcting it, resulting in incorrect display. .
[0084]
In other words, if the already recognized handwritten character is deleted or added, the recognition result character string changes, and the application needs to do some processing such as changing the character string shown to the user in response to the correction of the received recognition result. It becomes. That is, when the recognition result already notified to the application is corrected, the recognition engine side needs to notify the application of the correction content.
[0085]
In the following, an embodiment will be described in which the online character recognition device without a real time frame of the present invention notifies the application of the content whose recognition result has been corrected by canceling the confirmed segment.
[0086]
FIG. 15 is obtained by adding a recognition result correction information output unit 14 to FIG. 14 of the third embodiment. In the first and second embodiments, the optimal path generation unit 7 notifies the recognition result character string to the application. However, in this embodiment, the recognition result information output unit 13 selects either the recognition result character string or the recognition result correction information. One of them is notified to the application.
[0087]
The recognition result information output unit 13 includes the optimum path generation unit 7. When there is no correction, the recognition result character string is displayed. When there is a correction, the information indicating what character was corrected and the corrected recognition result character. The column (all recognition result character strings after the correction start position) is notified to the application.
[0088]
First, when the determined segment becomes an undetermined segment due to the correction of handwritten characters, the optimal path cancellation unit 10 deletes the path information of the segment that has become an undetermined segment from the path information held by the optimal path generation unit 7. However, at this time, the path deletion information indicating from which character the path is deleted is passed to the recognition result information output unit 13.
[0089]
The recognition result information output unit 13 determines that the recognition result character string generated by the optimum path generation unit immediately after receiving the path deletion information is a corrected recognition result character string, and determines from what character the character was corrected. When the correction start position information (= path deletion information) and the corrected recognition result character string are notified to the application and the path deletion information has not been received immediately before, the recognition result character string generated by the optimum path generation unit is corrected. It is determined that the character string is not a recognition result character string, and only the recognition result character string is notified to the application.
[0090]
Hereinafter, how the recognition result correction information is created will be described.
[0091]
It is assumed that the result is that the “frameless character recognition engine” has already been recognized. It is assumed that “recognition” of “recognition engine” is deleted and “input” is input by handwriting at the position where “recognition” was written. When “recognition” is deleted, the segment constituting the handwritten character “engine” becomes an undetermined segment. When “recognition” is deleted, the segment confirmation cancellation process is executed, and the optimum path cancellation unit 10 leaves the path up to the fifth character of “frameless character recognition engine”, and passes the path after the sixth character. The path deletion information indicating deletion and deletion from the sixth character is passed to the recognition result information output unit 13. After the optimal path cancellation process, the segment determination instruction unit 11 performs the processes from the segment determination unit 2 to the optimal path generation unit 7 again for the sixth and subsequent undefined segments, and the corrected six characters. The recognition result character string “input engine” after the first is generated as the recognition result character string of the optimum path generation unit 7. The recognition result information output unit 13 uses the path deletion information that the path has been deleted from the sixth character received from the optimum path cancellation unit immediately before as the correction start position information (= “6”) that the path has been corrected from the sixth character. Then, the display application is notified together with the corrected recognition result character string “input engine”. The display application that has already displayed “frameless character recognition engine” on the screen receives the correction start position information “6” and the corrected recognition result character string “input engine”, and displays the “no frame” The display from the sixth character of “Character recognition engine” is changed to “Input engine”, and “Frameless character input engine” is displayed again.
[0092]
With this configuration, even when the recognition result character string is corrected, the online handwritten character recognition device without real time frame of the present invention can notify the cooperating application of the correction content.
[0093]
(Embodiment 5)
Next, the program for realizing the real-time online frameless handwritten character recognition apparatus according to the embodiment of the present invention is a CD-ROM 90-1, a flexible disk 90-2, etc. as in the example of the recording medium shown in FIG. In addition to the program recorded in the portable recording medium 90, the program may be recorded in any of the other storage devices 91 stored at the end of the communication line, the recording medium 92 such as the hard disk or RAM of the computer, When the real-time online frameless handwritten character recognition apparatus according to the present invention is used, it is read by the computer 93 and executed on the main memory.
[0094]
By loading this program into a computer, a real-time online frameless handwritten character input device that can sequentially obtain recognition results during writing can be realized.
[0095]
【The invention's effect】
According to the real-time online frameless online recognition apparatus of the present invention, the recognition result up to a portion more than the predetermined distance from the writing position can be obtained with high accuracy in real time, and the user can wait for the frameless character recognition. It becomes possible to eliminate the need to worry about time.
[0096]
In addition, according to the real-time online no-frame handwritten character recognition apparatus of the present invention, the re-recognition / recognition processing function is added, so that if it is confirmed by mistake, it is not fatal and only the wrong part is redone. Therefore, even if the gap between the writing position and the recognition position is reduced (= threshold d is reduced), no problem occurs, and performance degradation due to an error in segment determination processing can be avoided.
[0097]
In addition, according to the real-time online frameless handwritten character recognition apparatus of the present invention, the next stroke is input even when the segment after the corrected character becomes unconfirmed due to the deletion or addition of the handwritten character. Without waiting, recognition processing for an indeterminate segment can be performed immediately.
[0098]
Moreover, according to the real-time online no-frame handwritten character recognition apparatus of the present invention, even when the recognition result character string is corrected, it is possible to notify the linked application of the correction contents.
[0099]
Also, by loading this program into a computer, a real-time online frameless handwritten character input device capable of sequentially obtaining recognition results during writing can be realized.
[Brief description of the drawings]
FIG. 1 is a configuration diagram of Embodiment 1 of the present invention.
Fig. 2 Data structure of input data (C language representation)
FIG. 3 is a diagram for explaining a method for improving character segmentation.
FIG. 4 is an operation flowchart of a segment determination unit.
FIG. 5 shows an example of operation of the segment determination unit.
[Figure 6] Segment data structure (C language representation)
FIG. 7 shows an example of the operation of the segment confirmation determination unit.
[Figure 8] Generation of character area candidates
[Figure 9] Generation of character area links
FIG. 10: Data structure of character area (C language representation)
FIG. 11 is a configuration diagram of Embodiment 2 of the present invention.
[Figure 12] Flow of segment confirmation cancellation processing
[Figure 13] Segment fixed cancellation processing
FIG. 14 is a configuration diagram of Embodiment 3 of the present invention.
FIG. 15 is a configuration diagram of Embodiment 4 of the present invention.
FIG. 16 is an exemplary diagram of a recording medium.
[Fig.17] Character cutout error for frameless recognition
[Figure 18] Character recognition without frame
FIG. 19 is a configuration diagram of Conventional Example 1;
20 is a configuration diagram of Conventional Example 2. FIG.
FIG. 21 is an operation flowchart of conventional example 2 of frameless character recognition.
FIG. 22 shows an example of character segmentation according to Conventional Example 2.
FIG. 23 is a diagram for explaining a method for improving character segmentation.
[Explanation of symbols]
1 Stroke input section
2 Segment judgment part
3 Segment temporary storage
4 Character area judgment part
5 Segment confirmation section
6 Individual character area recognition unit
7 Optimal path generator
8 Segment Confirmation Cancellation Department
9 Character area cancellation part
10 Optimal path cancellation section
11 Segment confirmation instruction section
12 Segment reconfirmation section
13 Recognition result information output part
30 Character string pattern input unit of Conventional Example 1
31 Segment determination unit of Conventional Example 1
32 Character area determination unit of Conventional Example 1
33 Individual character identification unit of Conventional Example 1
34 Optimal path generation unit of Conventional Example 1
40 Stroke input section of Conventional Example 2
41 Character boundary determination unit of Conventional Example 2
42 Character pattern generation unit of Conventional Example 2
43 Unrecognized character recognition unit of Conventional Example 2
44 Stroke temporary storage section of Conventional Example 2
90 Portable recording media such as CD-ROM and flexible disk
90-1 CD-ROM
90-2 Flexible disk
91 Other storage devices stored at the end of the communication line
92 Recording medium such as RAM / hard disk on computer
93 computers

Claims (2)

In an online handwritten character recognition device that does not have a character input area for each character,
A stroke input unit for inputting a stroke of handwritten characters and converting it to stroke data;
Said stroke data and based on an existing undetermined segment, the stroke data, new undetermined segment that should be added to the existing undetermined segment or determining segment determination unit,
Based on the segment, a segment determination unit that determines whether the segment is a fixed segment or an undefined segment, and determines the segment;
A character area determination unit that determines a character area for each character including a combination of determined segments based on the determined segments;
Based on the stroke data included in each character region, an individual character recognition unit that recognizes the handwritten character online for each character region and generates character candidates;
An optimal path generation unit that generates an optimal path of a combination of character candidates for each character area and outputs an optimal character combination as a recognition result ; and
When the stroke data is input to a confirmed segment, it is determined whether or not the stroke data is a deletion instruction. When the stroke data is a deletion instruction, the determined segment to which the stroke data is input is deleted and the subsequent determination is performed. If the segment is in an unconfirmed state and is not a deletion instruction, a segment confirmation canceling unit that places the confirmed segment after the confirmed segment in which the stroke data is input into an unconfirmed state;
A character area canceling unit that cancels the determination of the character area for the segment processed by the segment confirmation canceling unit,
An optimal path cancellation unit that cancels the optimal path for the segment processed by the segment confirmation cancellation unit;
If the stroke data is a deletion instruction, a segment confirmation instruction unit that instructs the segment determination unit to redo the determination of the segment returned to the unconfirmed state by the segment confirmation cancellation unit;
A recognition result information output unit for outputting correction information including a recognition result character string after the character corrected by the cancellation of the segment confirmation and correction start position information indicating from which character of the already output recognition result character string is corrected Comprising
When the stroke data is input to a confirmed segment, the segment confirmation cancellation unit, character area cancellation unit, optimum path cancellation unit, and segment determination instruction unit are processed. If the stroke data is not input to the confirmed segment, the correction information for the output recognition result character string is output to the linked application. Each time one stroke is input, the segment determination unit, the segment determination unit, the character region determination unit, the individual character recognition unit, and the optimum path generation unit are executed to perform character recognition of the segment portion whose input has been determined. Online handwritten character recognition device without real time frame. Computer to recognize online handwritten characters without real time frame,
Means for inputting strokes of handwritten characters and converting them into stroke data;
Segment means for determining, based on the stroke data and an existing unconfirmed segment, whether the stroke data is a new unconfirmed segment or is added to an existing unconfirmed segment;
Segment means for determining whether the segment is a confirmed segment or an indeterminate segment based on the segment and determining the segment;
A character area determination means for determining a character area for each character comprising a combination of determined segments based on the determined segments;
Based on the stroke data included in each character region, individual character recognition means for generating a character candidate by recognizing the handwritten character online for each character region;
The character generating an optimal path of a combination of character candidates of each region, the optimal path generation means for outputting a character combination of the perfect recognition result,
If the stroke data is entered in to confirm segment, the stroke data is determined whether or not deletion instruction, if the deletion instruction, the subsequent delete the determined segment, wherein the stroke data is input A segment confirmation canceling means for setting a confirmed segment in an unconfirmed state, and in a case where it is not a deletion instruction, a confirmed segment after the confirmed segment to which the stroke data is input is in an unconfirmed state;
Character area cancellation means for canceling the determination of the character area for the segment processed by the segment confirmation cancellation means,
Optimal path cancellation means for canceling the optimal path for the segment processed by the segment confirmation cancellation means,
If the stroke data is a deletion instruction, a segment confirmation instruction means for instructing the segment determination means to redo the determination of the segment returned to the unconfirmed state by the segment confirmation cancellation means; and
Recognition result information output means for outputting correction information composed of a recognition result character string after the character corrected by the cancellation of the segment confirmation and correction start position information indicating from which character of the already output recognition result character string is corrected , An online handwritten character recognition program without a real time frame to function as .

Images