KR100286709B1 - Method for separating ideographic character in alphabetic string - Google Patents

Abstract

PURPOSE: A method for separating an ideographic character in an alphabetic string is provided to rapidly and accurately recognize the ideographic character by accurately and simply abstracting the ideographic character without the interference of a character recognition unit. CONSTITUTION: A character string is inputted(ST30). An ideographic character is abstracted from the character string inputted(ST31). A small letter is separated from the ideographic character abstracted(ST32). An outline is tracked from the character(ST33). A capital letter is separated from the character(ST34). The character is classified using a feature of a hole. After that, the ideographic character separated is abstracted(ST35). A height and a margin of the character abstracted are detected. After that, the height and the margin are inputted to a recognition process(ST36).

Description

How to Separate Individual Characters in English Strings

The present invention relates to an individual character truncation in an English character string in a document recognition apparatus. In particular, even when characters are in contact with each other, the character string is precisely cut out without interference of the character recognition means to recognize a document more quickly. It is about a method of separating individual characters.

As a method for separating the input contact characters in the conventional document recognition apparatus, a plurality of separable points are selected from the contact characters and the recognition process is performed after each point. Will be cut off.

That is, as in the first diagram, when the letters "O" and "C" are input in contact with each other, the cut candidate positions are "2" and "3".

Therefore, if you cut the cut point "2" and recognize it, "O" is recognized as "C". Since this "C" is one of the English alphabets, the cut to the cut point "2" is considered appropriate and the cut is performed. To be recognized.

However, since the cutting of the contact character in the conventional document recognizing apparatus depends on the recognition, if the recognition is wrong, the individual character truncation, which is a higher process of the recognition, may be affected by the recognition, and the recognition and the truncation are mutually repetitive. Since the process takes a lot of processing time was a problem.

Accordingly, an object of the present invention is to accurately recognize a document by quickly cutting individual characters without interference of character recognition means even when the characters are in contact, as well as by simply cutting out contact characters using hole features. The purpose of the present invention is to provide a method of separating individual characters in an English character string that can recognize individual characters quickly and accurately.

The technical method of the present invention for achieving the above object is a method of scanning a document to binarize a document image, separating a picture region and a character region from the binary image, and extracting a character string from the character region to recognize a character. An individual character extracting step of extracting individual characters from the extracted character strings; Character classification step of distinguishing the lowercase letters for the individual characters of the cut out; An outline tracking step of tracking an outline of the touched character among the cut out characters; Character discriminating step of distinguishing the uppercase letters for the contact character; A character truncation step of cutting individual characters distinguished in the character discrimination step by using hall features; A margin detection step of obtaining a height of the cut individual character and obtaining a margin; And a recognition and post-processing step of recognizing the cut individual character and correcting a misrecognized portion of the recognition.

1 is an exemplary view for separating a conventional contact character into individual characters,

2 is a block diagram showing a separate character separation system of the present invention,

3 is a flowchart of recognition of a document image according to FIG.

4 is a detailed flowchart of the process of cutting individual characters of FIG.

5 is a detailed flowchart for obtaining the holes of FIG.

6 is an exemplary view showing the length of the black pixel run and the position of the black pixel in the character cutting step of FIG.

FIG. 7 is an explanatory diagram showing character fusion conditions in the character margin detection step of FIG.

<Explanation of symbols for main parts of drawing>

1 scanner 2 scanner interface means

3: host computer 4: host interface means

5: digital signal processing means 6: buffer member

7: DRAM control means 8: program memory

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

2 is a block diagram of a separate character separation system according to the present invention. As shown therein, an interface means for interfacing the binary image data of the scanner 1 with a scanner 1 for scanning a document and generating a binary image (2), digital signal processing means (5) for processing the image data transferred from the interface means (2) in accordance with the program execution of the program memory (8) and storing and outputting the data in the data memory (9); A buffer member 6 for buffering the data processed by the signal processing means 5, DRAM control means 7 for controlling and storing the image data output from the buffer member 6 by controlling the image memory 10; A host computer (3) for reading the output data of the buffer member (6) and controlling the entire system operation of the document recognition device in accordance with the control signal of the DRAM control means (7), and the control signal of the host computer (3). Interface Constitute a digital signal processing means 5 and the buffer member host interface means (4) for controlling (6).

FIG. 3 is a flow chart of recognition of a document image according to FIG. 2. As shown therein, a scanning process for generating a binary image from an input document (ST100) and a picture portion and a character portion are distinguished from the input image. An area division and string extraction process (ST200) for separating a character string for the character portion, an individual character extraction process (ST300) for extracting individual characters from the separated character string, and recognition for recognizing each of the individual characters extracted A process ST400 and a post-processing process ST500 for correcting a part misrecognized in the recognition process.

4 is a detailed signal flow diagram of the individual character cutting process ST300 of FIG. 3 to separate individual characters from a character string separated in the area division and string extraction process ST200 and input through a string input step ST30. An individual character cutting step (ST31), a character classification step (ST32) for distinguishing lowercase letters for the individual characters being cut out, an outline tracking step (ST33) for tracking an outline for the contacted character, and the contact character A character discriminating step (ST34) for distinguishing capital letters with respect to a letter, a character cutting step (ST35) for cutting individual letters distinguished in the character discriminating step (ST34) using hole features, and obtaining the heights of the individual letters cut out; A text margin detection step ST36 is performed to obtain a margin and input it into the recognition process ST400.

Thus, the operation and effect of the present invention made in detail with reference to Figure 2 as follows.

First, the input document is read by the scanner 1 under the control of the host computer 3 to store image information in the image memory 10.

Using this information, the digital signal processing means 5 separates the character strings according to the area division program of the program memory 8 and stores the character string information in the data memory 9.

Then, by using the individual character cutting program stored in the program memory 8, the digital signal processing means 5 performs an operation of cutting individual characters in the flow of FIG.

That is, in the individual character truncation step ST31 of FIG. 4, the black pixel density in the vertical direction is obtained for the character string input from the character string input step ST30, and the character is cut when the number of the black pixels is smaller than the threshold value.

The threshold here is "1".

An attribute of the cut block is determined using the height of the character string and the width / height information of the cut block for the cut character.

For example, if the height of the string is shgt, the height of the cut block is chgt, the width is cwid, and the maximum vertical run length of the string is max.

shgt-max <5 and cwid> max × 3/4

If shgt-max> = 5 and cwid> max × 4/5, the truncation block is marked as a contact character.

When the individual character cutting step ST31 is completed, the next process of cutting the contact character is performed.

That is, in the character classification step ST32, it is a step of determining whether a block classified as a contact character is a lowercase m or w.

The reason for distinguishing them is that they are misclassified as contact letters because their width is larger than other letters, but if the lowercase m and w are misclassified, they should not be cut, so the process of character classification step (ST32) is necessary. Done.

Here, the lowercase m is regarded as the lowercase m if there are three strokes in the horizontal direction in the middle height of the block, and if there is a continuous length of black pixels with a difference of 3 or less in height from the lower left quarter of the character width.

Here, the case of the lowercase letter w is as follows.

The character of the lowercase letter w is determined by the presence of two bends in the upper part.

The contact character existing after the process of the character classification step ST32 is the contour tracking step ST33 which cuts out the character by contour tracking.

In the above, the contour tracking is performed in the counterclockwise direction, and in the process, it is possible to cut off serif-type letters which cannot be cut at the vertical black pixel density.

As an example, in the case of OB, the left and right sides of B are squeezed, and when viewed in the vertical direction, the spool and the right side of O are not overlapped and separated, but since they constitute different letter clusters, they are cut according to the contour trace. Is possible.

Next, the process of character classification step 500 is performed.

This part is a routine that recognizes and cuts I and T when the uppercase letters 'I, T' are in contact with other characters (especially uppercase letters). In words with only uppercase letters, I and T often come into contact with other letters.

In this case, to accurately cut them, it is most accurate to recognize and separate I and T.

Thus, in this routine, I and T are divided and cut using a simple feature.

I has a run that is similar to the height of the string. T has a vertical run that is similar to the height of the string, but I and T can be distinguished because the upper and lower shapes of these characters are different from I and T.

First, the distinction of I is as follows. If the vertical run, similar to the height of the string, has the coordinates of the start and end points x1 and x2, the shapes of the upper and lower sides of x1 and the right and lower sides of x2 are symmetrical, and if the length of the upper and lower horizontal runs is less than or equal to the threshold Truncate the letter by recognizing it as I Division of T If a horizontal line exists at the top of the irradiation area and the length of the horizontal line is symmetrical with respect to xq and x2, it is regarded as T and cut off.

Next, in order to cut the contact character using the hole feature in the character truncation step ST35, a hole must first be obtained. This process will be described with reference to FIG.

First, for each column i, the first black pixel position in the vertical direction, the last black pixel position and the length of the first black pixel run are compared (ST4), and if the condition is satisfied, the starting point h_s of the hole is set to I ( ST5), the starting point of the hole is obtained, and then the starting point and the ending point of the hole are obtained by making I, which does not satisfy the condition ST4, as the end position of the hole (H_1) (ST6).

Next, in step ST7, the number of holes is obtained by comparing the size of the hole with the length Smax of the largest black pixel run in the character string.

The minimum black pixel position and the like of the black pixel run used here are shown in FIG.

Once the hole is found, the contact character is cut off at the position with the minimum black pixel density between the holes.

The following is a process of detecting the height and the margin of an individual character extracted by the process of the character margin detection step ST36.

This process is to find the height of each block, fuse the wrong cut letters, and find the blank to separate the words.

The height of the block is the difference between the position of the first black pixel and the position of the last black pixel, and the erroneously truncated character fuses two blocks that are considered to have been cut into two blocks by using the spacing of adjacent characters.

The fusion condition indicates a case in which one block is incorrectly cut into two blocks b and c, as shown in FIG. 7, and these blocks must be fused.

If the spacing of the characters is b2, the spacing of these characters and the left and right characters are b2 and b3, and the widths of each character are w1, w2 and w3.

If b2 × 2 〈= b1 and b2 × 2 〈= b3 and w2 〈max (w1, w3) × 3/2 and the difference of the starting y coordinate of the b and c blocks is less than or equal to 3, the two blocks are merged into one block. Make.

Here's how to find the margins to separate words.

Assuming that the average space between characters is greater than 2 ave_blank, the height of the string is str_hgt, and the height of the character for margin extraction is bhgt, the height of the character bhgt is obtained as follows.

bhgt = max (str_hgt / 2 〈bhgt = 0 if no characters satisfy the height <str_hgt-3) (condition) of all characters in the string.

If the height of the character (bhgt) is found, then if the test margin (test_intv) is greater than (ave_blank * 3/2), then bhgt> 0 and test_intv> bhgt / 3, then test_intv = blank, bhgt = 0, and test_intv> str_hgt / 3 Then test_intv = blank.

As described above, each individual character cut out in the individual character cutting process ST300 is recognized through the recognition process ST400, and the character misrecognized in the recognition process ST400 is corrected and supplemented in the post-processing process ST500. To be recognized.

Therefore, in the present invention, even when characters are in contact with each other, it is possible to recognize the document more quickly by accurately cutting the individual characters without interference of the character recognition means, and by simply cutting the contact characters using the hall feature, It has the effect of accurately recognizing individual characters.

Claims (2)

In the method of scanning a document and binarizing a document image, separating a picture area and a text area from the binary image, extracting a character string from the character area, and recognizing a character, the individual character is extracted from the extracted character string. Individual character extraction step; Character classification step of distinguishing the lowercase letters for the individual characters of the cut out; An outline tracking step of tracking an outline of the touched character among the cut out characters; Character discriminating step of distinguishing the uppercase letters for the contact character; A character truncation step of cutting individual characters distinguished in the character discrimination step using hole features; A margin detection step of obtaining a height of the cut individual character and obtaining a margin; And a recognition and post-processing step of recognizing the cut individual character and correcting a misrecognized part of the recognition. 2. The individual character of the character string of claim 1, wherein the hole feature in the character truncation step is to determine the presence or absence of the hole using the position of the black pixel in the vertical direction of the target block and the length of the black pixel run. Separation method.