KR100470346B1 - The method for clustering an image of a character and the method for high-speed inputting and correcting a character by using the same - Google Patents

Abstract

The present invention includes the steps of comparing a segmented bitmap image; Grouping bitmap images recognized as similar characters through the comparing step; Displaying the grouped bitmap images; As the user enters corresponding text of text for an appropriately grouped image from among the displayed images, inputting corresponding text of text into each of the appropriately grouped images and an inappropriately grouped image selected by the user. To a method of clustering an image, comprising the step of causing them to be corrected. According to the method of the present invention, instead of repeatedly inputting a small number of characters with high occurrence frequency, the images representing the same characters are recognized as the same characters and grouped, thereby repeating the same characters through one input operation. By being able to omit a large number of text input operations, it is possible to drastically improve the text input operations that are the basis for the computerization of Korean material.

Description

The method for clustering an image of a character and the method for high-speed inputting and correcting a character by using the same}

The present invention relates to a clustering method of an image and a method for fast input and correction of characters using the method. More specifically, the present invention uses the clustering method of the image that can efficiently cluster the images recognized as the same character, and converts the images recognized as the same character into text characters easily and quickly in one operation. , And, on the other hand, a method for effectively selecting and correcting inappropriately clustered images despite unequal characters.

Korean historical data recorded from the past is recognized for its academic value not only in the humanities but also in various fields. Therefore, computerization of these isolated data for academic research or in various fields is a very urgent and important task. In other words, by converting the original records such as ancient maps, ancient books, and ancient documents into a code system that can be understood by a computer and storing them in a file or a database, there is a need for the public to easily use them through a computer or the Internet. In general, the most important and basic task of computerizing Korean data is to convert the original Korean data into text. However, in order to computerize Korean data, it is necessary to convert images of a large amount of original materials into text characters corresponding to the images and input them.

For example, if you want to provide a service that can be easily accessed by the general public through the Internet, the representative document of the Joseon Dynasty, the representative literature of the Joseon Dynasty, first, users should be able to see the contents of the original records. A picture file should be provided that scans the original image for the chapter. However, simply providing a picture file scanned from the original image on the web does not satisfy the various needs of the user. Preferably, it should be possible to decipher the meaning of the letters used in the Annals of Joseon Dynasty, and also to be able to easily recognize how many times a particular phrase or word is used in the Annals of Joseon Dynasty. In order to provide such a high level of service, the computer system can understand the characters used in the Annals of Joseon Dynasty (generally most of the letters used in ancient books and ancient documents) and various additional information contents (KSC 5601, Entering text using Unicode, etc. must be preceded. Only when such text input is performed can the desired search or indexing work from the limited data and the in-depth study using the text become possible.

As mentioned above, it is very important and basic to input the contents of the records of Chinese characters as text, but these tasks are the vastness of the original materials, the difficulty of the written characters (usually Hanja or old Hangul), and the input tool. Problems such as inefficiency, the need for multiple corrections to incorrectly typed letters, require unnecessary labor, time, and effort. In addition, due to the nature of the task of manual input, the completeness of the input text data is difficult to guarantee as much as desired.

Background Art Conventionally, optical character readers (OCRs) have been widely used to decode characters in image form and convert them into text. However, OCR does not show its usefulness in the computerization of episodic data. The reason for this is as follows.

OCR basically assumes the establishment of a standard code system. In order to produce, store, and exchange information on a computer, codes and fonts must be unified. Therefore, countries around the world enact and promulgate national standard codes. Font is produced and used in application software. However, for efficient use of hardware and software resources, national standard codes are generally established in a universally reasonable form so that they can be used for everyday information exchange. In Korea, the current standard code system is a two-byte code system. According to KSC 5601, approximately 8,000 characters are designated as standard code sets. Of these, the Hangul complete form is 2350 characters, and the Chinese characters are 4888 characters. However, such limited standard codesets are not possible to computerize classical texts such as Korean literature. That is, in the classical literature, about 1.5 million combinations are used in the case of old Hangul or Hangul, and about 120,000 characters in the case of Chinese characters. In order to be able to recognize and read characters using OCR, standard codes This is because three must specify the number of characters. This problem is similar to the case of the Unicode scheme. In addition, OCR must go through the process of pattern recognition and learning. Pattern recognition and learning cannot be performed on characters that exceed the standard codeset, and characters that are unknown and appear only in classical literature. As a result, computerization of the Korean-language data is inevitably requiring a simple repetitive input operation by the inputter because it belongs to a special field to which the OCR cannot be applied.

Conventionally, the text input operation of the Korean traditional data has been a method of inputting textual characters corresponding to the unit characters of the original image in a one-to-one manner using a word program from an image of the original document. However, this input work is a simple and repetitive work and can not expect a breakthrough efficiency. In addition, the Korean-language data, etc. include a number of difficult-to-read characters in the letters used therein, which is not only easy to input, but also has to be accompanied by a number of precise calibration procedures after input. That is, since the level of understanding of the characters (kanji) differs depending on the inputter, the accuracy of the input text cannot be guaranteed when a large amount of data is input.

On the other hand, for the computerization of Korean data, a calibration system has already been developed and patented by the present inventors by the date control of Chinese characters (Korean Patent Publication No. 2001-15963). According to him, a method is disclosed in which different original images are recognized and input as the same character, and thus can be corrected efficiently. However, the Korean Patent Application Publication No. 2001-15963 discloses a method of efficiently grouping character images recognized as the same and efficiently converting the character images identified as the same among the grouped character images into text characters corresponding thereto. It does not mention. That is, even with the above-mentioned Patent Publication No. 2001-15963, it is inevitable that the text input operation must be repeatedly performed for each character in the computerization of the Korean language data.

If you look at the characters, especially Chinese characters, that appear in the Chinese literature, you can see that there are more than 90% of Chinese characters. Therefore, it is pointed out that repeatedly inputting Chinese characters with a high frequency of occurrence is not only very inefficient, but also significantly delays the speed of the entire input process. In addition, Chinese characters are a kind of ideographs, so it is very difficult to input them, and it is impossible to input them unless the exact number and number of strokes are known. In addition, even if the number of strokes and the number of copies are known, the probability of incorrect input is high, and an unnecessarily large amount of time and effort must be taken to correct the incorrectly inputted Chinese characters.

Therefore, it is very urgent to develop a groundbreaking method for minimizing unnecessary repetitive text input operations in computerization of Korean history data. In order to overcome the backwardness of the task of repeatedly inputting a small number of characters with a high frequency of occurrence every occurrence of the present invention, the present inventors use a program or a method capable of recognizing images representing the same character as the same character. The present invention has been made in light of the fact that it is possible to omit repetitive multiple text input operations for the same character through the operation.

One technical problem to be achieved by the present invention is to provide a useful method for supporting the computerization of Korean data effectively.

Another technical problem to be solved by the present invention is to provide an effective method for minimizing the input work that must be repeated unnecessarily in the textual work of the Korean material.

1 is a screenshot showing a step of opening a list of image files.

2 is a capture screen showing a step of selecting an image file to be segmented.

3 is a capture screen showing a step of giving an instruction of automatic segmentation.

4 is a capture screen illustrating a step of selecting a segmentation direction according to a document image.

5 is a screenshot showing a step of correcting a wrong segment.

6 is a screenshot illustrating a step of generating a PSD file by terminating a segmentation operation.

7 is a diagram illustrating an example of a bitmap image.

FIG. 8 is a diagram illustrating a state in which bitmap images are collected based on a bottom line of a segment.

9 is a screenshot showing a step of generating a LIK file.

10 is a diagram illustrating a step of setting image comparison criteria.

11 is a diagram illustrating a step of reading an ISB file generated through a bitmap generation program.

12 is a diagram illustrating the progress of grouping.

13 is a capture screen showing the step of opening a LIK file.

14 is a capture screen illustrating a step of simultaneously displaying a character image and a bitmap image corresponding to a text input object on a screen.

15 is a screenshot showing the step of selecting different characters.

16 is a capture screen illustrating the step of storing text input information in a PSD file.

17 is a flowchart illustrating a method for fast input and correction of characters according to the present invention.

In order to achieve the above technical problem, the present invention, comparing the segmented bitmap image; Grouping bitmap images recognized as similar characters through the comparing step; Displaying the grouped bitmap images; As the user enters corresponding text of text for an appropriately grouped image from among the displayed images, inputting corresponding text of text into each of the appropriately grouped images and an inappropriately grouped image selected by the user. Providing a clustering method for the image, comprising the steps of correcting the correction.

In order to achieve the above another technical problem, the present invention comprises the steps of generating a scanned original image; Segmenting the generated image; Converting the segmented image into a bitmap image; Correcting the segmented bitmap image to a constant size; Using the clustering method of the image, inputting text corresponding to each of the appropriately grouped, identical original images, and causing inappropriately grouped, unequal images to be corrected Provides a way to enter and calibrate.

In the clustering method of the image and the method of rapidly inputting and correcting characters according to the present invention (hereinafter referred to as the "method of the present invention" without distinction, except as specifically defined otherwise), the original image is first displayed. Requires segmentation work. Segmentation refers to segmenting the original images stored in the picture file by each unit character through scanning, and performing such segmentation in the method of the present invention deals with each unit character image rather than the entire image. This is because it is much easier. However, the method of the present invention does not merely segment the original image, but generates a file including an index, location information, and a place to input a code value for each segmented unit character. This step may be performed using a program developed by the inventor named Young-Nin Program, which will be described below.

The zero-in program performs segmentation for dividing the entire scanned image by unit characters included in the image. Although not necessarily a linear matrix, the entire image consisting of a matrix of characters is segmented by unit letters through a zero-in program, thereby simplifying the text input work that is the basis for computerization of the Korean traditional data.

Segmentation is performed by the character recognition module owned by Youngin program taking into account the shape of the character, the margin of space, and the size of the character. When the zero-in program is executed, simultaneously with segmentation of the entire image, a so-called PSD file including an index, position information, and a code value input location for the segmented unit letter image is generated. In a PSD file, an index indicates the order of characters in the entire image, and corresponds one-to-one with an index in an ISB file generated by a bitmap generation program later. In addition, the position information of the characters included in the PSD file serves as an element for determining the shape and size of the character image when creating a bitmap binary to be stored in an ISB file generated by a bitmap generation program. In addition, the PSD file contains a place to store the text of the image character, that is, the code value of the character. In the present invention, inputting text of a text corresponding to a segmented unit text image means storing a code value in a code value storage location in the PSD file. After all, a PSD file may be defined as a file including an index, location information, and a code value input location of each character when segmenting an entire image, such as a page of a piece of material, by character units. .

1 to 6 are screenshots for explaining segmentation performed by zero-in program, FIG. 1 shows a step of opening an image file list, and FIG. 2 shows a step of selecting an image file to be segmented. 3 shows a step of giving an instruction of an automatic segment, FIG. 4 shows a step of selecting a direction of a segment according to a document image, FIG. 5 shows a step of correcting a wrong segment, and FIG. 6 shows a segmentation operation. The steps to generate a PSD file are shown.

In the method of the present invention, the segmented image is converted into a bitmap image as the bitmap generation program is operated. The bitmap refers to a data structure that contains information about a rectangular image. The image constituting the bitmap consists of a plurality of lines, each of which consists of a plurality of pixels. That is, the bitmap image means an image expressed in bit units. A black and white bitmap (or mono bitmap) uses one bit to represent one pixel. Usually, the extension meaning bmp file is bmp.

The bitmap generation program is a program for generating a binary file of a bitmap image named ISB from a segmented image. Since the binary file of the bitmap image itself is not a bitmap image, but includes a binary array constituting the bitmap image, the bitmap image is generated and displayed on the screen in a later clustering process. Allows comparison of captured images.

In the method of the present invention, the bitmap generation program uses the positional information of the character image stored in the PSD file generated by the zeroin program to generate a PSD file segmented through the original image, that is, the zeroin program. It finds the position of the character from the original image used and creates an image bitmap as big as the size of the character, and stores the data forming the pixel of the bitmap in the ISB file. In the method of the present invention, the reason for generating the bitmap image for the unit character through the bitmap generation program is that, as described above with respect to the zero-in program, the processing of data becomes easier when the operation is performed in units of characters. to be. Generating a bitmap image for each unit character is an essential task to be performed later in performing a comparison step during clustering of images.

In general, the ISB file generated by the bitmap generation program is different from the PSD file generated by the zero-in program, except that the file name is the same. Thus, each ISB file is a bitmap file that has a one-to-one correspondence with a PSD file (only with extensions) having the same name.

In the clustering method of an image according to the present invention, when a segmented bitmap image is generated by the zeroin program and the bitmap generating program, the segmented bitmap image is then corrected to a constant size. This correction is intended to facilitate easy and accurate clustering of images later.

For the corrected segmented bitmap image, clustering of the image according to the present invention is performed. Hereinafter, a clustering method of an image according to the present invention will be described.

Clustering, commonly referred to as clustering, is one of the detailed techniques of data mining, which is a statistical term that describes the process of extracting useful information that is not easily revealed from large amounts of data. The process of grouping objects. The term "clustering" in the present invention shall be used to mean so-called broad clustering.

In other words, the clustering in the present invention, as well as the so-called clustering of grouping the same character image by comparing the bitmap image obtained by the bitmap generation program, in addition, the user to the appropriately grouped character image In accordance with the broad concept, as the corresponding text of text is entered, a corresponding text of text is entered into each of the appropriately grouped images, and the inappropriately grouped images selected by the user are corrected. do.

As described above, the step of correcting the bitmap generated by the zero-in program and the bitmap generation program to a constant size from the original image is described. As a result, the consensus image clustering is performed on the corrected segmented bitmap image. do.

By consensus clustering, as mentioned above, it means comparing the segmented bitmap images and grouping the images recognized with similar characters. That is, during the broad clustering phase, before the text input process. It refers to a step. According to the present invention, after the consensus clustering is completed, for each of the same character images that are appropriately grouped among the grouped character images, the corresponding character text is input, and an inappropriately grouped character image is selected and corrected. Proceeds subsequently.

The clustering step of consultation is a process of preparing basic data which is the basis for performing the subsequent inputting, selection and correction of text characters. According to an embodiment of the present invention, the basic data is an index file, which is named as a LIK file according to an embodiment of the present invention, and is generated from an ISB file generated through a bitmap generation program. That is, the negotiating clustering step reads bitmap binaries for each character from an ISB file and stores them in an array within a program, and uses them to compare monochrome bitmap files generated in memory with each other. Is implemented. This process is explained in more detail as follows.

First, it reads an ISB file, reads several bitmap binary files contained in the ISB file, and stores them in an array of structures in turn. Herein, the structure refers to one unit that stores each bitmap binary as one object and stores the index, file name, and bitmap binary of each object. After all the ISB files have been loaded and stored in the array, the first element in the array is compared to the last element in the second array. In this process, if an element of an array that is found to be the same as the first element is found, the information of the element of the array that is recognized as the same is stored in the structure of the first array, and the average of the binary array of the element is stored. , Deletes the elements of the array joined by the first array. Next, based on the elements of the second array, the third to last elements are compared with each other in the same manner as before. After that, the same process is repeated based on the third and subsequent elements. This process is repeated until the end of the array. The result of this comparison is that the final array is the grouped output, which contains the final information that will be stored in the LIK file later.

The specific principle of grouping similar bitmap images in the consensus clustering is to binarize the target image into black and white and compare the curve values of the mathematically calculated waveform from the binarized image.

According to an embodiment of the present invention, the principle of grouping similar images by comparing bitmap images is as follows.

First, a calculation is performed on a bitmap image corrected to a constant size. Bitmaps are represented on the computer by black spots and whites, with black spots being 1 and whites being binarized to zero. By applying this principle, bitmap images are collected in one direction with respect to one side. FIG. 7 illustrates an example of aggregated bitmap images, in which bitmap images are collected based on a left line of a unit segment. After collecting, the curve value of the waveform that borders black and white is mathematically calculated. That is, through mathematical calculations on individual segmented images, the curve values of the waveforms are calculated using an algorithm, and then similar images are grouped by searching for an equation having the curve values of the same waveform. In the present invention, the above process may be repeated a plurality of times using different reference lines to enable more accurate grouping. 8 is a graph of a bitmap image based on a bottom line of a segment. Similarly, a curve value of a waveform bordering a black and white boundary is mathematically calculated using an algorithm, and then a similar image having a curve value of the same waveform is generated. Can be found.

The inventors have found a surprising fact that the comparison of bitmap images applying this principle can group similar bitmap images very accurately. Although the present invention proposes a binarization method through black and white separation as an example, a method of determining similarity of an image composed of multiple colors through an algorithm capable of separating colors also belongs to the scope of the present invention.

In the following, the process of actually performing the clustering of negotiations will be described with reference to FIGS. 9 to 12.

9 illustrates a step of generating a LIK file, and FIG. 10 illustrates a step of setting image comparison criteria. This step is a step of determining a criterion for determining whether a program is the same image when comparing two images, and may differently determine the criterion appropriately according to the characteristics of the object on which image clustering is performed. 11 illustrates a step of reading an ISB file generated through a bitmap generation program. 12 shows the progress of grouping. At this time, if you want to cancel this grouping, click the Cancel button, while if you want to proceed, click the OK button after this process is finished.

In the method of the present invention, the bitmap images grouped through the comparing step are displayed, and the texts corresponding to the appropriately grouped images among the displayed images are input. The character text corresponding thereto is input to each of the displayed images.

In the present invention, a method for displaying a grouped bitmap image may be selected within a range without departing from the object of the present invention among known methods, and various interfaces may be adopted in consideration of the user's convenience.

In the method of the present invention, after the negotiation clustering is completed as described above, the text input process is subsequently performed.

The text input process uses the LIK file created and stored in the previous clustering process. 13 to 16 are capture screens illustrating a process of inputting text using a LIK file.

FIG. 13 illustrates a step of opening a LIK file, and FIG. 14 illustrates a step of simultaneously displaying a character image and a bitmap image corresponding to a text input on a screen. The steps shown in FIG. 14 are implemented by reading the bitmap binary from the appropriate ISB file and displaying the bitmap image on the screen with reference to the index and file path stored in the LIK file. According to an embodiment of the present invention, in the step shown in Fig. 4-2, it is preferable to represent the original character images to be a text input to a separate reference window on one side of the screen. In particular, it is preferable that the character images displayed in the reference window are arranged based on a predetermined criterion. FIG. 14 shows an example in which the characters appearing throughout the entire Korean material to which the method of the present invention is applied are arranged in frequency order. have. In addition, apart from the reference window, it is preferable to provide the input window at an appropriate position on the screen. According to an embodiment of the present invention, when selecting and clicking one of the text images listed in the reference window, the clustered bitmap images are displayed on the screen through the clustering step of the above discussion.

Next, if there is a different character among the displayed bitmap images, the user selects and checks that character. 15 shows the step of selecting different characters. After confirming that there are different characters, which are not properly grouped among the displayed bitmap images, the user inputs the character text corresponding to the displayed character image in the input window. This text entry is performed repeatedly until the last page. When the text input operation is completed, the text information is then stored in the PSD file. 16 illustrates a step of storing text input information in a PSD file. In this process, the LIK file index and file name information is used to find an appropriate PSD file, and the code value is stored at an appropriate location in the file. As a result, together with the PSD file in which the code value of the character is input, a WNG file in which the file names and indexes of the images that are incorrectly grouped are stored is generated. Finally, the WNG file is properly textified through a separate date correction method. Details of the date correction method are disclosed in detail in Korean Patent Laid-Open Publication No. 2001-15963, previously filed and published by the present inventors, the content of which is also adopted as a reference of the present invention to provide a part of the present invention. Configure.

According to another embodiment of the present invention, the present invention provides a method for inputting and correcting a text at high speed by using the aforementioned clustering method.

17 is a flowchart illustrating a method for fast input and correction of characters according to another embodiment of the present invention. With reference to FIG. 17, the fast input and correction method of the character which concerns on this invention is demonstrated.

First, an original image such as Korean traditional data is input and stored in the form of a picture file using a scanner or the like (steps 1 and 2). A zero-in program is operated on the stored image to segment by each character included in the image (step 3). The information of each segmented character is stored in the PSD file (step 4). As described above, the PSD file is a file that stores the index, location information, and code value of each segmented character. A bitmap generator is then used to generate a black and white bitmap for the segmented image (step 5). Generate the generated black and white bitmap as an ISB file (step 6). An ISB file is a file that contains information about a black and white bitmap of segmented images. Next, the clustering step of the above-mentioned image is performed. That is, the comparison criteria of the images are set (step 7), the same images are grouped together (step 8), and an index file (aka LIK file) is generated (step 9). The LIK file contains the location information and the index information of the grouped images, so that when the original image is displayed on the screen later, the corresponding ISB file and PSD file can be easily searched. Step 10 is a step of displaying the original image on the screen using the LIK file. Next, it is determined whether there is a different image among the displayed images (step 11). If there is a different image, the different image is selected and checked (step 12), and text characters corresponding to the appropriately grouped same images are input and stored (step 13). For the different images checked in step 12 above, a WNG file containing index information and paths for the images is generated (step 14). The WNG file is modified using a date correction program (steps 15-18).

According to the method of the present invention, instead of repeatedly inputting a small number of characters with high occurrence frequency, the images representing the same characters are recognized as the same characters and grouped, thereby repeating the same characters through one input operation. Multiple text input operations can be omitted. According to the method of the present invention, it is possible to drastically improve the text input work which is the basis for the computerization of Korean data.

Claims (6)

Comparing the segmented bitmap images; Grouping bitmap images recognized as similar characters through the comparing step; Displaying the grouped bitmap images and Inputting corresponding text of text into each of the suitably grouped images as the user inputs corresponding text of text to the appropriately grouped images in the displayed images. Comparing the segmented bitmap images; Grouping bitmap images recognized as similar characters through the comparing step; Displaying the grouped bitmap images; As the user inputs corresponding text of text corresponding to an appropriately grouped image among the displayed images, inputting corresponding text of text into each of the appropriately grouped images; and Causing the inappropriately grouped images selected by the user to be corrected. The method of claim 1 or 2, wherein comparing the segmented bitmap images and grouping bitmap images recognized as similar characters through the comparing step are performed using a binarization method using black and white separation. Characterized by the above. 4. The clustering method of claim 3, wherein the binarization method using black-and-white separation is characterized by comparing the curve values of a mathematically calculated waveform from the binarized image after binarizing the target image into black and white. Generating a scanned original image; Segmenting the generated image; Converting the segmented image into a bitmap image; Correcting the segmented bitmap image to a constant size, and A method of rapidly entering and correcting characters, comprising the step of inputting corresponding character text to each of the same original images, suitably grouped, using the clustering method of the image according to claim 1. Generating a scanned original image; Segmenting the generated image; Converting the segmented image into a bitmap image; Correcting the segmented bitmap image to a constant size; Using the clustering method of the image according to claim 2, as the user inputs the text corresponding to the appropriately grouped images among the displayed images, the text corresponding to each of the appropriately grouped images is input. And causing the inappropriately grouped images selected by the user to be corrected, on the other hand.