KR100719608B1 - Method and apparatus for recognizing serial number of paper money - Google Patents

Abstract

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and apparatus for recognizing a serial number of a banknote in a banknote counter or a counterfeit discriminator.

A method for recognizing a serial number according to the present invention comprises the steps of: receiving a bill image including a plurality of figures and characters and extracting a character string image of a portion where a serial number is displayed; Dividing the extracted character image into individual characters and dividing the extracted character image into character images having a pixel size of a predetermined range; Assigning the character images to a region having a predetermined pixel size, digitizing the values of the quantized pixels in the region, and arranging the quantized pixels into a series of numeric columns to generate each input data; Inputting the input data into a neural network that has already learned about the basic character data to provide a degree of matching between the input data and a plurality of basic character data classified by each character, Outputting a result of each probability value according to the number of cases of each character by the number; And selecting a case corresponding to the first order basic character data output from the neural network, and arranging the corresponding characters in order and outputting them as a character string.

Banknote, serial number, neural network, recognition

Description

TECHNICAL FIELD The present invention relates to a method and apparatus for recognizing a serial number of a paper currency,

BRIEF DESCRIPTION OF THE DRAWINGS The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate preferred embodiments of the invention and, together with the description of the invention given below, serve to further the understanding of the technical idea of the invention, And should not be construed as limiting.

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

FIG. 2 is an exemplary diagram illustrating a process of extracting a character string of a portion having a serial number in FIG. 1 together with a screen. FIG.

FIG. 3 is an exemplary diagram showing a character image extracted in FIG. 2 and divided parts. FIG.

Fig. 4 is a conceptual diagram showing the relationship between the output of the neural network and error judgment in Fig. 1. Fig.

5 is a schematic block diagram showing a configuration of a serial number recognizing apparatus according to an embodiment of the present invention.

6 is a schematic block diagram showing a configuration of a serial number-recognizing bill counter according to another embodiment of the present invention.

DESCRIPTION OF THE REFERENCE NUMERALS OF THE DRAWINGS

10 ... inlet 30 ... outlet

40 ... display unit 50 ... scan unit

100 ... serial number recognizing device 110 ... preprocessing section

120 ... Neural network function processor 130 ... Error detector

140 ... storage unit 200 ... input unit

300 ... output section

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and apparatus for recognizing a serial number of a banknote, and more particularly, to a method and apparatus for extracting and processing an image of a banknote in a banknote counter or a counterfeit discriminating apparatus to recognize a serial number.

The bill counter is for counting the number of bills in a banknote or the like. The conventional bill counter has merely functions for counting bills. However, in recent years, a bill counter with a function of discriminating authenticity of bills, Have been proposed.

The conventional method of recognizing the serial number of the banknote counter equipped with the serial number recognizing function includes a method of scanning the serial number display portion of the banknote as it is or using a method of detecting the character by detecting only the fragmentary characteristic of the serial number image Respectively. Also, as a more accurate serial number recognition method, a method of extracting a serial number image of a banknote and comparing it with a basic character image and recognizing the character string is used.

However, the conventional serial number recognition method has a problem that it is inconvenient to process the recognized serial number because the scanned image is outputted as it is, and the method of detecting and recognizing only the fragmentary characteristic of the character image has a problem of low recognition rate.

In addition, the method of recognizing the serial number string image of a bill by comparing it with the basic character image is a problem in which it is time-consuming and difficult to recognize freely in a universal use environment due to the deformation of the image depending on the environment of the damaged bills or the scan .

SUMMARY OF THE INVENTION The present invention has been devised to solve the problems described above, and it is an object of the present invention to provide a method and apparatus for recognizing serial numbers faster and more accurately by constructing a recognition algorithm using a mathematical process when acquiring images of banknotes and recognizing serial numbers .

Another object of the present invention is to provide a serial number recognition method and apparatus capable of determining the validity of character recognition by imposing an error detection function in the process of recognizing serial numbers of bills.

According to another aspect of the present invention, there is provided a method for recognizing a banknote serial number, comprising the steps of: (a) receiving a banknote image including a plurality of figures and characters, and extracting a character string image of a serial number; (b) dividing the extracted character image into individual characters and dividing the extracted character images into character images having a pixel size of a predetermined range; (c) assigning the character images to a region having a predetermined pixel size, digitizing the values of the quantized pixels in the region, and arranging the quantized pixels into a series of numeric columns to generate each input data; (d) inputting the input data to a neural network that has already learned about the basic character data to provide a degree of matching between the input data and a plurality of basic character data classified by each character, Outputting a result as each probability value according to the number of cases for each character by character data; And (e) arranging the probability values according to each character case output from the neural network in descending order, selecting a case according to first rank basic character data having the highest probability value, and arranging the corresponding characters in order And outputting it as a character string.

Determining whether the character recognition is valid by using a first rank probability value and a second rank probability value among the respective probability values and outputting the determined error probability data as error detection data in step (e) .

Preferably, the determination of whether or not the error exists may be performed when the first rank probability value is less than a predetermined reference value and when the first rank probability value is greater than a predetermined reference value, In this case, it is recognized as an error.

Preferably, the plurality of basic character data include a case where a plurality of basic character data is obtained by a combination of numbers, characters, symbols, or inverted characters.

In addition, it is preferable that the character image of the divided area is divided into a predetermined range of blocks, the pixel values of the corresponding block are averaged, and input data is generated using the average values of the blocks.

In addition, the step (a) includes a banknote image acquisition step, an extraction of an outline of the acquired image, and a step of extracting an area in which the serial number is located based on the center point of the image, as a character string image.

According to another aspect of the present invention, a banknote image including a plurality of graphics and characters is received, and a character string image of a portion where a serial number is displayed is extracted and divided into respective character images, A preprocessing unit for generating input data of the input data; Inputting the input data to a neural network that has already learned about the basic character data to provide degree of correspondence between the input data and a plurality of basic character data classified by character, And outputs the result to each probability value according to the number of cases of each character by the character string, and arranges the probability values according to the case of each character in descending order and according to the first order basic character data having the highest probability value A function processor for arranging characters corresponding to the selected characters and outputting them as a character string; And a storage unit for storing the input data.

Preferably, a function of determining whether an error is detected by using a first rank probability value and a second rank probability value among the respective probability values output from the function processing unit to determine the validity of the character recognition, and outputting the error detection data And an error detection unit for detecting the error.

According to another aspect of the present invention, there is provided a bill counter comprising a bill input port for inputting bills, a counting section for counting the number of bills inserted, a discharge port for discharging bills out of the apparatus, and a display section for displaying information of the counted bills A scan unit for scanning an image of a bill inserted through the input port, and a character image extracting unit for extracting a character string image of a portion where the serial number is displayed by receiving the bill image through the scan unit, dividing the character image into respective character images, A preprocessing unit for assigning the numerical values of the quantized pixels in the region to numerical values of the quantized pixels, and arranging the numerical values in a series of numeric columns to generate input data; Inputting the input data to a neural network that has already learned about the basic character data to provide degree of correspondence between the input data and a plurality of basic character data classified by each character of a banknote, Outputting the results to respective probability values according to the number of cases for each character by character data, sorting the probability values according to each character case in descending order, and outputting the first ranked basic character data having the highest probability value A function processor for selecting a case corresponding to the character string and outputting the character string corresponding to the selected character string; And a storage unit for storing the input data.

According to still another aspect of the present invention, there is provided a recording medium loaded on a device for recognizing a banknote serial number in a counterfeit discriminating device for discriminating a banknote counter counting the number of banknotes, A step of dividing the extracted character image into character images each having a pixel size of a predetermined range by dividing the extracted character image into individual characters, A step of assigning the character images to a region having a predetermined pixel size, digitizing values of quantized pixels in the region, and arranging the quantized pixels into a series of numeric columns to generate respective input data; A plurality of basic character data for each of the plurality of basic character data, Inputting the input data into a neural network that has already been subjected to wet processing and outputting a result to each probability value according to the number of cases of each character by the basic character data, A program for sorting the probability values according to the case in descending order and selecting a case based on the first rank basic character data having the highest probability value and arranging the corresponding characters in order and outputting them as a character string A computer-readable recording medium is provided.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. Prior to this, terms and words used in the present specification and claims should not be construed as limited to ordinary or dictionary terms, and the inventor should appropriately interpret the concepts of the terms appropriately It should be interpreted in accordance with the meaning and concept consistent with the technical idea of the present invention based on the principle that it can be defined. Therefore, the embodiments described in this specification and the configurations shown in the drawings are merely the most preferred embodiments of the present invention and do not represent all the technical ideas of the present invention. Therefore, It is to be understood that equivalents and modifications are possible.

FIG. 1 is a flowchart illustrating a process of recognizing a serial number of a bill according to an exemplary embodiment of the present invention.

Referring to FIG. 1, a method of recognizing serial numbers of banknotes according to an embodiment of the present invention starts with a banknote image input step (S10).

In the banknote image input step S10, banknotes including letters, numbers, symbols, figures, and the like are inputted into an image generated by scanning using a video image sensor. Here, bills are not limited to banknotes, but may include checks, lotteries, securities, gift certificates, and the like. In addition, when a bill is scanned, it can be converted into a digital image through an analog-to-digital converter (A / D converter) and input.

In the string image extracting step (S20), a process of extracting a character string of a portion in which a serial number, which is a unique identification number of the paper currency, is displayed in the image obtained in the step S10 is performed. In FIG. 2, the process of extracting the character string image of the portion having the serial number (S20) is shown in detail along with the banknote image. Referring to FIG. 2, first, in step S21, the image of the bill received in step S10 is acquired. The acquired image includes not only an image of a bill but also a portion other than a bill. In order to remove such an unnecessary part image, the outline of the banknote is extracted (step S22). In this case, only the area enclosed by the outline may be used, or only the string portion in which the serial number of the banknote is displayed may be extracted and used so as to further reduce unnecessary processing. In this embodiment, a process of extracting and using only the string portion will be described. A bill center point to be used as a reference point is calculated in order to grasp the position of the string portion in which the serial number is indicated (step S23). A direction and a distance are calculated based on the center point to calculate position coordinates and a character string image at that position is extracted (step S24).

In the image segmenting step S30, a process of dividing a character string including a serial number into characters is performed. FIG. 3 shows a character image and each divided character image. First, we average the pixels in the horizontal direction in the string image. The coordinates of the point at which the average value of the pixel averages decreases and is calculated. And is divided in the longitudinal direction based on this coordinate. The horizontal direction is also the same as the longitudinal direction division method described above, and each rectangular divided character image is generated.

In step S40, input data is generated so as to substitute each of the character images generated by the division into a function to be described later. All pixel values in the divided character image region may be generated as input data or the character image may be processed to generate input data. When all pixel values in the feature image region are generated as input data, the extracted feature image is allocated to a frame having a pixel size of a predetermined standard so that it can be input to a function to be described later. Each pixel included in the allocated frame is quantified and numerically defined as the value of individual pixels, and the pixels are arranged in a series of numerical sequences. For example, the pixel value quantified and quantified is defined as a value between 0 and 255 expressed in gray scale. In the processing of the character image area, there is a method of dividing the image area into blocks of a predetermined range and averaging the pixel values of the corresponding blocks. There may be a processing method of averaging pixel values in the horizontal or vertical direction of the image area have. If the input data is processed by the above-described method, the amount of data can be reduced, and more precise data can be generated by generating the input data using the pixel value as it is. And generates input data arranged in a series of numbers in the character images divided as described above.

In the application step of the neural network function (S50), the input data obtained in step S40 is inputted to the neural network function. The neural network will not be described in detail with reference to the known technology. First, an image prepared by combining all the cases of number characters, letters, symbols and their inverted inverted phases, which will be the basis of character recognition, is processed through the process of step S40, As shown in FIG. The neural network learns the basic character data and prepares it in advance. Backpropagation algorithm, Boltzmann machine learning method, or simulated annealing learning method can be used for learning of the neural network. When the input data is substituted into the already learned neural network function, a probability value corresponding to each case according to basic character data in all cases is output.

In step S80, the validity of the probability output in step S50 is determined, and when the validity is insufficient, an error signal is output. The process of determining the validity and detecting an error is schematically shown in Fig. Referring to FIG. 4, input data is input to a neural network function, and a probability value is output for each basic character. The respective probability values are sorted in descending order so that the first rank probability value and the second rank probability value are subjected to the validity judgment. First, when the first rank probability value does not exceed the probability value of the certain standard, it is determined as a character recognition error. When the first probability value is greater than or equal to a predetermined criterion value, the second probability value is compared with the first probability value. If the error between the two probability values falls within a certain range, the character recognition based on each probability value can not be relied upon and is determined as an error. If the first rank probability value exceeds a certain standard value and the error with the second rank probability value also deviates from a certain range, it is determined that character recognition is reliable. The predetermined reference probability value and the error range of the two probability values are preset to a predetermined value. The reference probability value and the error range are related to the character recognition rate and the judgment effectiveness. When the error range is reduced, the character recognition accuracy is increased. When the range is extended, the error detection amount is decreased. And outputs error detection data if it is determined to be an error as a result of the validity determination.

The recognizing character array step S60 selects a character corresponding to the first probability value among the probability values output in step S50 when there is no error or the error detection function is not imposed. The recognition characters sequentially selected are sequentially arranged to generate a character string.

Converts the arrayed character string into serial number recognition data and outputs corresponding recognition data (step S70). In addition, the recognition data may be a character string of a serial number recognized in a banknote, or may be a value corresponding to a character string.

FIG. 5 schematically shows a configuration of a serial number recognizing apparatus according to an embodiment of the present invention.

5, a serial number recognizing apparatus 100 according to an embodiment of the present invention includes a preprocessing unit 110 for receiving a banknote image from an input unit 200 and processing an image, A neural network function processor 120, an error detector 130 for detecting errors in character recognition, and a storage unit 140 for storing the processed data.

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

The preprocessing unit 110 extracts the banknote image scanned from the input unit 200, extracts the banknote outline, calculates the center point of the banknote, extracts the image of the serial number on the basis of the center point, And stores the generated input data in the storage unit 140.

The neural network function processor 120 substitutes the input data into the neural network function that has already been learned as the basic data for each character according to each case, outputs the probability data corresponding to each basic character data, Sort in order. The output unit 300 arranges the characters corresponding to the basic character data corresponding to the first rank probability value and outputs a character string.

The error detection unit 130 receives a probability value according to each case from the neural network function processing unit 120 and calculates error of the probability values to determine validity and outputs an error detection signal to the output unit 300 do.

The preprocessing unit 110 receives the image of the banknote from the input unit 200, converts the image into a digital image, and extracts an image of the portion having the serial number. The extracted image is divided for each character, input data is generated by the pixel value of the divided image, and then the input data is stored in the storage unit 140. The neural network function processing unit 120 reads the input data stored in the storage unit 140 and inputs the input data to the neural network function. The output function values are sorted in descending order and sent to the error detection unit 130 to determine validity. A character string corresponding to the corresponding function value is transmitted to the output unit 300. Also, if the error detection unit 130 determines that the validity is determined to be an error, the serial number recognition error signal is transmitted to the output unit 300.

6 is a schematic block diagram showing a configuration of a serial number-recognizing bill counter according to another embodiment of the present invention.

Referring to FIG. 6, the serial number recognition bill counter according to another embodiment of the present invention includes an input port 10 for inputting banknotes, a counting unit (not shown) for counting banknotes, (Not shown) for recognizing the serial number of the banknote, a display unit 40 for displaying the information of the banknote counted, a scanning unit 50 for scanning the image of the inputted banknote, .

The bill insertion slot 10 is configured to receive a bank-note bundle composed of a plurality of bills.

The counting unit counts the number of banknotes of the banknote bundle. The counting unit counts the number of banknotes at the same time as scanning the banknote images by the scanning unit 50 to be described later.

The discharge port 30 is formed in the form of a tray for collecting and discharging the counted bill in a single sheet.

The display unit 40 is a display window in which the serial number of the banknote and the counted numerical information are displayed.

The scanning unit 50 includes a video image sensor for scanning an image of a bill.

The serial number recognizing unit is located on the substrate on which the scanning unit 50 is installed and recognizes the serial number using the image extracted from the scanning unit 50. The serial number recognizing unit includes a function processing unit for recognizing the serial number using a neural network function, a preprocessor for generating input data by processing the image extracted from the scanning unit 50 for input to the function processing unit, For example.

When the banknote bundle is inserted through the input port 10, the scan unit 50 extracts the image of the banknote. The extracted banknote images are transmitted to the serial number recognizing unit and the banknotes that have passed through the scanning unit 50 are moved to the counting unit. The character string of the serial number is recognized in the serial number recognition unit of the transferred banknote image, and the serial number information of the corresponding banknote is transmitted to the display unit 40. The bill moved to the counting portion is counted continuously and discharged to the discharge port 30 to be stacked again. At this time, the counted information is transmitted to the display unit 40. The information transmitted from the serial number recognizing unit and the counting unit is converted into information that the user can confirm on the display unit 40 and displayed.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not to be limited to the details thereof and that various changes and modifications will be apparent to those skilled in the art. And various modifications and variations are possible within the scope of the appended claims.

According to the present invention, by using a function in a method of recognizing a serial number after receiving and processing a bill image, a serial number can be recognized as an output value at the same time as an input, Lt; / RTI >

In addition, in the serial number recognition process, the output values of the function are compared to determine the validity of the serial number recognition, and the error is checked to improve the reliability of the serial number recognition.

Claims (11)

(a1) a step 1-1 of obtaining a bill image including a plurality of figures and characters; (a2) extracting an outline of the obtained banknote image; (a3) searching for a region in which the serial number is located based on the center point of the banknote image; (a4) extracting, as a character string image, an area of the banknote image in which the serial number is displayed; (b) dividing the extracted character string image by each character and dividing the character-by-character image into blocks having a pixel size of a predetermined range; (c) averaging pixel values within the block for each of the divided blocks and generating input data for each character-specific image using the average values of the blocks; (d1) to provide a degree of correspondence between the input data and a plurality of basic character data classified by each character of the banknote, the learning of the basic character data is performed by a neural network function pre- 4-1 < / RTI > (d2) outputting each probability value according to the number of cases of each character by the basic character data as a result from the neural network function; And (e1) a step (5-1) of arranging the probability values according to each character case outputted from the neural network function in descending order; (e2) determining whether the first rank probability value is smaller than a predetermined reference value by using the first rank probability value and the second rank probability value among the respective probability values to determine the validity of the character recognition, Comparing the second probability value with a predetermined reference value, and determining that the recognition error is a character recognition error if it is within a predetermined error range; And (e3) outputting error detection data when it is determined as a character recognition error in the step 5-2, selecting basic character data corresponding to the first rank probability value if it is determined as normal recognition, and And arranging the characters in order and outputting them as a character string, In the steps 4-1 and 4-2, the plurality of basic character data includes numbers, letters, symbols, and combinations of numbers, letters, and symbols inverted in upper and lower bands. Way. delete delete delete delete delete Extracting an outline of the obtained banknote image, finding an area in which the serial number is located based on the center point of the banknote image, and then extracting the serial number from among the banknote images Extracting a character string image from the displayed area, dividing the extracted character string image into individual characters, dividing the character-by-character image into blocks having a pixel size within a predetermined range, A preprocessing unit for averaging pixel values of each block and generating input data for each character-by-character image using the average values of the blocks; Inputting the input data to a neural network function preliminarily learning about the basic character data to provide degree of correspondence between the input data and a plurality of basic character data classified by each character of a banknote, Each of the probability values corresponding to the number of cases of each character by the basic character data is outputted as a result from the neural network function, and the probability values according to the case of each character are sorted in descending order. If it is determined that the character recognition is normal A function processor for selecting a case corresponding to the first basic character data having the highest probability value and arranging the corresponding characters in order and outputting them as a character string; And determining whether the first rank probability value is less than a predetermined reference value and determining whether the first rank probability value is less than a predetermined reference value if the first rank probability value is less than a predetermined reference value, An error detector for determining a character recognition error and outputting error detection data when the second comparison probability is greater than a predetermined reference value and is within a predetermined error range; And And a storage unit for storing the input data, Wherein the plurality of basic character data includes numbers, letters, symbols, and a combination of numbers, letters, and symbols inverted up / down. delete delete 1. A bill counter comprising: a charging port for charging bills; a counting unit for counting the number of bills inserted; an outlet for discharging bills out of the apparatus; and a display unit for displaying information on the counted bills, A scan unit for scanning an image of a bill inserted through the input port, Extracting an outline of the obtained banknote image, finding an area in which the serial number is located based on the center point of the banknote image, extracting a character string image of the area where the serial number is displayed among the banknote images, For each character, dividing the image for each character into blocks having a pixel size of a predetermined range, and averaging the pixel values within the block for each of the divided blocks, A preprocessing unit for generating input data for a star image; Inputting the input data to a neural network function preliminarily learning about the basic character data to provide degree of correspondence between the input data and a plurality of basic character data classified by each character of a banknote, Each of the probability values corresponding to the number of cases of each character by the basic character data is outputted as a result from the neural network function, and the probability values according to the case of each character are sorted in descending order. If it is determined that the character recognition is normal A function processor for selecting a case corresponding to the first basic character data having the highest probability value and arranging the corresponding characters in order and outputting them as a character string; And determining whether the first rank probability value is less than a predetermined reference value and determining whether the first rank probability value is less than a predetermined reference value if the first rank probability value is less than a predetermined reference value, An error detector for determining a character recognition error and outputting error detection data when the second comparison probability is greater than a predetermined reference value and is within a predetermined error range; And And a storage unit for storing the input data, Wherein the plurality of basic character data includes numbers, characters, symbols, and a combination of numbers, letters, and symbols inverted up / down. delete

Images