WO2007011188A1

WO2007011188A1 - Method and apparatus for recognizing serial number of paper money

Info

Publication number: WO2007011188A1
Application number: PCT/KR2006/002879
Authority: WO
Inventors: Jae-Huan Park; Sang-Youl Jeon; Sang-Keun Seo; Seung-Hwan Seo; Gy-Yeop Kim
Original assignee: Seetech Co., Ltd.
Priority date: 2005-07-21
Filing date: 2006-07-21
Publication date: 2007-01-25
Also published as: KR20070011867A; KR100719608B1; TR200800352T1

Abstract

Provided are a method and an apparatus for recognizing a serial number of paper money. The method includes: receiving an image of paper money including a plurality of figures and a plurality of characters and extracting an image of a character sequence of a portion on which the serial number is inscribed; dividing the image of the character sequence into character images having a predetermined pixel size; allocating the character images to an area having a predetermined pixel size, converting values of quantitated pixels of the area into numerical values, and arranging the values in a series of numerical sequences to generate input data; inputting the input data into a neural network which has learnt a plurality of pieces of basic character data of each character to allow the input data to coincide with the plurality of pieces of basic character data of the each character and outputting probability values depending on a number of cases of the plurality of pieces of basic character data; and arranging the probability values output from the neural network in a descending order, selecting a case according to first basic character data having the highest probability value, sequentially arranging characters corresponding to the case, and outputting the characters as a character sequence.

Description

METHOD AND APPARATUS FOR RECOGNIZING SERIAL NUMBER OF

PAPER MONEY

TECHNICAL FIELD The present invention relates to a method and an apparatus for recognizing a serial number of paper money, and more particularly, to a method and an apparatus for extracting and processing an image of paper money in a paper money counter or a forged paper money discriminator to recognize a serial number of the paper money.

BACKGROUND ART

Paper money counters are used in banks or the like to count a number of paper money. General paper money counters have simple functions of counting paper money. Paper money counters having functions of discriminating the genuineness or spuriousness and denominations of paper money and recognizing serial numbers of the paper money have been suggested.

Such conventional paper money counters having functions of recognizing serial numbers use a method of scanning and outputting a portion indicating a serial number or sensing only fragmentary characteristics of an image of the serial number to recognize letters. The conventional paper money counters also use a method of extracting an image of a serial number of paper money and comparing the image with a reference letter image to recognize a letter sequence.

However, in the former method, the scanned image is output as it is. Thus, it is difficult to process the recognized serial number. Also, since only the fragmentary characteristics of the image of the serial number are sensed, a recognition rate is deteriorated.

In the latter method, recognition time is long, and a state of the image of the serial number is deformed depending on forged paper money or a scanning environment. Thus, it is difficult to freely recognize the serial number in a general- purpose use environment.

DESCRIPTION OF THE DRAWINGS The above and other aspects and advantages of the present invention will become more apparent by describing in detail exemplary embodiments thereof with reference to the attached drawings in which:

FIG. 1 is a flowchart illustrating a method of recognizing a serial number of paper money according to an embodiment of the present invention;

FIG. 2 is a view illustrating a process of extracting a character sequence of a portion in which a serial number is inscribed in the method illustrated in FIG. 1 ;

FIG. 3 is a view illustrating an image of the character sequence extracted in the process of FIG. 2 and character portions into which the character sequence is divided; FIG. 4 is a view illustrating a relationship between an output of a neural network and an error determination in the method illustrated in FIG. 1 ;

FIG. 5 is a schematic block diagram illustrating a configuration of an apparatus for recognizing a serial number of paper money according to an embodiment of the present invention; and FIG. 6 is a view illustrating a configuration of a serial number recognition paper money counter according to an embodiment of the present invention.

DETAILED DESCRIPTION OF THE INVENTION

TECHNICAL PROBLEM The present invention provides a method and an apparatus for constituting a recognition algorithm using a mathematical process to further quickly and precisely recognize a serial number of paper money through an obtainment of an image of the paper money.

The present invention also provides a method and an apparatus for detecting an error during a recognition of a serial number of paper money to determine a validity of a recognition of characters so as to recognize the serial number of the paper money.

ADVANTAGEOUS EFFECTS

As described above, according to the present invention, an image of paper money can be input and processed to recognize a serial number of the paper money. Thus, the serial number can be recognized as an output value so as to be further quickly recognized.

Also, output values of a function can be compared during the recognition of the serial number to determine a validity of the recognition of the serial number and check an error. As a result, reliability of the recognition of the serial number can be improved.

BEST MODE The present invention will now be described more fully with reference to the accompanying drawings, in which exemplary embodiments of the invention are shown. The invention may, however, be embodied in many different forms and should not be construed as being limited to the embodiments set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the concept of the invention to those skilled in the art.

FIG. 1 is a flowchart illustrating a method of recognizing a serial number of paper money according to an embodiment of the present invention. Referring to FIG. 1 , in operation S10, an image of paper money is input. In other words, the paper money including character, numerals, symbols, figures, etc. is scanned using an image sensor to generate the image of the paper money so that the generated image is input. Here, the paper money is not limited to a banknote but may correspond to a check, a lottery ticket, a security, a gift coupon, or the like. Also, when the paper money is scanned, an analog-to-digital converter (ADC) may convert the scanned image into a digital image and input the digital image. In operation S20, an image of a character sequence of a portion in which a serial number of the paper money as an identification number of the paper money is inscribed is extracted from the image of the paper money obtained in operation S10.

FIG. 2 is a view illustrating operation S20 of the method illustrated in FIG. 1. Referring to FIG. 2, in operation S21 , the image of the paper money is obtained. Here, the obtained image includes portions besides the image of the paper money. In operation S22, an outer line of the paper money is extracted to remove an image of such an unnecessary portion. Only an area enclosed by the outer line may be used. Alternatively, only the character sequence of the portion showing the serial number of the paper money may be extracted in order to further reduce unnecessary processing. In the present embodiment, a process of extracting only the character sequence will be described. In operation S23, a center of the paper money to be used as a reference point is calculated to check a position of the character sequence of the portion showing the serial number. In operation S24, a direction and a distance are calculated based on the center to calculate position coordinates, and the image of the character sequence in the corresponding position is extracted.

In operation S30, the image of the character sequence including the serial number is divided into characters. FIG. 3 is a view illustrating the image of the character sequence extracted in the process of FIG. 2 and images of characters. Averages of pixels in a horizontal direction in the image of the character sequence are obtained. Coordinates in a point in which average values of the averages of the pixels are decreased and then increased are calculated. The character sequence is divided into the characters in a vertical direction based on the coordinates. A division is performed in the horizontal direction using the same method as that performed in the vertical direction to generate rectangular images of the divided characters.

In operation S40, input data is generated to substitute the images of the characters for a function that will be described later. All of pixel values of the images of the characters may be generated as the input data or the images of the characters may be processed to generate the input data. If the all of the pixel values of the images of the characters are generated as the input data, the images of the characters are allocated to a frame having a predetermined standard pixel size so as to input the images of the characters into a function that will be described later. Pixels of the frame are quantitated to be defined as values of the pixels and arranged in a series of numerical sequences. For example, gray scale values "0" through "255" are defined as the values of the quantitated pixels. Also, the images of the characters may be divided into blocks having predetermined sizes, and pixel values of the blocks are averaged to process the images of the characters. Alternatively, pixel values in a horizontal or vertical direction of the images of the characters. If the input data is generated using the above method, an amount of data may be reduced. If the input data is generated using pixel values, more precise data may be generated. Input data arranged in a series of numercial sequences is generated from the images of the characters.

In operation S50, the input data obtained in operation S40 is substituted for a neural network function. A neural network is a known art, and thus its detailed description will be omitted. Operation S40 is performed with respect to images obtained through combinations of numerical characters, letters, symbols, and reverse images of the numerical characters, the letters, and the symbols to be references for recognizing letters in order to generate basic character data. The basic character data is learnt to provide the neural network in advance. A backpropagation algorithm, a

Boltzman machine learning method, or a simulated annealing learning method may be used to learn the neural network. If the input data is substituted for the neural network function which has been learnt, probability values of the basic character data are output.

A determination is made as to whether the probability values output in operation

S50 are valid. If it is determined that the probability values are not valid, an error signal is output in operation S80.

FIG. 4 is a view illustrating a relationship between an output of a neural network and an error determination in the method illustrated in FIG. 1. Referring to FIG. 4, input data is input into a neural network, and the neural network outputs probability values of basic characters. The probability values are arranged in a descending order so as to use first and second probability values for validity. If the first probability value is not more than or equal to a predetermined probability value, it is determined that an error is present in a recognition of characters. If the first probability value is more than or equal to the predetermined probability value, the first probability value is compared with the second probability value. If an error between the first and second probability values is within a predetermined range, the recognition of the characters using the first and second probability values is not reliable. Thus, it is determined that the error is present in the recognition of the characters. If the first probability value exceeds the predetermined probability value and the error between the first and second probability values is not within the predetermined range, it is determined that the recognition of the characters is reliable. The predetermined probability value and the error range are set to predetermined values. The predetermined probability value and the error range are related to a character recognition rate and a determination validity. If the error range is decreased, a character recognition precision is increased. If the error range is increased, an error detection amount is decreased. If it is determined that the error is present in the recognition of the characters, error detection data is output.

If the error is not present or an error detection function is not given, a character corresponding to the first probability value of the probability values output in operation S50 is selected in operation S60. The sequentially selected recognized characters are arranged in order to be generated as a character sequence. In operation S70, the character sequence is converted into serial number recognition data, and the serial number recognition data is output. Also, the serial number recognition data may be a character sequence of a serial number recognized from paper money or a value corresponding to the character sequence. FIG. 5 is a schematic block diagram illustrating a configuration of an apparatus for recognizing a serial number of paper money according to an embodiment of the present invention. Referring to FIG. 5, a serial number recognizing apparatus 100 includes a preprocessor 110, a neural network function processor 120, an error detector 130, and a storage unit 140. The preprocessor 110 receives an image of paper money from an input unit 200 and processes the image. The neural network function processor 120 applies the processed image to a neural network. The error detector 130 detects an error present in a recognition of characters. The storage unit 140 stores the processed image and basic character data.

The serial number recognizing apparatus 100 receives the image of the paper money from the input unit 200, recognizes a serial number of the paper money, and outputs the serial number and an error detection signal to an output unit 300.

The preprocessor 110 obtains the scanned image of paper money from the input unit 200, extracts an outer line of the paper money, calculates a center of the paper money, extracts an image of a portion on which the serial number is inscribed based on the center, generates pixel values of the extracted image as input data, and stores the input data in the storage unit 140.

The neural network function processor 120 substitutes the input data for a neural network function which has learnt basic data of each character, outputs probability values of the basic character data, and arranges the probability values in a descending order. The neural network function processor 120 arranges characters corresponding to the basic character data corresponding to the first probability value and outputs a character sequence to the output unit 300.

The error detector 130 receives the probability values from the neural network function processor 120, calculates errors of the probability values to determine a validity, and outputs an error detection signal to the output unit 300.

The operation of the serial number recognizing apparatus 100 according to the present embodiment will now be described. The preprocessor 110 receives the image of the paper money from the input unit 200, converts the image into a digital image, and extracts the image of the portion on which the serial number is inscribed. The preprocessor 110 divides the extracted image into images of characters, generates the input data using pixel values of the images of the characters, and stores the input data in the storage unit 140. The neural network function processor 120 reads the input data from the storage unit 140, inputs the input data into the neural network function, arranges function values output from the neural network in a descending order, outputs the function values to the error detector 130 to determine whether the recognition of the characters is valid, and if the recognition of the characters is valid, transmits a character sequence corresponding to the corresponding function values to the output unit 300. Also, if the error d€∑tector 130 determines that the error is present in the recognition of the characters, the error detector 130 outputs a serial number recognition error signal to the output unit 300.

FIG. 6 is a schematic view illustrating a configuration of a serial number recognition paper money counter according to an embodiment of the present invention. Referring to FIG. 6, the serial number recognition paper money counter includes an inlet

10, a counter 20, an outlet 30, a display 40, a scanner 50, and a serial number recognizer 60. Paper money is put into the inlet 10. The counter 20 counts the paper money. The outlet 30 discharges the counted paper money. The display 40 displays information as to the counted paper money. The scanner 50 scans an image of the paper money. The serial number recognizer 60 recognizes a serial number of the paper money.

The inlet 10 has a shape so as to accommodate a plurality of pieces of paper money.

The counter 20 counts a number of the plurality of pieces of paper money. A roller that is rotating separates each piece from the plurality of pieces of paper money to count the number of the plurality of pieces of paper money.

The outlet 30 has a shape of a stand case in which each of the counted plurality of pieces of paper money is discharged and accumulated.

The display 40 is a display window displaying serial numbers of the plurality of pieces of paper money and information as to the counted number.

The scanner 50 includes an image sensor scanning images of the plurality of pieces of paper money. The serial number recognizer 60 recognizes the serial numbers using the images extracted by the scanner 50. The serial number recognizer 60 includes a function processor, a preprocessor, and a storage unit. The function processor recognizes the serial numbers using a neural network function. The preprocessor processes the extracted images to be input into the function processor to generate input data. The storage unit stores the input data and data of the function processor.

The operation of the serial number recognition paper money counter will now be described. If a bundle of paper money is input through the inlet 10, the scanner 50 extracts images of the paper money. The extracted images are transmitted to the serial number recognizer 60, and the paper money which has passed the scanner 50 moves to the counter 20. The serial number recognizer 60 recognizes a character sequence of a serial number of the paper money and d transmits information as to the serial number of the paper money to the display 40. The paper money moved to the counter 20 is continuously counted and discharged to the outlet 30 to be accumulated as a bundle of paper money. Here, information as to the counted paper money is transmitted to the display 40. The display 40 converts the information transmitted from the serial number recognizer 60 and the counter 20 into information which can be checked by a user and displays the converted information.

MODE OF THE INVENTION According to an aspect of the present invention, there is provided a method of recognizing a serial number of paper money, including: receiving an image of paper money including a plurality of figures and a plurality of characters and extracting an image of a character sequence of a portion on which the serial number is inscribed; dividing the image of the character sequence into character images having a predetermined pixel size; allocating the character images to an area having a predetermined pixel size, converting values of quantitated pixels of the area into numerical values, and arranging the numerical values in a series of numerical sequences in order to generate input data; inputting the input data into a neural network which has learnt a plurality of pieces of basic character data of each character, so as to allow the input data to coincide with the plurality of pieces of basic character data of each character and outputting probability values depending on the number of cases of the plurality of pieces of basic character data; and arranging the probability values output from the neural network in descending order, selecting a case according to first basic character data having the highest probability value, sequentially arranging characters corresponding to the case, and outputting the characters as a character sequence.

The method may further include determining whether an error is present using first and second probability values of the probability values in order to determine the validity of a recognition of characters and outputting error detection data.

If the first probability value is less than or greater than a predetermined reference value and an error between the first and second probability values is within a predetermined range, it may be determined that the error is present. The plurality of pieces of basic character data may be obtained through combinations of numerals, characters, symbols, and reverse characters.

A character image of the area may be divided into blocks having predetermined sizes, pixel values of the blocks may be averaged, and input data may be generated using average values of the blocks. The receiving of the image of the paper money comprising the plurality of figures and the plurality of characters and the extracting of the image of the character sequence of the portion on which the serial number is inscribed may include: obtaining the image of the paper money; extracting an outer line of the obtained image; and extracting an area in which the serial number is positioned as the character image based on a center of the image.

According to another aspect of the present invention, there is provided an apparatus for recognizing a serial number of paper money, including: a preprocessor receiving an image of paper money including a plurality of figures and a plurality of characters, extracting an image of a character sequence of a portion on which the serial number is inscribed, dividing the image into character images, and generating pixel values of the character images as input data; a function processor inputting the input data into a neural network which has learnt a plurality of pieces of basic character data of each character so as to allow the input data to coincide with the plurality of pieces of basic character data, outputting probability values depending on a number of cases of each character, arranging the probability values in descending order, selecting a case depending on first basic character data having the highest probability value, sequentially arranging characters corresponding to the case, and outputting the characters as a character sequence; and a storage unit storing the input data and basic character data preprocessed by the preprocessor.

The apparatus may further include an error detector determining whether an error is present, using first and second probability values of the probability values output from the function processor and outputting error detection data in order to determine a validity of a recognition of characters.

According to another aspect of the present invention, there is provided a serial number recognition paper money counter in a paper counter including an inlet into which paper money is put, a counter counting the number of paper money, an outlet discharging the paper money, and a display displaying information about the counted paper money, including: a scanner scanning an image of the paper money put through the inlet; a preprocessor receiving the image of the paper money through the scanner, extracting an image of a character sequence of a portion on which a serial number is inscribed, dividing the image into character images, allocating the character images to an area having a predetermined pixel size, converting values of quantitated pixels of the area into numerical values, and arranging the numerical values in a series of numerical sequences in order to generate input data; a function processor inputting the input data into a neural network which has learnt a plurality of pieces of basic character data of characters of the paper money so as to allow the input data to coincide with the plurality of pieces of basic character data, outputting probability values depending on a number of cases of the characters of the plurality of pieces of basic character data, arranging the probability values in descending order, selecting a case depending on first basic character data having the highest probability value, sequentially arranging characters corresponding to the case, and outputting the characters as a character sequence; and a storage unit storing the input data and the plurality of pieces of basic character data preprocessed by the preprocessor.

According to another aspect of the present invention, there is provided a computer-readable recording medium having embodied thereon a computer program for executing the method.

Claims

1. A method of recognizing a serial number of paper money, the method comprising: receiving an image of paper money comprising a plurality of figures and a plurality of characters and extracting an image of a character sequence of a portion on which the serial number is inscribed; dividing the image of the character sequence into character images having a predetermined pixel size; allocating the character images to an area having a predetermined pixel size, converting values of quantitated pixels of the area into numerical values, and arranging the numerical valuers in a series of numerical sequences in order to generate input data; inputting the input data into a neural network which has learnt a plurality of pieces of basic character data of each character, so as to allow the input data to coincide with the plurality of pieces of basic character data of each character and outputting probability values depending on the number of cases of the plurality of pieces of basic character data; and arranging the probability values output from the neural network in descending order, selecting a case according to first basic character data having the highest probability value, sequentially arranging characters corresponding to the case, and outputting the characters as a character sequence.

2. The method of claim 1 , further comprising determining whether an error is present using first and second probability values of the probability values in order to determine the validity of a recognition of characters and outputting error detection data.

3. The method of claim 2, wherein if the first probability value is less than or greater than a predetermined reference value and an error between the first and second probability values is within a predetermined range, it is determined that the error is present.

4. The method of claim 1 , wherein the plurality of pieces of basic character data are obtained through combinations of numerals, characters, symbols, and reverse characters.

5. The method of claim 1 , wherein a character image of the area is divided into blocks having predetermined sizes, pixel values of the blocks are averaged, and input data is generated using average values of the blocks.

6. The method of claim 1 , wherein the receiving of the image of the paper money comprising the plurality of figures and the plurality of characters and the extracting of the image of the character sequence of the portion on which the serial number is inscribed comprises: obtaining the image of the paper money; extracting an outer line of the obtained image; and extracting an area in which the serial number is positioned as the character image based on a center of the image.

7. An apparatus for recognizing a serial number of paper money, comprising: a preprocessor receiving an image of paper money comprising a plurality of figures and a plurality of characters, extracting an image of a character sequence of a portion on which the serial number is inscribed, dividing the image into character images, and generating pixel values of the character images as input data; a function processor inputting the input data into a neural network which has learnt a plurality of pieces of basic character data of each character so as to allow the input data to coincide with the plurality of pieces of basic character data, outputting probability values depending on a number of cases of each character, arranging the probability values in descending order, selecting a case depending on first basic character data having the highest probability value, sequentially arranging characters corresponding to the case, and outputting the characters as a character sequence; and a storage unit storing the input data and basic character data preprocessed by the preprocessor.

8. The apparatus of claim 7, further comprising an error detector determining whether an error is present, using first and second probability values of the probability values output from the function processor and outputting error detection data in order to determine a validity of a recognition of characters.

9. The apparatus of claim 8, wherein if the first probability value is less than or greater than a predetermined reference value and an error between the first and second probability values is within a predetermined range, the error detector determines that the error is present.

10. A serial number recognition paper money counter in a paper counter comprising an inlet into which paper money is put, a counter counting the number of paper money, an outlet discharging the paper money, and a display displaying information about the counted paper money, comprising: a scanner scanning an image of the paper money put through the inlet; a preprocessor receiving the image of the paper money through the scanner, extracting an image of a character sequence of a portion on which a serial number is inscribed, dividing the image into character images, allocating the character images to an area having a predetermined pixel size, converting values of quantitated pixels of the area into numerical values, and arranging the numerical values in a series of numerical sequences in order to generate input data; a function processor inputting the input data into a neural network which has learnt a plurality of pieces of basic character data of characters of the paper money so as to allow the input data to coincide with the plurality of pieces of basic character data, outputting probability values depending on a number of cases of the characters of the plurality of pieces of basic character data, arranging the probability values in descending order, selecting a case depending on first basic character data having the highest probability value, sequentially arranging characters corresponding to the case, and outputting the characters as a character sequence; and a storage unit storing the input data and the plurality of pieces of basic character data preprocessed by the preprocessor.

11. A computer-readable recording medium having embodied thereon a computer program for executing the method of any one of claims 1 to 6.