KR20030018519A - The Easy Chinese Character Input and Correction Method using Image Retrieval Technologies - Google Patents

Abstract

PURPOSE: A method for correcting a Chinese character input is provided to input a Chinese character easily by extracting characteristic information of an image and using a search technique by a similarity comparison of the information without a difficult knowledge such as Chinese classic. CONSTITUTION: A scanning device(11) inputs data of an ancient document. An image duplication checking device(12) measures a similarity for avoiding a duplication of scanned data. An image correction device(13) removes a miscellaneous image for extracting information accurately in a scanned image and executes a horizontal and vertical set of a page. A segmentation device(14) cuts a character one by one in a sentence included in an image of a scanned page and creates character image information. An image storing device(15) stores an image of each extracted character in a storage. An image indexing device(16) extracts and stores characteristic information which represents each image applied to a stored character. An image searching device(17) arranges and displays similarity information according to a similarity to a given character. An automatic code endowing device(18) endows a character within a range of a similarity given in the searched result with a character code automatically. A manual code endowing device(19) endows many characters designated by a user in the searched result with a code value, respectively. A code value correction device(20) inputs a code value for checking an error of characters having a character code previously, displays a character image and a code value corresponded thereto, and corrects a wrong code value.

Description

Easy Chinese Character Input and Correction Method using Image Retrieval Technologies

The present invention is to automate the task of inputting an electronic file into a computer such as a PC in order to digitize the Chinese character-oriented study materials such as ancient documents by applying image retrieval technology using the characteristic information of the image and an easier method. This is how you can enter data.

In addition, the correction method for the Chinese character input result is not a comparison method between the original text and the input one-letter Hangul characters, but a correction method that makes it easier to find a wrong input character by collecting and comparing all segmented characters of characters already input by the Chinese character code To provide.

The existing automated input method attempts to reduce the human hand part by using the character recognition technology (OCR), but the result is good for the Chinese character data due to the technical limitation of the OCR technology and the limitation of the processing capacity of the OCR algorithm. The reality is not getting it.

In the case of Chinese characters, all the materials are written by human handwriting, so the handwriting is different depending on the person who created the same letter, and it has been kept for several hundreds of years. It's not easy. Moreover, Chinese characters have various fonts and it is very difficult to process them with OCR technology.

In addition to the automatic input of a large number of Korean-language data, an efficient calibration method that checks and corrects the inputted contents is required. The conventional method was to check each letter by hand and eyes one by one.

Accordingly, it is an object of the present invention to provide a method of easily inputting Chinese characters without difficult knowledge such as Chinese literature as a retrieval technique by extracting feature information of an image and comparing similarities thereof.

In addition, another object of the present invention is to provide a method for reducing errors that occur frequently in the recognition or reading by the expert by applying the image processing technology to collect the same and similar letters at a glance.

In addition, another object of the present invention is the basic Chinese character code input method is input by increasing the number of characters in proportion to the number of characters input by one character unit, the technique of the present invention by inputting a code value to all or part selected from the search results at the same time To provide a way to explode airborne exponentially.

In addition, another object of the present invention is to apply an automatic code input method that can be substituted for OCR technology by applying a method for automatically assigning a code value to a character below a certain threshold based on the similarity value measured by image similarity comparison. To provide.

1 is a diagram showing the functional flow of an image-based input calibration system;

2 is a view showing the result of the image redundancy check apparatus,

3 is a diagram illustrating vertical spectral analysis of an image calibration apparatus;

4 is a diagram illustrating line separation from an image of an image segmentation device and subsequent single letter separation;

5 is a diagram showing an example of an index information generation result by the image indexing apparatus;

6 is a diagram illustrating an example of a user interface screen for an image retrieval device;

7 is a view showing an example of a user interface screen of the automatic code assignment device;

8 is a view showing an example of a user interface screen of the manual code assignment device;

9 is a diagram illustrating an example of a user interface screen of a code value calibrating apparatus;

10 illustrates one preferred example in which all these devices are implemented and operated within a computer;

11 is a graph showing actual Chinese character frequency measurement data to show the effect of image processing.

According to the present invention for achieving the above object, there is provided an image search-based Chinese character input correction system. The image retrieval-based Chinese character input correction system includes: a scanning device for inputting data of an old document to be input; Image redundancy inspection device by measuring image similarity to avoid duplication of many scanned materials; An image correction device for removing horizontal noise and aligning the page horizontally and vertically for accurate information extraction from the scanned image; A segmentation device for generating character image information by cutting out characters one by one from sentences or multiple characters included in the scanned one page image; A storage device for storing an image of each letter extracted one letter at a storage device; An image indexing apparatus for extracting and storing feature information representing each image from the stored characters; An image retrieval device that lists similarity information in order from the most similar to a given letter (image); An automatic code assigning device for automatically assigning a letter code to a letter within a range of a given similarity value in the searched result; A manual code assigning device for assigning a code value to a plurality of characters designated by a user in a search result; In order to check for errors of the characters that have already completed the character code value, the device includes a correction device that inputs a code value and displays a corresponding character image and code value to correct an incorrect code value.

Preferably, the scanning device for inputting the data of the old document stores the data in the file system of the storage device in a monochrome information representation format in units of one page of the old document.

More preferably, the file of the scanned image information may be directly managed in the internal memory of the image-based Chinese character input calibration system so that the user may process the file within the system without a separate storage management and transfer process.

Preferably, the duplication inspection apparatus for eliminating duplication of the scanned data is an apparatus for efficiently managing and processing a large amount of image data by judging by similarity measurement based on image retrieval technology. It is a device for managing only image data without duplicates by extracting a duplicated stored image that may be generated by an incorrect management process during a scanning operation by several workers from a database or the like.

Here, the image retrieval technique may apply various known algorithms, and a well-implemented example may be configured by applying a similarity measurement algorithm or region shape algorithm based on edge component extraction.

Preferably, the image correction device performs a process for eliminating noise and aligning the page horizontally and vertically for accurate character segmentation in the image scanned by one page. The purpose of this device is to prevent the various damages caused by the long-term neglect of long-term storage and management of the old books, which would adversely affect the automatic image segmentation.

In order to level the scanned image of one page, the horizontal spectrum representing the coordinates of the x-axis of the image plus the values of pixels (pixel: 0 or 1) of the y-axis is determined. In other words, because the letters are written as vertical letters, the parts with letters appear high spectral values, and between one and the next lines of written letters becomes empty space without letters, so the spectrum at this time has the lowest value. . If the scanned image is skewed, the spectrum will show a sinusoidal curve or flat shape with less variation on the x-axis, and will have the highest and lowest spectral values if it is not tilted and is exactly horizontal.

In this case, the scanned image of the page is aligned vertically and horizontally, and the image correction process is completed by removing ghosts in a part not related to the text by manual or automated methods.

Preferably, the segmentation apparatus for generating the character image information included in the scanned one page image is to cut out the characters one by one from a sentence or composed of several characters, for each of the separated character image You can also bring location information to restore the text position on the original page.

More preferably, the segmentation of the letters is automatic, but in some cases, manual work is required, and the segmentation error generated by the automatic processing can also be provided through the manual process.

Preferably, a storage device for storing each character image separated by one letter in a storage device may be stored in a computer memory or a file, or by using a database or the like to process a large amount of characters at the same time. .

As described above, in order to process the original position information of the letter, the position information is also stored together so as not to change the order or meaning in the process of restoring the original sentence.

Preferably, the image indexing apparatus extracts feature information representative of each character image to generate index information to enable similarity comparison. Index information can be judged as the same letter even if the font or shape of the letter is slightly different, or it is better to define characteristic information that can be distinguished from other letters, and among the various embodiments, Edge can be applied well. Similarity measurement algorithm based on component extraction, or Region Shape algorithm.

More preferably, the index information by this algorithm is expressed in multiple dimensions, for example, vertical component x%, horizontal component y%, slash (/) diagonal component z%, backslash (\) diagonal component w%, etc. As a result, it is determined how similar each letter is to each of these components.

The index information of the image of each character thus determined is also stored in the storage device or file of the database to form the index information of the entire character.

Preferably, the image retrieval apparatus showing the similarity information provides as a result of the character images determined to be the most similar to the character (image) presented by the user, wherein the similarity determination is based on the index information generated by the image indexing apparatus. Decide by using.

More preferably, various algorithms may be applied for the similarity determination, and in general, a vector space model or the like may be used. In order to show the result determined to the user is shown in order from the letter which is determined to be the most similar, the similarity measurement result can also be presented to help the user to judge.

Preferably, the automatic code assigning apparatus for the searched results applies a method for automatically assigning a character code to a character image within a predetermined range based on a threshold similarity value predetermined and input by a system user or administrator. can do. This method is a new method that can provide the same function as the existing optical character recognition (OCR).

More preferably, as the user uses the system, the threshold similarity value may be adjusted according to the state of the characters or the performance of the system to obtain better results.

Preferably, the manual code assigning apparatus provides a method of assigning a code value to various characters designated by a user by hand in the searched result. In the search results, some other letters may be interposed between the desired letters, and sometimes appear far from the order of similarity, which is a useful way to do this. The manual selection method provided to the user may include a method of selecting several simultaneously, a method of not selecting several simultaneously, a method of selecting one by one, and a method of not selecting one by one.

Preferably, a correction device capable of correcting an invalid code value is for checking an error of characters that have already been completed with a code value, and when a character code value is input, a function of displaying an image of all corresponding letters that are given the same letter code is provided. To provide. This allows the user to easily find the wrong letter in the same set of letters, and provides the ability to immediately correct.

More preferably, the correction of wrong letters can be easily performed by selecting various letters, i.e., selecting several letters at the same time, not selecting several letters at the same time, selecting one by one, not selecting one by one, etc. It provides the ability for calibration to take place.

1 is a diagram illustrating a preferred example of an image search-based Chinese character input correction system according to an embodiment of the present invention. The image retrieval-based Chinese character input correction system includes a scanning device 11, an image redundancy inspection device 12, an image correction device 13, a segmentation device 14, an image storage device 15, and an image indexing device 16. , An image retrieval device 17, an automatic code assignment device 18, a manual code assignment device 19, and a code value correction device 20.

Figure 2 shows an example of the results actually provided by the implementation of the image redundancy check apparatus 12 of the present invention, it can be seen that the scanned data of the same image is gathered from 21 to 25. Each image is the result of several people scanning the same contents, and it can be seen that the contents are the same but the positions of the images are different.

3 shows a function corresponding to the image correction device 13 of the present invention, and preferably shows an example of measuring tilt by vertical spectrum analysis. Preferably, measuring the vertical spectrum with respect to 31 images yields the same result as 32. This flatness indicates that the image is skewed, and measuring the vertical spectrum with respect to 33 results in 34. Therefore, it is possible to determine whether the scanned image is exactly vertically and horizontally scanned using the obtained spectrum values.

4 is a diagram illustrating a process in which the segmentation device 14 separates the pieces after passing the lines in the image. Preferably, 41 shows the vertical horizontal spectrum (projection) described in FIG. 3, and the segmentation apparatus first performs line separation as shown in 42 and 43, and then performs word separation as shown in 44 and 45. FIG.

5 shows an example of a method of generating good index information on a character image by the image indexing device 16. Characters like 51 or 53 will generate index information like 52 and 54. As can be seen from the example, it can be seen that the horizontal component of 51 (the value shown in 52) is larger than the horizontal component of 53 (the value shown in 54).

6 shows a user screen of the image input correction system in a good case, 61 shows an image selected by the user, and those judged as similar images (letters) by the search function are displayed in the area 63. FIG.

FIG. 7 is a screen for automatically assigning codes by the automatic code assigning device 18. Preferably, images collectively selected by a preset similarity threshold are automatically assigned code values. Supports batch code assignment of the characters 71 manually selected by the user for the remaining unprocessed characters.

8 is a good user interface screen of the manual code assigning device 19. The letter 81, in which the code value is already input, is not the object of the present device, and the user directly reads the letter 82 which is not given any code. It allows you to enter.

9 is a good user interface screen of the code value correcting apparatus 20. When a character is inputted into the window 91 through the Chinese character input by the phonetic value, which is a general Chinese character input method, and the character is given the code value 94 The results are displayed in the flourish, and letters such as 92 and 93 that are incorrectly coded are easily visible and can be easily modified.

10 shows a good example of such devices configured in a computer system, all devices being loaded from the hard disk 114 into the main memory 113, interfaced with the user through the monitor 111, and storing or indexing images. In addition, retrieval devices store or recall necessary information on a hard disk. The image scanning device is connected to an external scanner 112 through a communication port, and the CPU 115 manages all these processing.

Korean characters written in Chinese characters have been difficult to construct as well as to decipher unless they are experts. The problem of Korean data processing is that it is difficult to decipher due to various letters, various fonts, and transfer characters. Actually, an input process is required to digitize and store these data. Problems in the input process require a lot of input maneuver and in the process of input is not made accurate input by similar letters, wrong letters and the like. Since these inputs are made by simple input operators, not by experts, the process of calibration is inevitable. The difficulty of proofing is that it is not easy to compare the input text with the original text, so it is not easy and it can only be compared to a non-letter picture unless you are an expert.

It is an object of the present invention to solve this problem and solves the problem of the Korean data processing pointed out above. In other words, avoiding the difficulty of decoding by applying the technology of processing Chinese characters into images, by applying the method of processing Chinese characters by image indexing and retrieval techniques, and devising a batch input method of retrieved images, it is easy for the general public It provides a way to work, reducing the area of work of specialists so that they can study in more specialized areas.

The present invention applies an image processing method as well as an input to provide an effective method for correcting and contrasting the inputted material, thereby contributing to the digitization of more complete isolated data.

As shown in the graph of Fig. 11, the effectiveness of the present invention shows that the processing of more than 95% of the entire data is possible even if only 1,900 women are processed by the Chinese character frequency rank. It shows how much of the data processing can be handled easily.

Claims (10)

A scanning device for inputting data of an old document to be input; And Image redundancy inspection device by measuring image similarity to avoid duplication of many scanned materials; And An image correction device for removing horizontal noise and aligning the page horizontally and vertically for accurate information extraction from the scanned image; And A segmentation device for generating character image information by cutting out characters one by one from sentences or multiple characters included in the scanned one page image; And A storage device for storing an image of each letter extracted one letter at a storage device; And An image indexing apparatus for extracting and storing feature information representing each image from the stored characters; And An image retrieval device that lists similarity information in order from the most similar to a given letter (image); And An automatic code assigning device for automatically assigning a letter code to a letter within a range of a given similarity value in the searched result; And A manual code assigning device for assigning a code value to a plurality of characters designated by a user in a search result; And An image-based input calibration system comprising a calibration device that can correct an incorrect code value by inputting a code value and displaying the corresponding character image and code value to check for errors in characters that have already been completed. The image overlapping apparatus of claim 1, A way to check for image duplication with an image index on a full page basis. The image correcting apparatus of claim 1, Full page-by-page image vertical and horizontal spectral analysis that allows you to examine image corrections and modify images horizontally. The apparatus of claim 1, wherein the image segmentation device comprises: Full page image vertical and horizontal spectral analysis that cuts the text components contained in an image to create a text image in single-letter units. The image storage device of claim 1, wherein the image storage device comprises: A method of storing information generated as a single character image by a segmentation device for indexing and searching in a database or the like. The method of claim 1, wherein the image indexing device, A method of extracting index information by analyzing vertical, horizontal, and diagonal components of all stored single letter images. The apparatus of claim 1, wherein the image retrieval device is A method of calculating the similarity by vector space measurement based on the index information of each character, and ranking the most similar. The apparatus of claim 1, wherein the image automatic code assigning device is provided. Character images collectively selected by the preset similarity threshold are automatically assigned a code value, and a method of enabling batch code assignment of characters manually selected by the user for the remaining characters not processed by such automatic code assignment . The apparatus of claim 1, wherein the image manual code assigning device comprises: A method that allows users to directly enter a code value for characters that have not been given a code, except for characters that have already been automatically entered with a code value. The apparatus of claim 1, wherein the code value correcting apparatus comprises: If you enter a code in the character input window through the Chinese character input by the phonetic value, which is a general Chinese character input method, and search for a character with the same code value, the results will be displayed. A identifiable and easy way to modify the code of these letters.