KR19990052967A - Korean Recognition Method Using Window and Projection Information - Google Patents

Abstract

The present invention relates to a Korean character recognition method using window and projection information. Conventionally, when characters are recognized using contour information, as shown in FIG. Often, false information is extracted, such as occurrence, and in this case, the recognition information and efficiency are deteriorated because the recognition time is consumed by using the projection information which is less sensitive to the result of separation of the phoneme and using it for recognition. . Therefore, the present invention compares a value obtained by calculating a cumulative black pixel in a horizontal direction of a document with a threshold value and a character string to separate a character string from a binary image input from a scanner. A second step of separating individual characters with respect to the second character; a third step of separating Hangul characters and vowel recognition from the characters separated in the second step; and a second step of separating the phoneme images separated in the third step into windows. Step 4, the fifth step of extracting the vertical and horizontal components for each window divided in the fourth step, and by using the components extracted in the fifth step is checked whether the phoneme recognition is possible to recognize the phoneme A sixth step of reducing the number of candidate letters by grouping other phonemes; and a seventh step of recognizing phonemes using projection information on the candidate letters grouped in the sixth step. And the eighth step of storing the result recognized in the sixth and seventh steps in the buffer, and extracting and determining only a part of the information obtained in the projection information extraction process in advance using the characteristics of the Hangul alphabet. Some phonemes that satisfy them export the recognition result immediately, while other phonemes are grouped into some extracted information to reduce the number of candidate characters to be compared, and then extract and recognize the remaining projection information to increase the speed and recognition rate. It is to be.

Description

Korean Recognition Method Using Window and Projection Information

The present invention relates to a Korean character recognition method using window and projection information. Particularly, if a specific condition is satisfied using only some information of the projection information extraction process, the recognition result is immediately output and the other characters are compared by grouping candidate characters. The present invention relates to a method of recognizing Hangul using window and projection information to improve the recognition speed and recognition rate by reducing and recognizing the remaining projection information.

In the high information society of the future, the amount of information must be generated, transmitted, and processed in an incremental manner.

For this purpose, an interface between a computer and a human is essential, and the present invention is limited to an input device of a computer among the interfaces, and the Korean language is used for a character / document recognition system in which a human can substitute an input using a keyboard directly. In order to implement the recognition method, the present invention will be described later.

1 is a flowchart illustrating a conventional Hangul recognition method. As shown in FIG. 1, a character string is separated by comparing a threshold value and a value obtained by accumulating black pixels in a horizontal direction of a document with respect to a binary image input from a scanner. A first step, a second step of separating individual characters for each of the character strings separated in the first step, a third step of separating Hangul characters and vowel recognition from the characters separated in the second step; A fourth step of recognizing the phoneme separated in the third step, and a fifth step of storing the result recognized in the buffer.

Looking at the prior art made of each step as follows.

When a binary image of a document is input through a scanner in a character recognition system, a cumulative black pixel in the horizontal direction of the document is obtained with respect to the input binary image.

By comparing the accumulated black pixel value and the threshold value, the upper and lower positions of the string are obtained, and the string is separated from the document.

After the string separation is completed in step S11, individual characters are separated for each of the separated strings.

From the individual characters separated in step S12, the phoneme of Korean characters, i.e., primary, neutral, and final, is separated and separated, and the phoneme separated from the previous character is sent to the phoneme recognition unit (S13).

Then, the phoneme recognition unit extracts the extracted feature of the phoneme image and how to use the same in the recognition process. Contour information, such as the number of blocks, and projection information for determining the presence of horizontal, vertical, and diagonal components are used.

As described above, the phoneme recognition unit recognizes the phoneme image using the contour information and the projection information (S14), and then generates a Hangul character from the recognized results (collection and superspecies) and stores it in the buffer. (S15)

However, in the prior art as described above, when the character is recognized by using the contour information, incorrect information such as loss of hall information and excessive occurrence of unnecessary concave may be extracted by incorrectly recognizing "and" as shown in FIG. 2. Frequently occurring, in this case, there is a problem in that recognition time is consumed because recognition time is consumed by re-extracting and using the projection information which is less sensitive to the result of the separation of the elements.

Accordingly, an object of the present invention for solving the conventional problems as described above is to output a recognition result immediately after satisfying a specific condition using only some information of the projection information extraction process using the characteristics of the Hangul alphabet, and grouping the remaining characters. By reducing the number of candidate characters to be compared and extracting and recognizing the remaining projection information, the present invention provides a method for recognizing Korean characters using window and projection information to increase recognition speed and recognition rate.

Another object of the present invention is to provide a Hangul recognition method using a window and projection information to use a large classifier to reduce the number of candidates in the extraction process using projection information.

1 is an operation process diagram for a conventional Hangul recognition method.

2 is an explanatory diagram showing an error state when "wa" characters are extracted using contour information;

Figure 3 is an explanatory diagram showing a process of performing a window process when the phoneme input in the present invention.

4 is a flowchart illustrating a method of recognizing a Hangul using a window and projection information of the present invention.

The present invention for achieving the above object is a first step of separating the character string by comparing the threshold value and the value obtained from the accumulated black pixels in the horizontal direction of the document with respect to the binary image input from the scanner, and in the first step A second step of separating individual characters for each of the separated character strings, a third step of separating Hangul characters and vowel recognition from the characters separated in the second step, and a phoneme image separated in the third step The fourth step of dividing into nine windows, the fifth step of extracting the vertical and horizontal components for each window divided by the fourth step, and the component extracted in the fifth step is checked to determine whether the phoneme recognition is possible. The sixth step of reducing the number of candidate characters by recognizing the phoneme and grouping other letters is possible, and using the projection information on the candidate characters grouped in the sixth step. And a eighth step of recognizing the phoneme, and an eighth step of storing the result recognized in the sixth and seventh steps in a buffer.

Hereinafter, the present invention will be described in detail with reference to the accompanying drawings.

4 is a flowchart illustrating a method of recognizing a Hangul using a window and projection information according to an embodiment of the present invention. As shown therein, a value and a threshold obtained by calculating a cumulative black pixel in a horizontal direction of a document with respect to a binary image input from a scanner A first step (S21) of separating the character strings by comparing values, a second step (S22) of separating individual characters for each of the strings separated in the first step (S21), and the second step ( A third step (S23) of recognizing the Hangul alphabet and the vowel from the character separated in S22), a fourth step (S24) of dividing the phoneme image separated in the third step (S23) into a window, and The fifth step (S25) for extracting the vertical and horizontal components for each window divided in the fourth step (S24), and by checking whether the phoneme recognition is possible using the components extracted in the fifth step, the phoneme recognition is possible. Recognize and other groupings A sixth step (S27) of reducing the number of candidate letters, a seventh step (S29) of recognizing the phoneme of the candidate characters grouped in the sixth step (S27) using projection information, and the sixth step ( S27) and the eighth step S30 of storing the result recognized in the seventh step S29 in the buffer.

Referring to the operation and effect of the present invention configured as described in detail as follows.

When a binary image of a document is input through a scanner in a character recognition system, a cumulative black pixel in the horizontal direction of the document is obtained with respect to the input binary image.

By comparing the accumulated black pixel value and the threshold value thus obtained, the upper and lower positions of the character string are obtained, and the character string is separated from the document.

After the string separation is completed in step S21, individual characters are separated for each of the separated strings.

Recognize vowels after separating them into individual letters, that is, initial, neutral, and final, from the individual characters separated in step S22.

In this way, the phoneme image of the separated Hangul is divided into windows. (S24)

That is, when the image of the phoneme is inputted, as shown in FIG. 3, after dividing into nine windows, the horizontal (-), vertical (-), diagonal (1), and diagonal (2) angles of the respective windows are shown. Extract the components of and use this information as a feature of each image.

Therefore, when one image comes in, four pieces of information are extracted from nine windows.

However, the method used in the present invention, once the image is entered, only the presence of the horizontal component for all the windows to confirm the vertical in the windows (1,4,7), (3,6,9) in FIG. Extract the ingredients. (S25)

Next, we check for the presence of empty spaces between the horizontal components only if all the horizontal components exist in the nine windows.

In this process, images satisfying the following conditions are exported with their respective recognition results.

First, window 1,2: presence of horizontal components

Windows 3: Image Exists

Windows 6: 9: vertical component present

Remaining windows: If there are a few, it is recognized as "ㄷ".

Second, window 1,4: vertical component present

Windows 7: Image Exists

Windows 8, 9: Horizontal component present

Remaining windows: If there are a few, it is recognized as "b".

Third, window 2, 3, 8, 9: horizontal component present

Windows 1,7: Image Exists

Windows 4: vertical component present

Remaining windows: If there are a few, it is recognized as "ㄷ".

Fourth, there is a horizontal component in every window

Empty Margins Between Windows 1,2 and Windows5

If there is a blank space between windows 5 and 6 and windows 8 and 9, it is recognized as "ㄹ".

Fifth, there is a horizontal component in every window

Empty Margin Between Windows 2,3 and Windows 5,6

If there is a blank space between windows 5, 6 and 8, 9, it is recognized as "E".

As described above, if the phoneme satisfies the five conditions is immediately recognized (S29), the images that do not satisfy all five conditions group the characteristics of the extracted information (S27).

The grouped phonemes except for the phoneme that has undergone the phoneme recognition above use a large classifier to reduce the number of candidate characters to be compared.

Then, using the projection information to recognize the rest of the phonemes (S29).

The phoneme thus obtained is stored in a buffer (S30).

In this way, the part of the phoneme that satisfies a specific condition by extracting and determining only a part of the information obtained in the process of extracting the projection information using the characteristics of the Hangul phoneme is immediately outputted as a recognition result, and the other phonemes are grouped into some extracted information. After reducing the number of candidate characters to be compared and extracting and recognizing the remaining projection information, the recognition speed and recognition rate are increased.

Accordingly, the present invention has the effect of increasing the recognition speed and the recognition rate by allowing the projection to be recognized using some projection information and a window.

Claims (2)

The first step of separating the character strings by comparing the threshold value and the value obtained by calculating the cumulative black pixels in the horizontal direction of the document with respect to the binary image input from the scanner, and the individual character for each character string separated in the first step A second step of separating, a third step of separating Hangul characters and vowel recognition from the characters separated in the second step, a fourth step of dividing the phoneme image separated in the third step into a window, The fifth step of extracting the vertical and horizontal components for each window divided in the fourth step, and by using the components extracted in the fifth step to check whether the phoneme recognition is possible, the phoneme recognition can recognize the phoneme and other phonemes A sixth step of reducing the number of candidate characters by grouping, a seventh step of recognizing the phoneme of the candidate characters grouped in the sixth step using projection information, and the sixth step Hangul recognition method using the window and projection information, characterized in that comprising the steps of claim 8 to store the result of recognition in step 7 in the buffer. The method of claim 1, wherein the window is divided into nine windows in the fourth step.