KR20010026560A - Korean Stroke Recognition Method with Unrestricted Writing Pattern - Google Patents

Abstract

PURPOSE: A Hangul stroke recognition method is provided to recognize the kind of a current stroke by referring to a stroke library at a current stroke branch point, and select an optimal combination stroke among a candidate stroke combination list so that it can make it possible for a user to instantly identify an input stroke, for example, in a PDA or a smart phone. CONSTITUTION: The method comprises steps of receiving continuous coordinate values input by a user with a stylus(S10), converting the input coordinate values into recognizable data by eliminating unnecessary input points to determine a shape of an input stroke(S20), identifying the kind of the currently input stroke by considering a stroke library at the current stroke branch point according to a Hangul syllabic combination rule, and determining if the identified stroke is included in a candidate stroke combination list(S30), selecting sequentially an optimal candidate stoke according to a stroke combination probability(S40), determining if the selected candidate stroke is recognized as a Hangul syllable(S50), displaying the stroke with prior recognized strokes if the selected candidate stroke is recognized as a Hangul syllable or storing the stroke at a memory unless the selected candidate stroke is recognized as a Hangul syllable(S60), and constructing and storing an estimated candidate stroke combination list for a next input stroke(S70).

Description

Korean Stroke Recognition Method with Unrestricted Writing Pattern

The present invention relates to a method of recognizing Hangul 없는 without writing constraints. More specifically, the stroke recognizer is difficult to recover an error when recognition is terminated. If the problem is outputted, the problem is solved by using back tracking in the combination of strokes, and it can be used for all input devices. It is about.

Due to the increasing demand for small personal terminal devices (eg, PDAs, Palm PCs, Sub NoteBook PCs, and Smart Phones) that are difficult to use keyboards, input methods using online text recognition are becoming more and more common. However, since one character is composed of several strokes, unlike keyboard input, recognition of input of each stroke does not start and recognition can be started only after the input of the entire character is finished. In this case, it is common to take a method of waiting for a delay time after pressing a specific character or performing stroke writing as a method of notifying that the character input is completed.

However, it is cumbersome to press the button that indicates the end of the input, and the method of waiting for a fixed time after writing the stroke is boring to the writer if this time is long, and if the time is short, the recognition starts before the character is correctly inputted. Do.

In order to prevent the user from waiting for a certain time without any special operation to inform the type of the character input, in English, the concept of 'UNISTLOCK' is introduced so that all characters can be written in one stroke to check the stroke as soon as the input is completed. So that you can enter any character you want (e.g. 'Graffiti', which is used in Palmpilot, or Windows CE's built-in recognizer).

Hangul also attempted to provide a recognizer that interprets the result of handwriting and recognizes the input immediately. For example, in the case of 'Hantip' of 'Lay Solution', a character recognizer using a combination of phonemes may be developed by introducing the English Unisclock concept into Korean by expressing each phoneme forming a Hangul character. In this case, the proposed phoneme is more deformed from the original phoneme shape, so it is easy to write wrong, and such a constraint is a burden on the writer. For example, writing '??' as shown in FIG. 1 (a) and writing the phoneme '??' as shown in FIG. 1 (b) are possible in one stroke. This problem occurs because the Hangul alphabet is composed of several strokes unlike the alphabet. As a way of solving this handwriting constraint to some extent, a method that can recognize strokes while supporting more natural writing than Uniscrock is presented in 'Wisephone' of 'Open Solution'. These two methods are basically a method of outputting characters by directly using a stroke recognition result in a combination of phonemes. In both cases, Neutral makes sure that the basic stroke is written correctly. In the case of neutral, writing is possible with only horizontal strokes and vertical strokes, and the stroke order is almost constant.

Hangul input order is generally the order of the first consonant, the neutral, and the final consonant, and the neutral can be classified into three types (vertical, horizontal, and complex) by using the relative position of the first consonant. The recognizer is operated. In the case of 'Hiptip', the input strokes are mapped in order of initial, neutral, and final. The absolute position of the stroke is used to distinguish the first and last species, and each element of the first and final species is determined in one form, so the progress between the elements is easy. However, if the last character of the character ending in '??' is the initial character of '??', the distinction between the final character and the initial character is not good. In this case, a gesture or the like indicating the end of the character input should be written. In addition, the starting position of the initial is fixed.

In the case of Wise Pen, one phoneme can be written in multiple strokes, so it waits for the recognition result until the correct phoneme is input. For example, if you enter 4 strokes '??', there is no reaction until 3 strokes are entered, but after the last stroke is input, '??' is output. In addition, because each element has limitations, '??' or '??' consisting of several strokes should be written in a predetermined stroke order, and the final character '??' of the character 'mouth' can be written in two types of four strokes. However, in the case of the character 'Eup', the final character '??' should be written in four strokes of one kind. In the case of '??', only three fixed forms can be written as shown in FIG.

In the case of 'Hiptip', it is necessary to memorize the unnatural phoneme shape so that the user has to memorize the shape, and even if it memorizes, there is anxiety about whether the user writes the correct form every time a character is entered. It is a burden. In the case of 'Wise Pen', which solves this problem to some extent, the consonants such as 4 strokes, which are frequently confused, differ in the types of strokes that are allowed depending on the neutrality, and the complex consonants '??' As in the case, there are limitations that must be taken only in the prescribed form. For example, when writing 'eup', 'o + ??' is written and '?? + ??' is written. In this state, even if the '??' stroke is outputted, the stroke is not recovered in the 'Yu' state, and the stroke '??' is regarded as the first stroke of the next character or is regarded as an error. At the same time, the '??' of the 'mouth' can be written in four strokes, which can confuse the user. In addition, stroke errors cannot be overcome at all. This is a problem that often occurs when a writer writes 'za', and when writing horizontal lines of '??' close to horizontal, 'gi' is already output when 'ji' is inputted. If it's written, the stroke is ignored or considered the next character. When entering 'week', often enter '?? The result of typing + ?? 'is expected to be' off ', so even if you enter the last' ?? 'stroke, the output remains' off' and '??' is ignored or considered the next character. Since this does not overcome the stroke recognition error, such a case occurs frequently during actual writing, which causes a problem of poor stability as a product.

Accordingly, the present invention has been made to solve such a problem, and in order to provide a Hangul stroke recognizer capable of taking natural handwriting while obtaining an input result immediately upon handwriting, such as a keyboard, the Hangul is basically not burdened by the writer. Four strokes with different stroke order and three strokes with different stroke order, and two strokes with two frequently written strokes I also want to support. In addition, the possible writing forms in the initial and the final consonants are the same, and the objective of the present invention is to provide a consistent writing method to the user by supporting the multiple consonants in the extended form of the terminal sound.

Another object of the present invention is to convert the 'yu' to 'eup' and 'war' to 'eup' when writing 'eup' and so on. It allows you to take notes with almost no restrictions. In particular, it is possible to recover an error for frequently used strokes, thereby providing a user with a satisfactory recognition result.

1 is an exemplary diagram showing an example of a phoneme written in a modified cursive;

2 is a flowchart illustrating a preferred embodiment of a Hangul stroke recognition method without handwriting constraints according to the present invention;

3 is an exemplary diagram illustrating a process of waiting for a stroke for outputting or predicting a corresponding phoneme for each input stroke;

FIG. 4 shows a first combination example of the overall stroke combining process when the final word '??' of the Hangul 'eup' is written in the order '??', '??', '??', '??' Example,

FIG. 5 shows a second combination example of the overall stroke combining process when the final word '??' of the Hangul 'eup' is written in the order '??', '??', '??', '??' Example,

6 illustrates an example of recovering a stroke recognition error in a combination process when the writer takes the horizontal stroke of '??' close to a horizontal line when writing the ruler.

In order to achieve the above object, the Hangul 휙 recognition method without writing constraint according to the present invention identifies a new stroke type by referring to a stroke library at a stroke branch point where writing for a new stroke is completed, and the stroke for which the identified stroke is expected. After determining whether it is included in the combination candidate list and selecting the best stroke combination candidate from the expected stroke combination candidate list, performing stroke recognition according to the Hangul alphabet combination rule, and then until the next stroke input is not authorized, Create a stroke combination candidate list containing at least one stroke combination candidate expected to be entered into and store in memory, and if the identified stroke is not in the expected stroke combination candidate list, write at this point from the current stroke bifurcation A back-tracking process is performed to the stroke branch point just before the initial character of the character being determined is determined. And when the stroke to be included in the combination of the candidate list of the expected stroke identifies the stroke path combinations to track up to this point is derived from the respective branch point stroke is characterized by performing a stroke recognition.

Hereinafter, with reference to the accompanying drawings, preferred embodiments of the Hangul stroke recognition method without writing constraints according to the present invention will be described in detail as follows.

2 is a flowchart illustrating a preferred embodiment of a Hangul stroke recognition method without writing constraints according to the present invention.

According to an exemplary embodiment of the method for recognizing a Hangul stroke without handwriting constraint according to the present invention, as shown in FIG. 2, a user writes a writing unit using a writing means on an input window of a predetermined size of a device capable of detecting an input such as an LCD. Receiving the received coordinate values (S10);

Converting the inputted coordinate values into data that can be used by an actual recognizer by removing only the input points that are necessary for determining the stroke shape and removing the remaining input points (S20);

Each stroke combination is written to form a phoneme, and each phoneme is entered in the order of initial, neutral, and final characters to form a character. Identifying a type of the new stroke with reference to and determining whether the identified stroke is included in an expected stroke combination candidate list (S30);

If the identified stroke is included in the expected stroke combination candidate list, the optimal stroke combination candidate in the expected stroke combination candidate list using the result of the phoneme and stroke combination up to the present time and the relative position of the identified stroke Sequentially selecting according to the stroke combination probability (S40);

Determining whether the selected optimal stroke combination candidate is recognized as a phoneme (S50);

When the selected optimal stroke combination candidate is recognized as a completed phoneme, the corresponding phoneme is output in combination with the phonemes recognized so far in a phoneme unit, and when the selected optimal stroke combination candidate is recognized as an incomplete phoneme. Storing in the memory without outputting the identified stroke (S60);

After determining whether the recognized stroke is recognized for the identified stroke, based on the result of the phoneme and stroke combination up to the present time, at least one stroke combination candidate expected to be input until no further stroke input is authorized Preparing a recognition for the next stroke to be input by preparing and including a stroke combination candidate list in the memory (S70); And

If the identified stroke is not included in the predicted stroke combination candidate list, the next stroke combination candidate is selected according to the stroke combination probability from the current stroke branch point to the stroke branch point immediately before the initial constellation of the character being written at this point is determined. When the identified strokes are included in the predicted stroke combination candidate list by tracking the stroke combination paths derived from each stroke branch to the present time while performing the back-tracking process of selecting the strokes one by one sequentially. It is configured to perform the step (S80) to perform the stroke recognition.

Referring to the accompanying drawings, a procedure of performing a preferred embodiment of the method for recognizing a Hangul stroke without writing constraints according to the present invention configured as described above will be described.

Prior to describing the implementation procedure of the present invention, in order to avoid confusion of the terms used herein, the concept of the term 'phoneme', 'letter' is defined, first, in Korean, 'jamin' refers to consonants and vowels. It is the basic unit that forms the initial, neutral, and final species, and at least one stroke is gathered into one phoneme. Here, the vowel is composed of a single vowel and a double vowel, and the consonant is also composed of a double consonant and a terminal sound. In addition, 'letter' refers to a combination of phonemes to form a meaningful letter, which is formed in the order of initial, neutral, and final to form a single letter. It is well known that one letter can be gathered together to form a single letter and the first and neutral stars can be gathered together.

First, in step S10, a series of coordinate values written by a user are received by using a writing means in an input window of a predetermined size of a device capable of sensing an input such as an LCD. In this case, since the relative position information of the phoneme should be used, writing is preferably performed at a fixed position to some extent.

Thereafter, in step S20, only the input points necessary for determining the stroke shape are left and the remaining input points are removed to convert the input coordinate values into data that can be used by the actual recognizer.

Next, in step S30, the phoneme is formed by the respective stroke combinations written, and each phoneme is inputted in the order of initial, neutral, and final, and the writing of the new stroke is performed according to the Hangul phoneme combination rule for forming letters. The type of the new stroke is identified by referring to the stroke library at the completed current stroke branch point, and it is determined whether the identified stroke is included in the expected stroke combination candidate list.

Normally, there are seven basic strokes for Hangul phonemes, which are vertical ('??'), horizontal ('??'), diagonal ('/'), and historical lines ('＼) found in the initial, neutral, and final modes. ')' And '??', '??', and '??' found only on the first and last planets. In addition to the stroke for the two-stroke '??' of the form (d) of Figure 1, and one stroke '??', '??', '??' The stroke recognition is the process of separating them. In the recognition process, dynamic matching and rules are used. A new stroke is identified by referring to a stroke library previously learned.

On the other hand, the list of expected stroke combination candidates is created by referring to the Hangul alphabet combination rule and the stroke library. Briefly, the principle of writing the strokes is as follows: First, first stroke is input, and then neutral is input. A stroke for forming a primary or final star must be input, and a stroke for forming a double vowel may be applied after the short vowels constituting the neutral.

As a result of the determination in step S30, if the identified stroke is included in the expected stroke combination candidate list, in step S40, the result of the phoneme and stroke combination up to the present time point and the relative position of the identified stroke are used. The optimal stroke combination candidates are sequentially selected based on the stroke combination probability from the predicted stroke combination candidate list.

Here, the distribution of stroke combination probabilities gives a high probability to the formation of the completed phoneme of the identified stroke, and if the identified stroke is expected to be part of the unfinished phoneme, then there are many combinations of possible stroke combinations. It is desirable to give priority to things.

Subsequently, in step S50, it is determined whether the selected optimal stroke combination candidate is recognized as a phoneme.

As a result of the determination in step S50, when the selected optimal stroke combination candidate is recognized as a completed phoneme, in step S60, the corresponding phoneme is output in combination with the phonemes recognized so far in a phoneme unit and outputted. If the best stroke combination candidate is recognized as incomplete, then the identified stroke is stored in memory without outputting.

After determining whether the phoneme is recognized for the identified stroke, in step S70 at least one expected to be input next until no further stroke input is applied based on the result of the phoneme and stroke combination up to the present time. A stroke combination candidate list including the above stroke combination candidates is prepared and stored in the memory to prepare for recognition of the next stroke to be input.

As a result of the determination in step S30, if the identified stroke is not included in the expected stroke combination candidate list, in step S80, the stroke immediately before the initial character of the character being written at the present time is determined from the current stroke branch point. The stroke combination paths derived from each stroke branch point are traced to the present point while performing the back-tracking process to select stroke candidates sequentially one by one according to the stroke combination probability up to the branch point. Stroke recognition is performed until the identified stroke is included in the expected stroke combination candidate list.

In order to help the understanding of the present invention, the process of the step S50 to the step S80 is further added. In the combining step, if the input stroke turns out to be a phoneme, the process of waiting for the remaining stroke otherwise To perform. At this time, the back-tracking technique is applied to correct the effective stroke combination and stroke recognition error. If it is determined that the combination of the identified strokes constitutes a proper phoneme, the corresponding phoneme is output, and when a character is formed, stroke recognition for the next character is performed.

Table 1 shows the number of terminal sounds supported by the present invention.

Jaso ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? ?? amount One One 2 2 2 3 One One One One One 3 One 3

For example, for the consonant 'ㄼ', '??' 2 types and '??' Up to six stroke combinations are possible in three combinations. Neutrality follows the number of strokes used and the order in which they are commonly used. Table 2 shows the number of strokes for each neutral.

neutrality ?? ?? ?? ?? ?? ㅚ ㅙ ㅘ ?? ?? ?? ?? ?? ?? ㅢ ㅟ ㅝ ㅞ ?? ?? ?? Strokes One 3 4 2 3 3 5 4 2 3 2 3 3 4 2 3 4 5 2 3 One

The Hangul stroke recognizer is expected to have written strokes in the order of initial, neutral, and finality. When the expected strokes are input, the corresponding characters are matched and the characters are finally output through the combination process. As shown in FIG. 3, the recognized character is immediately provided while outputting the corresponding phoneme for each stroke input or waiting for an expected stroke. Progress to the next character is determined using the result of the phoneme combination and the relative position of the identified stroke.

Since the phoneme is formed by the combination of strokes and the letters are formed by the combination of phonemes, the phoneme proceeds from a broad perspective and the stroke proceeds within each phoneme. At this time, a stroke that is not included in the expected stroke combination candidate list, that is, a stroke different from the expectation may be input, and some special processing for this is required. This is the case where the Hangul stroke recognizer proceeds with a combination that is different from the intention of the writer for a stroke capable of two or more combinations, and there is a misperception among strokes that have already been recognized.

First, a case in which two or more phoneme combinations can be processed will be described with reference to FIGS. 4 and 5.

As in FIGS. 4 and 5, when the finality '??' of 'eup' is written as '??', '??', '??', '??', the '??' If the stroke '??' is input while the + ?? 'is written, it may be the neutral' ?? 'or the first stroke of the neutral' ?? 'and the final star as in the stroke branching point of FIG. In other words, in this case, the stroke '??' is included in the list of expected candidate stroke combinations. In addition, the stroke '??' is likely to proceed in the form of a plurality of stroke combinations. In this case, in order to predict the exact stroke direction in a shorter time as one stroke direction must be selected, a plurality of standard Korean characters can be used. Select stroke combination population and calculate stroke combination probability according to stroke combination frequency offline and store it in memory for reference.

Subsequently, when the next '??' stroke is written, when viewed by a user other than the writer, this is 'Yes' as shown in FIG. Therefore, in the predicted stroke combination candidate list, stroke input for forming the finality and the consonant is included in the combination candidate list. However, as shown in (b) of FIG. 4, when the user sees the next horizontal line, the user expects the user to write 'eup'. In this case, if the stroke is different from the intention of the writer and receives an unexpected stroke as input, the trace proceeds back to the previous stroke branch point and sequentially selects the next stroke combination candidates one by one to recognize the stroke. tracking) process. This traceback process tracks the stroke combination paths derived from each stroke bifurcation point to the present time and performs stroke recognition until the identified stroke is included in the expected stroke combination candidate list.

In this case, the recognizer can anticipate that the next row will attempt to take the 'eup' at the same time as selecting the second stroke combination candidate at the immediately preceding stroke junction.

Then naturally '?? + ?? + ?? + ?? + ?? 'is expected to be up to 5 strokes of' eup ', expecting the next' ?? 'character, and normal character combinations can be terminated.

In the case of FIG. 5 (a), the phoneme combination is proceeded up to the initial '??' and the neutral 'ㅝ' and then traced back to the next stroke '??' and the final '??' to finally determine 'eup'. You can also proceed.

Second, the recovery of the stroke recognition error in the combining process will be described with reference to FIG. 6.

As shown in FIG. 6, when a writer writes a ruler, the horizontal stroke of '??' is often written close to horizontal. When inputting up to '??' as shown in (a) of FIG. 6, since the recognizer has already completed the phoneme combination, the next stroke '??' is regarded as an invalid input in the conventional stroke recognizer. This stroke is considered to be the beginning of finality without discarding, discarding, or using location information. This error cannot be solved even in the same phoneme area, neutral, but must be traced back to the beginning and combined again. This can be achieved by keeping separate candidates for strokes that do not have a clear recognition result and then using this information for backtracking.

Through the above two processes, problems that cannot be solved by the conventional stroke recognizer are overcome, which provides an accurate recognition result in almost all cases.

Terminologies used herein are terms defined in consideration of functions in the present invention, which may vary according to the intention or customs of those skilled in the art, and the definitions should be made based on the contents throughout the present application. will be.

Although specific embodiments of the invention have been described and illustrated, it will be apparent that the invention may be embodied in various modifications by those skilled in the art.

In addition, since the present invention has been described through the preferred embodiment of the present invention, in view of the technical difficulty aspects of the present invention, those having ordinary skill in the art can easily be different from another embodiment of the present invention. As modifications may be made, it is obvious that both the embodiments and modifications cited in the above description belong to the appended claims.

As described in detail above, the present invention refers to the stroke library at the current stroke branch point where the writing of the new stroke is completed, identifies the type of new stroke, and determines whether the identified stroke is included in the expected stroke combination candidate list. Selects an optimal stroke combination candidate from the list of expected stroke combination candidates, performs stroke recognition according to the Hangul alphabet combination rule, and then adds at least one or more of the expected stroke inputs until no further stroke input is authorized. Create a stroke combination candidate list containing stroke combination candidates and store them in memory, and if the identified stroke is not in the expected stroke combination candidate list, immediately before the entry of the character being written at the present time is determined from the current stroke branch point. At each stroke junction, performing a back-tracking process up to According to the present invention, which tracks the completed stroke combination path to the present time and performs stroke recognition until the identified stroke is included in the predicted stroke combination candidate list, it can be used for all input devices and can immediately know the result of the writing result. But there is an advantage that there is no writing constraint.

In particular, these stroke recognizers require minimal resources and provide recognition results in real time, making them the same level of input methods that keyboards provide on desktop PCs, making it difficult to provide even a small device terminal and soft keyboard that are difficult to mount. It can be used as an optimal input device in a smart phone.

Claims (3)

Each stroke combination is written to form a phoneme, and each phoneme is entered in the order of initial, neutral, and final characters to form a character. Identifying the type of the new stroke with reference and determining whether the identified stroke is included in an expected stroke combination candidate list; If the identified stroke is included in the expected stroke combination candidate list, the optimal stroke combination candidate in the expected stroke combination candidate list using the result of the phoneme and stroke combination up to the present time and the relative position of the identified stroke Sequentially selecting according to stroke combination probabilities; Determining whether the selected optimal stroke combination candidate is recognized as a phoneme; When the selected optimal stroke combination candidate is recognized as a completed phoneme, the corresponding phoneme is output in combination with the phonemes recognized so far in a phoneme unit, and when the selected optimal stroke combination candidate is recognized as an incomplete phoneme. Storing in the memory without outputting the identified stroke; And After determining whether the recognized stroke is recognized for the identified stroke, based on the result of the phoneme and stroke combination up to the present time, at least one stroke combination candidate expected to be input until no further stroke input is authorized And preparing to recognize a stroke to be input next by creating a stroke combination candidate list including the stored stroke combination candidate list in the memory. The method of claim 1, If the identified stroke is not included in the predicted stroke combination candidate list, the next stroke combination candidate is selected according to the stroke combination probability from the current stroke branch point to the stroke branch point immediately before the initial constellation of the character being written at this point is determined. When the identified strokes are included in the predicted stroke combination candidate list by tracking the stroke combination paths derived from each stroke branch to the present time while performing the back-tracking process of selecting the strokes one by one sequentially. Handwriting constraint free Hangul 이 recognition method characterized in that it further comprises the step of performing a stroke recognition. The method of claim 1, Receiving continuous coordinate values written by a user using a writing means in an input window of a predetermined size of a device capable of detecting an input; And And converting the inputted coordinate values into data that can be used by the actual recognizer by removing only the input points that are necessary for determining the stroke shape and removing the remaining input points. Way.