KR100367580B1 - Device for recognizing on-line character of stroke order independence - Google Patents

Abstract

PURPOSE: A device for recognizing an on-line character of a stroke order independence is provided to improve an efficiency of a memory calculating speed by recognizing written character being inputted at different orders using a database inputted by one exact writing order , thereby preventing an appendix of new database. CONSTITUTION: When a stroke of a character is inputted by a copyist, a distance filtering unit(10) removes duplicated dots and dots existing within a predetermined distance. An area separation unit(20) separates an area by searching a starting dot of a stroke shape variation with respect to input data through a distance filtering. A characteristic extracting unit(30) extracts a characteristic value with respect to the separated area. When the extracted characteristic value is compared with strokes inputted in a database having characteristic values of reference characters, a recognition unit(40) recognizes characters, and outputs character codes of the recognized character.

Description

Independent online character recognition device

The present invention relates to an on-line character recognition device capable of recognizing characters irrespective of stroke order, in particular, even though the data is input in the correct stroke order and looks the same as this data, but is input in different stroke orders. The present invention relates to an independent on-line character recognition device capable of recognizing characters in comparison with the correct stroke order data.

In the conventional learning type character recognition device configuration, as shown in Figure 1, the pre-processing unit (1) for removing the point of noise and noise input from the input points when entering the text through the tablet, and the pre-processing A feature extractor 2 for extracting features determined for character recognition with respect to the input points transmitted through the unit 1 and a feature value extracted by the feature extractor 2 are stored in the database 4. It is composed of a recognition unit (]) that recognizes the same stroke compared to the reference value and combines the recognized strokes to form letters.

Looking at the prior art configured as described above in detail.

When the user writes letters through the tablet, the preprocessor 1 receives the x and y coordinate values of each stroke of the letter, and then removes the meaningless points from the input contacts first to make the recognition quick and accurate. After removing the input point and outputs to the feature extraction unit (2).

Accordingly, the feature extractor 2 extracts the predetermined features for character recognition from the input points from which the noise is removed from the preprocessor 1. The feature extractor 2 changes the stroke direction or the distance between the input points. Will be used.

The recognition unit 3 receiving the feature values extracted from the feature extraction unit 2 determines and recognizes which one is similar to the reference value of the database 4 in which the reference stroke is already stored. After combining the letters, make the output.

However, in the conventional technology as described above, the database in which the reference stroke is stored in advance for the character recognition is made based on the stroke order, so that the stroke order is incorrect even if the user writes the same letter through the tablet. This makes it impossible to recognize, and for this reason, the same character must have a database of characters inputted in different stroke order, thereby causing a problem of inefficiency in memory waste and calculation speed.

Therefore, an object of the present invention is to recognize the handwriting inputted in one order with the database inputted in one correct order, and to prevent the addition of a new database to increase the efficiency of memory and calculation speed. It provides a recognition device.

The apparatus of the present invention for achieving the above object is, as shown in Figure 2, the distance filtering means for removing the overlapping points and the points existing within a certain distance when inputting the stroke of the character by the scriber, distance filtering Region separation means for finding a point where a change in the shape of the stroke starts with respect to the input data which has passed through the data, and feature augmentation means for extracting the feature value for the separated region; As a recognition means for recognizing not only the characters input in a certain stroke order but also the characters input in other stroke orders and outputting the character code of the recognized characters when comparing the input strokes with the database where the feature values of the standard characters are already stored. Configure.

When described in detail with respect to the operation and effect of the present invention configured as described above.

When a writer writes a text through a tablet, strokes of the text are input to the distance filtering unit 10 one by one, and the distance filtering unit 10 removes the overlapping points and the points existing within a certain distance.

At this time, the distance value, which is the standard of distance filtering, may vary according to the resolution and the stroke size. Generally, when the resolution is high and the stroke size is large, a large value is used.

When distance filtering is performed as described above, the effect of sampling the input points at regular time intervals regardless of the writing speed can be obtained, the change from the writing speed can be absorbed, and the number of input points can be reduced. Can reduce the aroma.

When the input data, as shown in (a) of FIG. 3, which has been filtered by the distance filtering unit 10 is transferred to the area separator 20, the area separator 20, as shown in (b) of FIG. It is divided into three vice versa, and the domain separation is separated when the following three conditions are satisfied.

1) When another stroke is input

2) The cumulative angle is 360 degrees

3) the stroke changes its angle from clockwise to counterclockwise, or vice versa.

As shown in (b) of FIG. 3, when divided into three regions and outputted to the feature extractor 30, the feature extractor 30 determines feature values for each region. Consists of information.

Information about the size of the area: i) RT, ii) RB

In characterizing information about the size of an area in an input stroke, the ratio of the size of the area given in the input stroke is represented by a value. That is, the value is calculated by the following equation, and then the magnitude is quantized to an integer between 0-15 and normalized.

C _xmin , C _ymin : Smallest X, Y coordinates in the input stroke

C _xmaxm , C _ymax : Largest X, Y coordinates in the input stroke

R _xmin , R _ymin : Smallest X, Y coordinates in the area

R _xmax , R _ymax : Largest X, Y coordinates in the area

Information about the beginning and end of the area: iii) RS, iv) RE

A value for characterizing the location of the entire stroke that characterizes the selected region, which characterizes where the starting and ending points of the region are located in the overall stroke to characterize it. Quantize and normalize to an integer between

R _xs , R _ys : the coordinates of the starting point in the area

R _xe , R _ye : the coordinates of the end point in the region

Information about the direction of stroke in the area: v) RC

Characterized in the written direction of the stroke in the region, the feature value when the stroke in the region is a straight line uses an eight-way code that separates the direction between the start and end points of the stroke in the region at 45 degree intervals. When the stroke in the turning area is a curve, the feature values are assigned 8 for clockwise rotation and 9 for counterclockwise rotation.

As a result, the characteristic values of one region are defined as RT, RB, RS, RE, and RC.

When recognizing data made of multiple strokes, the area is separated for each stroke, and each feature value is designated according to the input stroke order. That is, if the first stroke is divided into three regions, each feature value is composed of F ₁₁ , F ₁₂ , F ₁₃ , the second stroke is divided into two regions, and the feature value is composed of F ₂₁ , F ₂₂ , The feature value of the input consisting of two strokes is represented by five feature values of F = {F ₁₁ , F ₁₂ , F ₁₃ , F ₂₁ , F ₂₂ }.

When the feature value represented by the five feature values described above is transferred to the recognizer 40, the recognizer 40 has already stored the feature value of the reference character of the database 50 and the feature value of the input stroke. By calculating the dissimilarity of, the reference character with the lowest dissimilarity is output as a recognized character.

In case of recognizing the strokes input by the recognition unit 40 in the state of segmentation and making a feature value, the comparison algorithm is composed of a database consisting of the same number of regions as the total number of input strokes. Because of the same number of feature areas, character recognition can be accomplished with simple calculations.

Among the feature values, a comparison is first made on the information RC on the direction of the stroke in the region.

If the result of the comparison indicates that there is a difference in direction over a certain value, it is regarded as a different character and ignored.

When the number of input strokes is n, when the feature value of the i-th region of the input stroke is F _ci and the feature value of the reference character is F _rc , the dissimilarity E _i is calculated as follows.

F _ci = {cRT, cRT, cRT, cRE, cRC}

F _ri = {rRT, rRB, rRS, rRE, rRC}

The average dissimilarity is the average of the above values by the number of regions.

Therefore, a candidate character is one whose average dissimilarity of the reference character is equal to or less than a predetermined value, and a character having the smallest average dissimilarity among the candidate characters is a recognition character.

Recognition in the above manner is a process of recognizing characters written in order of stroke.

If the recognition of the stroke sequence fails, the stroke sequence is assumed to be changed, and the stroke sequence independent character recognition process is performed. This recognition process is performed between the input stroke and the reference character having the same number of areas as shown in FIG. The first step is to find the strokes of the same number of strokes in the text, calculate the dissimilarity between the strokes, and select the most similar strokes to determine the paired strokes. Find the average dissimilarity.

The stroke order independent character recognition process selects one stroke from the first input stroke and selects all strokes having the same number of regions in the database.

Second, calculate dissimilarity for each stroke.

Third, select the pair with the smallest dissimilarity in pairs and repeat the first procedure for the other input stroke.

Fourth, if all pairs are determined, the dissimilarity is calculated.

Therefore, as in FIG. 4, (S1, S'1), (S2, S'2), (S3, S'3), and (S4, S'4) are determined in pairs.

As described in detail above, the present invention enables recognition of written characters in the order of stroke and indefinite stroke, thereby preventing mistakes of the writer, improving recognition speed, and improving efficiency in memory usage. It works.

1 is a block diagram of a conventional learning type character recognition device.

2 is a block diagram of an independent on-line character recognition device of the present invention.

FIG. 3 is an explanatory diagram in FIG. 2 in which the input stroke is divided into three zeros.

(A) is the result of distance filtering of input stroke,

(B) is the result of domain separation.

4 is an explanatory diagram showing a stroke independent character recognition process in FIG.

******* Explanation of symbols on the main parts of the drawings *******

10: distance filtering unit 20: area separation unit

30: feature extraction unit 40: recognition unit

50: database

Claims (3)

Area filtering means for removing overlapping points and points existing within a certain distance when writing strokes of characters by handwriting, and finding the point where the shape of stroke starts for input data that has been filtered through the means. A region separating means for separating a word, a feature extracting means for extracting feature values for a region separated by the region separating means, and data in which the feature value extracted by the specific extracting means and the feature value of a reference character are already stored. In comparing the base and input strokes, on-line character recognition of independent stroke order, comprising recognition means for recognizing not only characters input in a certain stroke order but also characters input in other stroke orders and outputting a character code of the recognized characters. Device. The method of claim 1, wherein the area separating means comprises: 1) when another stroke is input; 2) when the cumulative angle is 360 degrees; 3) when the stroke changes the angle from clockwise to counterclockwise or vice versa. An independent on-line character recognition device, characterized in that the area is separated when satisfied. The apparatus of claim 1, wherein the recognition unit outputs the recognition character and the candidate character by measuring dissimilarity using the feature values of the input areas.