KR100707943B1 - System and Method for recognition of financial statement using optical character recognition - Google Patents

Abstract

Disclosed are a method and a system for recognizing financial statements using an optical character recognition (OCR) function, and databaseting the recognized financial statement information. Financial statement analysis system according to the present invention, the OCR processing unit for scanning the financial statements, extracting the characters from the scanned image, pattern matching the extracted characters with the standard words used in the financial statements to determine the correct words for the extracted characters A standard word judging unit for judging, and a data arranging unit for rearranging the OCR processed data according to the determination result. According to such a financial statement analysis system, the accuracy of the optical character recognition is increased, thereby automating the accounting book processing, thereby greatly reducing the time required for manual processing. In addition, the present invention can increase the reliability of the bookkeeping process by inspecting the error for each region of the OCR processed image using the accounting principle.

Optical Character Recognition, Financial Statements, Pattern Matching

Description

System and Method for recognition of financial statement using optical character recognition}

1 is a conceptual diagram according to an example of a financial statement analysis system according to the present invention;

2 is a diagram illustrating an example of a balance sheet;

3 is a block diagram according to an example of the standard word determination unit illustrated in FIG. 1;

4 is a block diagram according to an example of the check unit shown in FIG. 1;

5 is a block diagram according to another embodiment of the present invention, and

6 shows a flowchart according to an embodiment of the present invention.

* Description of the symbols for the main parts of the drawings *

OCR Processing Unit: 10 Scanner: 11

Character Recognition Program: 12 Standard Word Decision Unit: 20

Checker: 30 Data sorter: 40

The present invention relates to a method and system for recognizing a financial statement using an optical character recognition (OCR) function, and a database of the recognized financial statement information, and more specifically, correcting the imperfection of the optical character recognition function. The present invention relates to a method and system for DBing the recognized financial statement information by improving the accuracy of recognized financial statement information by minimizing the possibility of error in optical character recognition.

Typically optical character recognition aims at extracting character information from an optically scanned image. Recognition of a character is performed by dividing the scanned image into predetermined units, and then mapping the separated image to an arbitrary character by a predetermined algorithm.

In general, various character recognition algorithms have been developed to increase the accuracy of optical character recognition, and various attempts have been made to increase the character recognition rate. Among such character recognition algorithms, there is a method of pattern matching.

However, the accuracy of optical character recognition cannot be 100% perfect, which means that when there are many kinds of characters to be recognized or fonts are unusual, the recognition rate is greatly reduced and the input efficiency is lower than that of manual input. . In addition, when the quality of the scanned image is poor, the recognition efficiency is further lowered. In particular, when only 10 or 52 characters are recognized, such as numbers or alphabets, the recognition accuracy is relatively high. However, when there are many kinds of letters to be recognized, such as Chinese characters or Korean characters, the recognition degree is relatively low.

Financial statements are the performance indicators of companies widely used in accounting, taxation, finance, etc., and important parts of the form and content are legislated or customary.The analysis of financial statements shows the financial soundness, safety of loans, and Eligibility for investment can be determined. Typically, a company registered on the stock exchange can obtain the financial statements of the company as digitalized data online, and can easily make a DB (Data Base) based on the obtained information.

However, in the case of corporate financial statements printed on paper rather than digitized data, it must be digitized by inputting them for various analysis. This input process is time consuming and is a direct factor that significantly reduces the ability to analyze financial statements.

Therefore, there is an urgent need to develop a technique for quickly and accurately recognizing financial statements using optical character recognition methods for use in analyzing financial statements.

Therefore, an object of the present invention is to register the various field names constituting the financial statement in the optical character recognition process in order to increase the accuracy of the optical character recognition, and then optical character recognition by a method of contrasting the recognized image with the reserved word. The present invention also provides a method and system for recognizing financial statements to increase the accuracy of the financial statements. Further, another object of the present invention is to recognize financial statements by using a special test formula unique to the financial statements in order to find recognition errors in the recognized financial statements. To provide a method and system for recognizing financial statements that improves performance.

The above object is an OCR processing unit for scanning a financial statement, extracting a character string from the scanned image, and comparing the extracted character string with a standard word used in the financial statement according to the present invention. And a data sorting unit for rearranging the OCR-processed data according to the selected standard word.

Preferably, the apparatus further includes a checker that extracts at least one numeric string from the OCR data, and accounts for and checks the numeric string associated with the extracted standard word.

Preferably, the checker further includes an arithmetic processing unit that divides the scanned image into at least two areas according to the position where the number string is arranged, and adds or subtracts the number string of each partitioned area.

The standard word is preferably a word used in the balance sheet, the income statement, the statement of retained earnings, and the cash flow statement.

The standard word determination unit may include a standard word storage unit having information on the standard word, and a standard word most similar to the standard words stored in the standard word storage unit by comparing the standard word and the extracted OCR string. It is preferable to include a standard word selection unit for selecting.

The financial statement is preferably one of a balance sheet, an income statement, a statement of retained earnings, and a cash flow statement.

Preferably, the data sorting unit generates a file in which the standard word and the numeric string are linked.

It is preferable that the said file is an Excel file.

According to the present invention, the scanning unit for extracting the image by scanning the financial statement, the standard word judgment for selecting the correct word for the extracted character by pattern matching the image with the standard word used in the financial statement And a file generation unit that generates a file consisting of standard words according to the determination result.

The standard word determination unit may include a first memory configured to temporarily store the scanned image, a standard word storage unit including an image of the standard word, and the standard word, and an image of the temporary stored image and the standard word. It is preferable to include a standard word selector that selects the closest standard word by pattern matching.

Preferably, the financial statement is a standard balance sheet, and the standard word storage unit includes a word described in the standard balance sheet as a standard word.

Preferably, the apparatus further includes a checker that extracts at least one string of numbers from the scanned image, and accounts for and checks the string of numbers associated with the selected standard word.

The standard word is preferably a word used in the balance sheet, the income statement, the statement of retained earnings, and the cash flow statement.

Preferably, the file generation unit generates a file in which the standard word and the number string are linked.

According to the present invention, the object of the present invention is to scan a financial statement, OCR processing the scanned financial statement to extract a string of letters and numbers, and compare the extracted string with a plurality of standard words used in the financial statement. Selecting the closest standard word, and generating a financial statement file containing the selected standard word.

Preferably, the method further includes performing a check using an accounting principle on the numeric string associated with the standard word among the extracted numeric strings.

The performing of the checking may include: dividing the sequence of numbers according to the positions arranged in the financial statements, and subtracting the numbers of the divided regions according to an accounting method, and transferring the resultant values. It is preferable to include the step of determining.

The method may further include generating a file in which the standard word and the string of numbers are matched when the processed result value is normal.

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

1 is a conceptual diagram according to an example of a financial statement analysis system according to the present invention.

As shown, the financial statement analysis system according to the present embodiment includes an OCR processing unit 10 for processing OCR (Optical Character Recognition) of A4 paper size financial statements submitted to a financial institution such as a bank. To compare the extracted letters (or strings) and numbers (or strings of numbers) with the standard words commonly used in accounting books (for example, financial statements (balance sheets, income statements, statement of retained earnings, and cash flow statements). A standard word determination unit 20, a checker 30 that adds and subtracts numeric strings (or numbers) extracted from an OCR processed document to determine whether each OCR processed numeric string is correct, and matches a standard word and a numeric string. And a data alignment unit 40 to generate a file. In general, the OCR processing unit 10 reads the financial statements optically, and the read rate of the letters (or strings) or numbers (or strings) is in the 90% range, and the font, size, and printing of the characters described in the financial statements are maintained. There is a disadvantage that the read rate is changed depending on the state. Accordingly, the present embodiment registers the OCR processed data as a standard word with words such as debits, credits, total assets, current assets, fixed assets, and current assets, which are commonly used in bookkeeping. And OCR are processed to compare the extracted text data to find the closest character (or string) and numeric string (or number) and merge them into one file. For example, a character (or a character string, which will be omitted hereinafter) obtained by applying the character recognition program 12 to the image scanned by the scanner 11 is "illusion asset" and registered in the standard word determination unit 20. When the standard word is "current asset", the standard word determination unit 20 determines that the "CRIM" is processed as "current assets" used in the accounting book. That is, the standard word determination unit 20 registers words commonly used in the account book and compares them with the OCR-processed characters to reduce errors generated during OCR processing.

The checker 30 determines whether or not the recognized number (or string of numbers) is accounted for the number among the OCR processed letters or numbers. All books of account have a certain relationship between deposits and withdrawals. This will be described with reference to FIG. 2 together. 2 shows an example of a balance sheet. As shown, the balance sheet debits all assets, the liabilities and capital on credit, and the sum of the debit and credit is equal. Normally, the balance sheet is divided into left and right, the divided left side is the vehicle and the right side is the credit. Accordingly, the checker 30 adds the numeric strings on the left side and the numeric strings on the right side, respectively, and then determines whether the two sides have the same value. Can be determined. In addition, the checker 30 performs an OCR process using the following accounting principles, for example, the debit representing the asset and the balance average of the credit representing the sum of the debt and the capital, and using the same value. Can be determined.

1) Total Assets = Total Liabilities + Total Capital

2) Total Assets = Current Assets + Fixed Assets

3) Total liabilities = Current liabilities + Fixed liabilities

4) Current Asset = Quick Asset + Inventory

5) Fixed Assets = Investment Assets + Tangible Assets + Intangible Assets

In addition, the checker 30 may determine whether the OCR has been processed by using the following check equation using the accounting principle of the balance sheet.

6) Gross profit = revenue-cost of sales

7) Operating Income = Gross Profit-Sales and Maintenance

8) Recurring Income = Operating Income-Non-operating Cost + Operating Cost

9) Profit before Income Tax = Ordinary Income + Special Income-Special Loss

10) Net income = Profit before income tax-Income tax

In addition, if the checking assets consist of cash, ordinary deposits, accounts receivables, and VAT payments in the financial statements, the checking assets = cash + ordinary deposits + accounts receivables + VAT payments. It is also possible to determine whether or not the OCR processed number is applied. This embodiment enumerates the above ten accounting principles, but in addition, the accounting principles used in the balance sheet, the income statement, the retained earnings statement, and the cash flow statement can be applied. In addition, the accounting principle as described above may be applied for each area scanned by the scanner 11 to determine whether the OCR processing result for each area is good or bad. The data sorter 40 merges the word determined by the standard word determiner 20 and the number checked according to the accounting principle by the checker 30 into one file and converts the file into an Excel-like format. Of course, the format of the converted file can be a variety of file formats in addition to Excel. Such a file generation can be usefully used to computerize the accounting book and make it into a database in a financial institution such as a bank.

FIG. 3 is a block diagram illustrating an example of the standard word determiner illustrated in FIG. 1.

The illustrated standard word determiner 20 includes a standard word store 21 and a standard word selector 22. The standard word storage unit 21 is implemented by a storage medium such as RAM, ROM, and Hard Disk Drive, and currently exists in a standardized account book (eg, asset total, Total liabilities, total capital, current assets, fixed assets, current assets, inventory sales, cost of sales, net income, etc.). In this case, the standard word storage unit 21 may additionally register a standard word in addition to the registered standard word. The standard word selector 22 selects the best matched word by comparing the letters and numbers that are subjected to OCR processing from the OCR processor 10 with the standard words registered in the standard word storage 21. The standard word selected by the standard word selector 22 replaces the letters and numbers provided by the OCR processing unit 10 and is provided to the data sorter 40. Accordingly, the data sorter 40 is provided with the letters and numbers corrected by the error in the standard word selector 22 and merges them into a single file.

4 is a block diagram illustrating an example of a check unit shown in FIG. 1.

The illustrated check unit 30 is composed of a buffer 31 and an arithmetic processing unit 32, and the arithmetic processing unit 32 includes a partition setting unit 32a and an arithmetic unit 32b. The buffer 31 temporarily stores the image output for the number from the OCR processing unit 10. The numbers stored in the buffer 31 are applied to the division setting unit 32a, and the division setting unit 32a divides the numbers into several numbers according to the characteristics of the account book (for example, the balance sheet). For example, the partition setting unit 32a may divide the left and right of the image as shown in FIG. 2 into 50:50. The calculation unit 32b calculates the left and right sides of the image divided by the partition setting unit 32a, respectively, and determines the identity of the calculated result. As a result of the determination, it may be determined that the numbers written on the left and right sides are correct, and in the opposite case, an error may occur in the number recognition. This is based on the accounting principle that the sum of the numbers on each of the left and right sides of the balance sheet is the same.

Meanwhile, the partition setting unit 32a may partition the OCR processed image by referring to the word determined by the standard word determination unit 20. Taking the accounting book as shown in FIG. 2 as an example, in accordance with the accounting principle, the current asset is a current asset + inventory, using the term "current assets" and "current assets" among the words determined by the standard word determination unit 20. It is also possible to set the numbers on the right side of "and" the "inventory assets", "61,419,500", "57,419,500", and "4,000,000" to one compartment, and perform arithmetic processing on the numbers provided in this compartment. In this case, since current assets are quick assets + inventories, the numbers associated with these words must satisfy 61,419,500 = 57,419,500 + 4,000,000.

5 is a block diagram according to another embodiment of the present invention.

The financial statement analysis system using the optical character recognition function according to the illustrated embodiment includes a scanner 50, a standard word selector 61, a checker 70, and a data sorter 80, and select a standard word. The unit 61 includes a buffer 61a, a scaler 61b, and a pattern matching unit 61c. The standard image storage 62 is a word commonly used in an accounting book (e.g., financial statements (balance sheet, income statement, statement of retained earnings, and cash flow statement) (e.g., total assets, total liabilities, total capital, current assets). , Fixed assets, checking assets, inventory sales, cost of sales, net income, etc.) and images corresponding to words. The buffer 61a temporarily stores image data transmitted from the scanner 50. The scaler 61b normalizes the letters or numbers temporarily stored in the buffer 61a to the same size as the letters and numbers stored in the standard image storage 62. The pattern matching unit 61c determines a most similar image by pattern matching the image of letters or numbers normalized by the scaler 61b and image data previously stored in the standard image storage unit 62. For example, when the scaler 61b provides image data recognized as "two-sided assets", the pattern matching section 61c performs a pattern matching comparison of the images stored in the standard image storage section 62 with the "two-sided assets". Selecting the most similar "investment asset" reduces errors in the scanned image data.

On the other hand, if the financial statements authorized by the scanner 50 are standard financial statements (for example, standard balance sheet) according to the form set by the country, the account subjects used in the standard financial statements (for example, current assets, current assets, short-term financial instruments, short-term) Loans, receivables, advance payments, inventory, temporary goods, stores, land, fixed assets, investment assets, debentures, constructs, long-term trade receivables, long-term loans, and opening expenses, etc. Accordingly, it is preferable that the standard image storage 62 include image data of the account subject described in the standard financial statement. Standard financial statements define the name and location of each accounting subject. Therefore, when comparing the extracted character and the image data, the pattern matching unit 61c performs pattern matching by referring to the position information where the characters are scanned, thereby reducing the error due to the pattern matching.

6 shows a flowchart according to an embodiment of the present invention.

First, the account book submitted to a financial institution or an accounting office is scanned through the scanner 11 (S101). The scanned image for the book is pattern matched by the OCR program to read letters and numbers (S102). In this case, the OCR program for performing pattern matching on the image acquired through the scanner 11 is preferably installed in a separate computer connected to the scanner 11. Next, the correct word is selected by the standard letter selection unit 24 for letters and numbers obtained by pattern matching by the OCR program. In this case, the following two methods may be used to determine letters and numbers.

1) A method of selecting the most similar word by comparing letters and numbers obtained after OCR processing with previously stored standard words.

2) A method of pattern matching between an image of a character obtained after OCR processing and an image of a pre-stored standard word to select the most similar word.

After the letters and numbers are selected (S103), the numbers are partitioned into a predetermined number, and arithmetic processing is performed for each section to determine whether the selected number is in error. Since this has been described in detail above, a detailed description thereof will be omitted. Next, when the check result is normal (S104), the standard words and numbers are rearranged according to the position scanned by the original scanner 11, and converted into a file of an Excel-like format (S105). The converted file may be stored in a database of a bank, an accountant having a mass storage medium, and an organization processing other account books (S106).

As described above, the present invention significantly reduces the time that is manually processed by automating the accounting book by increasing the accuracy of optical character recognition. Accordingly, it is possible to generate and accumulate digitized data even on financial statements printed on media such as paper. In addition, the present invention improves the reliability of the accounting book process by inspecting errors for each area of the OCR processed image using the accounting principle.

Claims (18)

An OCR processor for scanning the financial statements and extracting character strings from the scanned image; A standard word determination unit selecting a correct word for the extracted string by comparing the extracted character string with a standard word used in the financial statement; A data sorting unit for rearranging the OCR processed data according to the selected standard word; And And a checker for extracting at least one numeric string from the OCR data and performing an accounting process on the extracted numeric string associated with the extracted standard word. The financial statement analysis system comprising an optical character recognition function. delete The method of claim 1, The checker unit, An arithmetic processing unit for dividing the scanned image into at least two regions according to the position where the sequence of numbers is arranged, and adding or subtracting the sequence of numbers of each segmented region; and using an optical character recognition function. Financial Statement Analysis System. The method of claim 1, The standard word is Financial statement analysis system using the optical character recognition, characterized in that the words used in the balance sheet, income statement, statement of retained earnings and cash flow statement. The method of claim 1, The standard word determination unit, A standard word storage unit having information about the standard word; And A standard word selector which selects the most similar standard word among the standard words stored in the standard word storage unit by comparing the standard word and the string extracted by the OCR process and using the optical character recognition function Financial Statement Analysis System. The method of claim 1, The financial statements, A financial statement analysis system using optical character recognition, characterized in that the balance sheet, income statement, retained earnings statement, and cash flow statement. The method of claim 6, The data sorting unit, A financial statement analysis system using the optical character recognition function, characterized in that for generating a file associated with the standard word and the numeric string. The method of claim 7, wherein The file is Financial statement analysis system using optical character recognition, characterized in that the Excel file. A scanning unit which scans the financial statement and extracts an image; A standard word determination unit for pattern matching the image with a standard word used in the financial statement to select an exact word for the extracted character; A file generation unit generating a file composed of standard words according to the determination result; And And a checker for extracting at least one numeric string from the scanned image and performing an accounting process on the numeric string associated with the selected standard word. The financial statement analysis system comprising an optical character recognition function. The method of claim 9, The standard word determination unit, A first memory for temporarily storing the scanned image; A standard word storage unit having the standard word and an image of the standard word; And And a standard word selector configured to select the closest standard word by pattern matching the temporarily stored image and the image for the standard word. The method of claim 10, The financial statements are standard balance sheets, The standard word storage unit, A financial statement analysis system using the optical character recognition function, characterized in that the word described in the standard balance sheet as a standard word. delete The method of claim 10, The standard word is Financial statement analysis system using the optical character recognition, characterized in that the words used in the balance sheet, income statement, statement of retained earnings and cash flow statement. The method of claim 10, The file generation unit, A financial statement analysis system using the optical character recognition function, characterized in that for generating a file associated with the standard word and the numeric string. Scanning the financial statements; Extracting a string of letters and numbers by OCRing the scanned financial statements; Selecting the closest standard word by comparing the extracted character string with a plurality of standard words used in financial statements; Generating a financial statement file including the selected standard word; And Performing a check using an accounting principle on the numeric string associated with the standard word among the extracted numeric strings; and analyzing the financial statements using the optical character recognition function. delete The method of claim 15, The step of performing the check, Dividing the sequence of numbers according to the positions arranged in the financial statements, and adding or subtracting the numbers of the divided regions according to an accounting method; And Determining whether or not the result of the subtraction process result; Financial statement analysis method using an optical character recognition function comprising a. The method of claim 17, If the processed result is normal, generating a file in which the standard word and the string of numbers are matched; and analyzing the financial statement using the optical character recognition function.

Images