KR930012140B1 - Recogntion method of on-line writing down character using stroke automata - Google Patents

Abstract

Sixteen strokes are defined and the phoneme automata is executed according to the strokes to reduce recognition speed. The method comprises a pre-processing step for removing data unnecessay for recognition from data input from a tablet, a character point extraction step for detecting characteristic point from pre-processed data, a phoneme automata execution step for executing consonant and vowel automata of the initial, middle and last phonemes of input stroke, and a Hangul recognition step for recognizing input character according to code value of the phoneme automata execution step.

Description

On-Line Handwritten Character Recognition Method Using the Phoneme Automata

1 is a conventional character recognition processing flow chart.

2 is a block diagram of an online handwritten character recognition apparatus using the present invention autogram.

3 is a flowchart of an online handwritten character recognition method using the present invention autogram.

FIG. 4 is a process flow diagram for performing Hangul alphabet automata in FIG.

5 is a consonant automata flow chart of the present invention.

6 is a vowel automata flow chart of the present invention.

7 is a table of strokes and their identification codes defined by the present invention.

8 is a consonant chart showing the performance of the present invention automata.

* Explanation of symbols for main parts of the drawings

1: Tablet I / O 2, 4, 6, 8: Memory

3: preprocessing unit 5: feature point extracting unit

7: Stroke recognition part 9: Hangul alphabet automata

10: Hangul Recognition Department

The present invention relates to an online handwritten character recognition method using a phoneme automata. In particular, it is possible to improve the recognition speed and reduce the memory size in character recognition by defining 16 strokes and performing the phoneme automata. It relates to an online handwritten character recognition method using a phoneme automata.

The conventional character recognition processing flow chart includes a preprocessing step of removing unnecessary data from the character information input as shown in FIG. 1, extracting feature points from the preprocessed data, and extracting the extracted feature points. According to the stroke recognition step of recognizing the stroke and the data recognized by the stroke recognition in the above step, the most similar characters to the standard characters are recognized and recognized through the large classification by the number of strokes, the middle classification by the stroke matching, and the small classification by the fuzzy inference. It consists of a character recognition step of outputting the result.

When a user writes a text (some text information) on a tablet as if he or she writes it on a paper, it remembers the written text information. This input information contains information that is not necessary to actually recognize the text. It performs preprocessing that only processes information. Let's look at it.

To remove the hooks that occur at the start and end points of the pen when writing on the tablet, and the wrong sample points when writing, the process of removing all but the first sample point except the first sample point Eliminate unnecessary input of information. The pre-processing information is used to extract the feature points necessary for recognizing the stroke using the already defined stroke and the six feature points, and by using the extracted feature points, the respective defined strokes are recognized and recognized. Character is recognized through the specified stroke.

In order to reduce the number of characters to be recognized, the character recognition classifies only candidate characters having the same number of strokes according to the number of strokes, and then performs a middle classification that performs stroke matching among candidate characters selected in the major classification process. The final recognition result is obtained by using the relative position information between strokes among candidate characters extracted from the middle classification to extract the final recognition result in the small classification process. Fuzzy inference is used here.

In this way, only the sperm and a few squiggles can be recognized. However, in the conventional stroke matching method, 2350 characters of standard Korean must be created as a database, which takes up a lot of memory and a large amount of computation when the number of strokes increases due to the introduction of fuzzy inference. There was a problem that the recognition speed was falling.

In view of such a conventional deficiency, the present invention recognizes which type of stroke the character information input by the pen is, and arranges the recognized stroke according to a predetermined rule to recognize the phoneme. An online handwritten character recognition method using a phoneme automata to recognize what type of text is input information is described below. Referring to the accompanying drawings, the following description is made.

2 is a block diagram of an online handwritten character recognition device using the present invention automata, a memory 2 for storing data (information) input through the tablet I / O unit 1, and the memory 2 as shown in FIG. Is used for stroke recognition in accordance with the preprocessing section 3 which removes unnecessary information at the time of actual recognition from the information stored in the memory, and selects and stores only the necessary information in the memory 4, and the stored information stored in the memory 4; A feature recognition unit (5) for extracting feature points, a stroke recognition unit (7) for recognizing strokes in accordance with storage information through a memory (6) for storing feature points extracted by said feature point extraction unit (5), and said stroke The information input in the order of the Hangul alphabets (9) for recognizing the phonemes by arranging the strokes recognized by the recognition unit (7) according to a predetermined rule, and the order of the phonemes recognized through the phonemes (9) It is composed of a Hangul recognition unit 10 for outputting to the tablet input and output unit 1 to recognize what letters.

Referring to the operation, effects of the present invention configured as described above in detail.

In the tablet input / output unit 1 of FIG. 2, data written from a pen down to a pen up by a user, that is, information of a stroke, is received and stored in the memory 2. Since the information stored in this memory 2 contains a lot of unnecessary information to actually recognize Hangul, the unnecessary information is removed and preprocessing is performed to find only the necessary information. Here, the distance filter is used. Sample points are selected.

This is because sample points at close distances are not meaningful for real recognition and cause only a large amount of computation. And the angle filter between the sample points is selected more than 45 ° in order to extract the bending point with a large change using the angle filter.

In addition, the hook (hook), which is a slight twist at the start point and the end point that significantly affects stroke recognition, is removed and stored in the memory 4.

The feature point is extracted using the information stored in the memory 4, and the stroke is recognized using the extracted feature point. The stroke recognition unit 7 defines 16 strokes, the defined 16 strokes being a seventh stroke. Same as the table shown in FIG.

Recognition strokes shown in FIG. 7 are stored in the memory 8. In the stored method, each stroke is stored so that each stroke can be input in stroke units in the Hangul alphabet. In the Hangul alphabet automata 9, the respective stages are input by the input stroke.

Recognize which type of stroke is the information inputted by the phoneme automaine pen, arrange the recognized strokes according to a predetermined rule, recognize the phoneme, and input information by the order of the recognized phonemes. Refers to the recognition of cognition.

Here, the flowchart of the Hangul alphabet automata 9 is shown in the consonant automata flow chart in FIG. 5 and the vowel automata flow chart in FIG. 6, in which numbers refer to values of recognized strokes, and q means each stage.

Now, the Hangul recognition sequence by performing the automata will be described with reference to FIG.

In FIG. 2, when a stroke is input to the Hangul alphabet automata 9 through the stroke recognition unit 7 and the memory 8, the remaining stroke is checked after the initial consonant is recognized by the consonant automata. By recognizing the neutrality, check the remaining stroke again, if there is a stroke, perform the consonant automata, and if there is the remaining stroke, it is the double consonant, so it performs the consonant automata and recognizes the Hangul. If not, it recognizes Hangeul and exits.

Referring to FIG. 5 and the stroke code table defined in FIG. 7 above, the stroke of # 0 is recognized (a) and the next stroke is # 0, # 14, and # 6. If a stroke other than comes, it is recognized as ＂ㄱ＂ first.

For example, after stroke # 0 is recognized, if stroke # 0 is too old, it is recognized as '0'. After stroke # 0 is recognized, stroke # 15 is recognized (in Fig. 7). If-(middle of the previous stroke) comes, it is recognized as ＂ㅋ＂. In addition, after stroke # 0 is recognized, stroke # 14 (-(below the previous stroke) in Fig. 7) comes and then stroke # 1 (b in Fig. 7) comes again. Recognize it as ＂ㄹ＂. Likewise, when stroke # 0 is recognized and stroke # 6 (＼ in Fig. 7) comes, and strokes other than # 0 \ are coming, the previous stroke is recognized as '＂' and then again '6' When the stroke comes, it is recognized as ＂ㅉ＂.

As mentioned above, if stroke # 0 is recognized first, and strokes other than # 0, # 14, and # 6 come, it is recognized as 'B' and stroke # 1 is If it comes, it will be recognized as ＂B＂. If stroke # 3 is coming, it will be recognized as ＂ㄹ＂. If stroke # 7 is coming, it will be recognized as ＂0＂. The next stroke # 2 is recognized. If a stroke other than # 2 is next, it is recognized as 'B' but if it comes back to stroke # 2, it is recognized as 'B'.

If stroke # 12 is recognized and strokes other than stroke # 1, # 14 \ and # 10 \ are received, it is recognized as #_ and the stroke # 12 comes and again If stroke # 1 comes, it is recognized as '＂' and if this stroke is recognized again, stroke # 15 (in Figure 7-(middle of the previous stroke)) is recognized as '＂ ㅌ ＂'.

If stroke # 8 is recognized and stroke # 0 and stroke # 14 continue, it is recognized as stroke. If stroke # 8 comes after stroke # 8, then stroke # 8 Recognize it as ＂.

In the same way as above, all consonants are recognized to recognize the consonants. When the consonants are recognized and the vowels are recognized, the vowels are recognized using the vowel automata and the stroke chart shown in FIGS. 6 and 7, as follows.

When stroke # 9 (the right side of the previous stroke) comes, it recognizes it. When the stroke # 13 follows (-(the right side of the previous stroke)), it recognizes it as. If stroke # 9 is recognized and strokes # 13, # 14, and # 9 continue, it is recognized as.

For example, when the character information of "guest" is inputted, the Hangul alphabet automata (9) first performs the phoneme automata and recognizes "0", and performs the next vowel automata, "9", "13", Recognizes # 14 and # 9, and performs consonant automata again to recognize # 0 and # 1, respectively, and outputs the Hangul recognition unit 10 sequentially. Accordingly, the Hangul recognition unit 10 recognizes the stroke according to the recognition code value shown in FIG.

If it is recognized in this way, all vowel recognition is possible. When the vowel recognition and the neutral recognition are completed, the consonant automata is performed again to recognize the finality, and if there is a remaining stroke, the consonant automata is performed to recognize the final consonant. Same as one.

In other words, the consonants of Korean characters appear in the consonant automata, which is not recognized by the consonant automata. The recognition of the consonants uses the number of strokes remaining after the neutral recognition. Run the automata. In this way, the consonant is recognized by combining the first consonant and the second consonant.

In other words, in order to recognize Hangul in Korean alphabet automata, it is necessary to first recognize consonants. Recognize consonants and use the rest of the stroke to recognize neutrals or vowels. This is also done by running a vowel automata. And Jong-sung recognizes super-neutrality, uses better strokes and performs consonant automata. If there are no strokes remaining after the super and neutral recognition, the letters end with neutral. If the recognized result is exported to the output as a combination code, the output will be represented in Hangul.

As described in detail above, the present invention can replace the keyboard in the existing desktop computer, and when the user writes the Hangul on the liquid crystal tablet as if he or she writes a simple memo on paper, the memory is stored in the memory device. Compared to the stroke matching method, the memory size can be reduced, and the Hangul alphabet recognition method can be implemented to increase the recognition speed by reducing the calculation amount.

Claims (2)

A preprocessing step of removing unnecessary data at the time of recognition and selecting only necessary information when data information inputted through the block input / output unit 1 is input, and a feature point extracting step of extracting feature points according to the preprocessed data information; A stroke recognition step of recognizing a stroke according to the feature point extracted in the step; a consonant automata performing step of recognizing the phoneme by performing consonants and vowel automata for initial, neutral, and final according to the stroke input recognized in the step; And a Hangul recognition step of recognizing the Hangul according to the recognition code value of the stroke recognized in the step and outputting it to the tablet input / output unit (1). The method of claim 1, wherein the phoneme automata performing step comprises: a consonant recognition step of performing consonant automata when a stroke is input, recognizing a consonant; a stroke checking step of checking whether there is a remaining stroke recognized in the step; A neutral recognition step for recognizing neutrality by performing vowel automata if there is a stroke remaining in the step; and a final recognition step for recognizing finality by performing consonant automata if there is a stroke after neutral recognition in the step; If there is no end, and if there is a remaining stroke, the consonant automata performed by the consonant autorecognition step to recognize the consonant consonants, characterized in that the online handwritten character recognition method using the phoneme automata.