JP5604284B2 - Paper piece identification system, paper piece identification device - Google Patents

Description

  The present invention relates to a paper piece registration / identification method and a paper piece registration / identification method for authenticating a piece of paper. Hereinafter, a piece of paper is sometimes simply referred to as paper.

  In recent years, advanced counterfeit tickets such as banknotes, securities, and stamps can be easily created due to improvements in copying technology, which has become a social problem. In general, forgery prevention requires squeezing technology and special ink, which increases printing costs.

  For example, Patent Document 1 discloses a method for printing characters of securities with magnetic ink and preventing copying by a copying machine. Patent Document 2 discloses a method for authenticating by inserting a magnetic material. Patent Document 3 discloses a method of resisting deterioration of a paper piece in a paper piece discrimination method using transmitted light. Patent Document 4 discloses a method for registering a paper piece including tag information of a paper piece input from a keyboard or the like in addition to the feature amount of the paper piece based on the transmission image in the paper piece discrimination method using transmitted light. Patent Document 5 discloses a method that eliminates the need for strict adjustment of the illumination direction and the position / posture of the paper in the paper piece discrimination method using transmitted light. Patent Document 6 discloses a method of printing a two-dimensional code corresponding to a condition when the reading condition is to be changed during operation in an authenticity determination method using reflected light of a piece of paper.

JP 2010-9482 JP2007-164293 JP2007-11700 JP 2006-18525 A JP 2006-338330 A JP 2006-338331 A

  The method described in Patent Document 1 or 2 is an effective method for preventing forgery. However, there is a problem that it is necessary to use special ink or to insert special fibers into paper, which increases the cost of preventing forgery. On the other hand, the methods described in Patent Documents 3 to 6 do not require any special processing on the paper, so that the cost of preventing forgery can be suppressed. However, the methods described in Patent Documents 3 to 6 have the following problems.

  The first problem is how to determine the authenticity criteria for judging whether a given paper is genuine or fake. Generally, the rate at which genuine paper is mistakenly determined to be fake paper (hereinafter also referred to as false rejection rate, FRR) and the rate at which false paper is mistakenly determined to be genuine paper (hereinafter referred to as false alarm) Rate, FAR) is in a trade-off relationship, and it is difficult to make it completely zero. Therefore, when designing a system that determines the authenticity of paper, the false rejection rate and false alarm rate are set to certain values so that there are no operational problems. When judging the authenticity of paper, the numerical value is used as a reference value for authenticity judgment. Hereinafter, this numerical value is referred to as a threshold value, and determining the authenticity determination result based on the threshold value is referred to as threshold processing.

  On the other hand, even if the paper is printed for the same application, the composition may change depending on the time of papermaking, even when using paper with a different composition for each papermaking company. . In the methods described in Patent Documents 3 to 6, a reflected image or a transmitted image of paper is collected and the authenticity is determined using these images. However, the influence of variations in the composition of the paper on these images, There is no disclosure about a method of performing authentication with high accuracy against such variations. In addition, fluctuations due to individual differences of scanners that collect reflected images and transmitted images are not disclosed in the conventional method.

  Patent Document 4 describes a method for determining the authenticity of a paper by threshold processing. However, only a method for determining one value in advance by experiment is disclosed as a method for determining a threshold, and this method corresponds to the above variation. I can't.

  The second problem is how to determine the authenticity with high accuracy when the paper characteristics change due to aging. After paper making, the paper surface or internal properties change due to wear, bending, dirt, and the like. In addition, the color of ink printed on the paper surface or the color of fibers inside the paper changes due to exposure. In the methods described in Patent Documents 3 to 7, a reflection image or a transmission image of paper is collected and the authenticity is determined using these images. However, the reflection image and the transmission image also change due to such a secular change. , It affects the accuracy of paper authenticity judgment.

  Patent Document 3 discloses a method of taking a frequency distribution of feature values and measuring the degree of paper stains and deterioration based on the degree of distortion (distortion degree) of the distribution shape. That is, a method is disclosed in which another portion of paper having a low degree of distortion is used for authenticity determination based on the assumption that the degree of deterioration is high when the degree of distortion is large with respect to the transmitted image. However, this method cannot cope with the discoloration of the ink due to abrasion, folding, dirt, and exposure throughout the paper. Further, the same patent document discloses a method in which a pixel having a pixel value far from the average value of the transmission image is not used for authenticity determination. However, when deterioration is severe, there is a problem that most pixels cannot be used for authenticity determination.

  The third problem is how to make authenticity determination with high accuracy when characters or photos are printed on paper, characters are handwritten, or stamps are pressed. It is. When a transmission image of a piece of paper is observed, it can be seen that the difference between the pixel value of the portion where the characters are printed and the plain pixel value is larger than the distribution width of the pixel value of the plain portion. Therefore, in order to accurately determine the authenticity of the paper, it is necessary to accurately align the printed characters between the paper to be determined authenticity and the real paper. However, when a transmission image is picked up by a scanner, there may be a shift of several pixels due to the accuracy of the paper transport system. Even if the scanning position is accurately adjusted using a highly accurate paper conveyance system, it is inevitable that the character width in the image fluctuates by several pixels due to fluctuations in the photoelectric conversion characteristics of the scanner. As a result, the pixel values used for authenticity determination may differ greatly, and in this case, real paper may be erroneously determined to be fake. In the methods described in Patent Documents 3 to 6, there is no disclosure regarding authenticity determination using a plain paper portion or printing of characters and photographs.

  The fourth problem is how large the image used for authenticity determination, that is, how many pixels should be used. If the number of pixels is increased, a more precise distribution of pixels can be obtained, so it is expected that false alarms will be reduced, but conversely, the influence of paper misalignment during scanning will increase and false rejection will increase. It is done. In addition, the processing time increases as the number of pixels increases. Therefore, it is common to find and adjust the false rejection rate and false alarm rate during preliminary experiments and design.

  However, as described in the first problem, even if paper is used for the same purpose, the composition of the paper varies depending on the paper manufacturer, and may vary depending on the timing of the paper. As a result, the transmission image and reflection image of the paper fluctuate, and the number of pixels determined during preliminary experiments and design may not be optimal. Patent Document 6 discloses a method of dynamically changing image reading conditions and authenticity determination conditions. However, only a method for manually specifying the number of sensors to be used and the gradation of an image from a computer screen has been disclosed, and the optimum number of pixels could not be set automatically.

  In the paper piece identification system according to the present invention, in the paper piece identification system including a paper piece registration device and a paper piece identification device, the paper piece registration device includes a first feature quantity extraction processing extraction unit for extracting a feature quantity from the paper piece, and the first piece. A printing unit that prints information on the feature amount extracted by the feature amount extraction processing unit on the piece of paper, and the paper piece identification device includes a second feature amount extraction processing extraction unit that extracts the feature amount from the piece of paper. A feature amount acquisition unit that acquires information about a feature amount printed on the piece of paper by the printing unit from a piece of paper, a feature amount decoding unit that extracts a feature amount from the information about the feature amount, and the feature amount decoding unit And a similarity calculation unit for determining a similarity between the feature quantity extracted by the feature quantity extracted by the second feature quantity extraction unit.

  According to a second aspect of the present invention, there is provided a parameter estimator that estimates one or more of the unique characteristics of the paper piece registration device, the unique characteristics of the paper piece identification device, and the unique characteristics of the paper piece. The authenticity determination unit determines the authenticity of the piece of paper based on the similarity and the characteristics estimated by the parameter estimation unit.

  Furthermore, in the third aspect of the present invention, the paper piece registration apparatus includes a statistic calculation unit that calculates a statistic of the feature amount from the feature amount extracted by the first feature amount extraction processing unit. The printing unit prints information relating to the statistic calculated by the statistic calculating unit on a piece of paper, and the paper piece identifying device extracts the statistic from the information relating to the statistic printed on the piece of paper. And a feature amount correction unit that corrects the feature amount extracted by the second feature amount extraction unit based on the decoded statistic, and the similarity calculation unit includes the feature amount decoding unit. The similarity between the feature quantity extracted by the conversion section and the feature quantity corrected by the feature quantity correction section is determined.

  According to the present invention, since it is not necessary to register a piece of paper image in the server, the authenticity determination can be performed at high speed. Further, according to the second aspect of the present invention, even when the composition of the paper fluctuates or when the characteristics of the paper piece registration / identification device fluctuate from one individual to another, the authenticity determination of the paper can be performed with high accuracy. . In addition, according to the third aspect of the present invention, it is possible to accurately determine the authenticity of a paper even when the characteristics of the paper fluctuate with time.

It is a figure which shows the basic structural example of the paper piece registration and identification apparatus used for each Example of this invention. It is a figure which shows the structural example of the paper piece registration and identification system of 1st Example of this invention. It is a figure which shows the structure of the process of the 1st Example of this invention. It is a figure which shows the example of the permeation | transmission image in the 1st Example of this invention. It is a figure which shows the procedure of the paper piece registration process in the 1st Example of this invention. It is a figure which shows the procedure of the feature extraction process in the 1st Example of this invention. It is a figure which shows the procedure of the authenticity determination process in the 1st Example of this invention. It is a figure which shows the structural example of the paper piece registration and identification system of 2nd Example of this invention. It is a figure which shows the structural example of the paper piece registration and identification system of 2nd Example of this invention. It is the figure showing the histogram of the pixel value of the transparent image in the 2nd Example of this invention, and the figure showing the relationship between a histogram and elapsed time. It is the figure showing the histogram of the hue value of the transmission image in the 2nd Example of this invention, and the figure showing the relationship between a histogram and elapsed time. It is a figure which shows the calculation procedure of the statistics of the feature-value in the 2nd Example of this invention. It is a figure which shows the calculation procedure of the statistics of the feature-value in the 2nd Example of this invention. It is a figure which shows the procedure of the authenticity determination process in the 2nd Example of this invention. It is a figure which shows the procedure of the feature extraction process in the 3rd Example of this invention. It is a figure which shows the procedure of the feature extraction process in the 4th Example of this invention.

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

  As a first embodiment of the present invention, a paper piece authenticity determination system will be described below.

  FIG. 1 is a configuration example of a paper piece registration / identification device. In this apparatus, at least one computer, image acquisition means, printing means, and communication means are connected via a bus. That is, the paper piece registration / identification apparatus includes a transmission image acquisition unit 101 that acquires a transmission image of a paper piece, a reflection image acquisition unit 102 that acquires a reflection image of a paper piece, a calculation unit 103, a communication unit 104, and a storage unit 107. , A printing unit 120 and a bus 106, and can communicate with an external device via the network 105. The transparent image is used when extracting a feature amount for identifying a piece of paper. The reflected image is used for recognizing a feature amount printed on a piece of paper as a code, a piece of paper identifier, or a mark. Here, the identifier is a character string unique to each piece of paper. The mark is a mark to be printed on a piece of paper in order to designate a region from which a feature value used for identifying the piece of paper is extracted.

  The transmission image acquisition unit 101 and the reflection image acquisition unit 102 can be realized as an image input terminal for the paper piece registration / identification device, and for example, a commercially available optical scanner can be used.

  The calculating means 103 can be realized as a general-purpose MPU (Micro Processor Unit), and the feature extracting means 111, the parameter estimating means 112, and the similarity calculating means 114 execute necessary calculations.

  The processing for recognizing the identifier and detecting the mark is stored in the storage means 107 as a program. The storage means 107 includes an identifier recognition means 108, mark detection means 109, perturbation processing means 110, feature extraction means 111, parameter estimation means 112, identifier decoding means 113, similarity calculation means 114, identifier issue means 115, feature quantity. A program describing the processing procedures of the decryption means 116, the parameter decryption means 117, the paper piece registration means 118, and the authenticity determination means 119 is stored.

  Each element of the printing unit 120 can be realized by using a commercially available printer or printer module. The mark printing unit 121 is a unit that prints a mark on a piece of paper, the identifier printing unit 122 is a unit that prints an identifier on a piece of paper, the feature amount printing unit 123 is a unit that prints a feature amount on a piece of paper, and a parameter printing unit Is a means for printing parameters on a piece of paper.

  In FIG. 1, the paper piece registration / identification device is described as a device having both the function as the paper piece registration device and the function as the paper piece identification device. However, the paper piece registration device and the paper piece identification device have the respective functions. It may be provided as a separate device.

  FIG. 2 is a configuration example of a paper piece authenticity determination system using a network. In this system, at least one paper piece registration / identification device and a management server for managing paper piece identifiers are connected via a network. That is, this system includes a first paper piece registration / identification device 201, a second paper piece registration / identification device 202, a third paper piece registration / identification device 203, and a paper piece identifier management server 205 having a paper piece identifier management DB 206. 204 is connected.

  FIG. 2 shows an example in which one paper piece identifier management server is connected to a plurality of paper piece registration / identification devices via a network. However, the load on the network 204 is reduced for backup of the paper piece identifier management DB 206. Therefore, a plurality of paper piece identifier management servers may be provided. Further, a configuration may be adopted in which a paper piece identifier management server is provided for each paper piece registration / identification device. Also, registration of a piece of paper may be performed by the first piece of paper registration / identification device 201, and identification of the piece of paper may be performed by a second piece of paper registration / identification device 202. Also, a device for registering a piece of paper and a device for identifying a piece of paper may be separated and connected via a network.

  FIG. 3 is a diagram showing a more specific configuration and flow of each processing program of the first embodiment. As shown in the figure, it is roughly divided into a paper piece registration process 302 and a paper piece identification process 310.

  The paper piece registration process 302 is a process of extracting a characteristic amount unique to a paper piece from a transmission image of the paper piece, quantifying the characteristic quantity, and printing it on the paper piece. The transmission image acquisition unit 303 is a unit for obtaining a transmission image of the paper piece 303. The feature extraction processing unit 304 is a unit that extracts a feature amount from the transmission image. Further, the parameter estimation means 305 estimates the characteristic of the paper piece or paper piece registration / identification device. The feature quantity / parameter / mark printing means 306 is a means for printing feature quantities, parameters, and marks on a piece of paper. As a form of printing, a numerical value may be printed as a character string, or may be printed with a code such as a barcode or a two-dimensional code. The identifier printing unit 308 is a unit that prints the identifier on a piece of paper. The identifier generation unit 307 may generate a character string unique to the paper piece inside the paper piece registration process 302 or may cause the identifier management server 205 to generate a character string unique to the paper piece via the network 105.

  The paper piece identification process 310 is a process for determining the authenticity of a paper piece by extracting the characteristic quantity unique to the paper piece from the transmission image of the paper piece, digitizing the characteristic quantity, and comparing it with the characteristic quantity printed on the paper piece. is there. The reflected image acquisition unit 311 is a unit that obtains a reflected image of the paper piece 309. The parameter decoding unit 312 is a unit that decodes parameters included in the reflected image. When the numerical value is printed as a character string as the feature amount / parameter / mark printing means 306, a known character recognition technique can be used. Even when printing is performed using a code such as a barcode or a two-dimensional code, a known barcode or two-dimensional code reading technique can be used. The feature amount decoding unit 313 is a unit that decodes the standard feature amount from the reflection image of the piece of paper obtained by the reflection image acquisition unit 311. The mark detection unit 315 is a unit that detects a mark from the reflection image obtained by the reflection image acquisition unit 311.

  The feature amount extraction coordinates specified by the mark may be misaligned during the mark printing process or when the reflected image is acquired. Therefore, the coordinates for feature extraction are shifted little by little by the perturbation processing unit 316, and the feature quantity is extracted by the feature extraction processing unit 317 from the shifted region. In order to shift, at least one method is stored in the paper piece identifying means 310 in advance. Then, the similarity calculation means 318 calculates the degree of similarity (similarity) between the obtained feature quantity and the standard feature quantity obtained by the feature quantity encoding means 313.

  As a method of calculating the similarity, a correlation function or a cosine similarity can be used. It can also be the reciprocal of the Euclidean distance or Mahalanobis distance. In the authenticity determination process 319, if the similarity obtained by any of the perturbation processes is equal to or greater than the threshold value θ, the paper piece 309 is regarded as genuine and the result is output. Conversely, if it is less than the threshold value θ, the paper piece 309 is regarded as a fake and the result is output. In the authenticity determination process 319, the threshold value θ is corrected not by simply comparing with the threshold value θ but by the characteristic characteristic of the paper piece obtained by the parameter decoding unit 312 or the characteristic characteristic of the paper piece registration / identification device. A specific correction method will be described later in the description of the procedure of the paper piece authenticity determination process.

  FIG. 4 is an example of a transmission image of a piece of paper. This image is an image of high-quality paper captured at a resolution of 600 dpi using a commercially available transmissive scanner. The shading is emphasized for easy visual inspection.

  The paper piece registration means is shown in FIG. First, a transparent image of a piece of paper to be registered is taken (502), and then a feature amount used for identifying the piece of paper is extracted from the transparent image (503). Then, a parameter is calculated from the feature amount (504). Details of the feature extraction process and the parameter calculation process will be described later. Subsequently, an identifier is issued (505), and the feature amount, parameter, and mark are printed on the piece of paper (506). The marks are printed at the four corners of the area of the piece of paper that has undergone feature extraction. Alternatively, the coordinates of the feature extraction position may be encoded and printed on a piece of paper in the form of a barcode or a two-dimensional code. Finally, the identifier is printed on the piece of paper (507).

  FIG. 6 is a diagram showing the procedure of the paper piece feature extraction processing. In the feature extraction process, first, a transmission image is acquired (602). Next, based on the feature extraction area specified by the mark printed on the piece of paper, a partial image is cut out from the transparent image (602). Subsequently, normalization is performed (605). At this time, the image is subjected to projective transformation in accordance with a rectangle of a predetermined size and cut out. Next, a low-pass filter (LPF) is applied to the image (607), and the image in the transmission image is removed. Further, histogram normalization is performed on the obtained image (607). This process is a process of obtaining a histogram of the input image and converting the pixel value so that the histogram becomes flat. This process makes it possible to correct a difference in characteristics between the transmission image captured at the time of paper piece registration and the transmission image captured at the time of paper piece identification. Finally, numerical features are extracted from the image obtained as a result of histogram normalization (607) (608). For example, the value of each pixel is used as the numerical feature quantity. In this case, if the output of image segmentation / normalization is a rectangular RGB color image with horizontal w pixels and vertical h pixels, a feature quantity having a dimension number of 3 × w × h can be obtained. In the case of a gray image, a feature amount having a dimension number of w × h is obtained. In addition, an image filter can be convolved with the image that has been cut out and normalized, and the output value can be used as a feature amount. If the feature extraction area is not rectangular, projective transformation is used as necessary.

  For the image cutout / normalization processing, the method described in Patent Document 5 can be used. That is, the feature extraction area is designated by the coordinates of the four corners in the transmission image. In this case, the feature extraction area is not necessarily a standard rectangle. It may also be tilted. Therefore, by using projective transformation, the parameters of the projective transformation can be used so that the coordinates of the four corners in the feature extraction region coincide with the coordinates of the four corners of the standard rectangle.

  Here, the perturbation process will be described. In this embodiment, the perturbation means that the range of the image used for feature extraction is changed minutely, and that the image range is changed vertically and horizontally. In this embodiment, the perturbation method p (i), where (1 ≦ i ≦ n) is registered in the n method, and features are extracted from the image to which the perturbation of p (i) is added. This makes it possible to correct misalignment caused by paper scanning. The method described in Patent Document 5 can be used for specific types and implementation of perturbations.

  The mark is a notation indicating on which position a pixel from which a feature amount used for authenticity determination is to be extracted is located. For the specific mark mounting, the method described in Patent Document 5 can be used.

  FIG. 7 is a diagram showing the procedure of the paper piece authenticity determination process. First, a reflected image of a piece of paper is obtained in a reflected image acquisition process 702. Next, in a transparent image acquisition process 703, a transparent image of a piece of paper is acquired. Subsequently, an identifier is recognized from the reflected image or the transmitted image (704). Subsequently, the feature amount is decoded from the obtained reflection image or transmission image (706). Next, the mark is recognized from the reflected image or the transmitted image (707). Next, the following processing is repeated in the loop 708 for the number of perturbation methods. First, the coordinate x of the region where feature extraction is performed is slightly shifted by the i-th perturbation method. Then, the feature amount f of the shifted feature region coordinates is extracted (710). Subsequently, the paper piece image of the region coordinate y adjacent to the feature region x is corrected by the i-th perturbation method (711), and the feature amount g is extracted from the paper piece image of the feature extraction region coordinate y (712). Then, the similarity θ between the standard feature amount decoded by the feature amount decoding process 705 and the feature amount f is calculated (713), and the similarity Δ between the feature amount f and the feature amount g is calculated (714). Then, it is determined whether the similarity θ is greater than or equal to the threshold th1 (715). If it is greater than or equal to the threshold, it is further determined whether the similarity Δ is greater than or equal to the threshold th2 (716). The ticket is output (717), and the process is terminated (718). Regardless of the perturbation method, if the degree of similarity θ, Δ is less than the threshold value, a counterfeit note is output as the determination result (719), and the process is terminated (720).

  Here, the parameters will be described. The parameter is a set value given from the outside in order to control the operation of the paper piece authenticity determination process. Specifically, the similarity and the thresholds th1 and th2 in FIG. The threshold th1 can be obtained as follows. First, two positions on a piece of paper to be registered are selected, and feature amounts v1 and v2 are calculated by the feature amount calculation means 503 in FIG. Next, the similarity of those feature quantities is calculated and set to th1. Alternatively, an appropriate constant α may be prepared, and α × th1 may be changed to th1. In the above, the method of extracting features by selecting two positions and calculating the similarity between those feature quantities was described. However, the feature extraction by selecting three or more positions and the similarity of those feature quantities was performed. The threshold may be calculated by calculating the degree. In that case, a plurality of threshold values th1-1, th1-2,... Are obtained. By calculating the threshold value in this way, it is possible to estimate a characteristic unique to the piece of paper, and to define the operation of the paper piece authenticity determination process according to the characteristic.

  The threshold th2 can be obtained as follows. First, one position on the piece of paper to be registered is selected, and the feature amount w1 is calculated by the feature amount calculation means 503 in FIG. Next, without moving the piece of paper, the transmission image capturing 502 in FIG. 5 is performed, and the feature amount calculation unit 503 calculates the feature amount w2. Then, the similarity between the feature quantity w1 and the feature quantity w2 is calculated, and this is set as th2. Alternatively, an appropriate constant beta may be prepared, and β × th2 may be changed to th2. In the above, the threshold value was calculated from the captured images of two times, but feature extraction was performed from the captured images of three times or more, and the similarity between these feature quantities, that is, the similarity between w1 and w2, and the similarity between w1 and w3 ,... May be calculated as threshold values. In that case, a plurality of threshold values th2-1, th2-2,. By calculating the threshold value in this way, it is possible to estimate the characteristic unique to the paper piece registration / identification device, and to define the operation of the paper piece authenticity determination process according to the characteristic.

  Since the threshold value for authenticity determination is set by using the paper or the device that actually performs the authenticity determination, the authenticity determination can be performed with high accuracy while suppressing the influence due to the variation of the paper composition or the individual device.

  As a second embodiment of the present invention, a paper piece authenticity determination system will be described below.

  FIG. 8 shows a configuration example of the paper piece registration / identification device. In this apparatus, at least one computer, an image acquisition unit, a printing unit, a time calculation unit, and a communication unit are coupled via a bus. That is, the paper piece registration / identification device includes a transmission image acquisition unit 801 that acquires a transmission image of a paper piece, a reflection image acquisition unit 802 that acquires a reflection image of a paper piece, a calculation unit 803, a communication unit 804, and a storage unit 807. , A printing unit 820, and a bus 806, and can communicate with an external device via a network 805. The transparent image is used when extracting a feature amount for identifying a piece of paper. The reflected image is used for recognizing a feature amount printed on a piece of paper as a code, a piece of paper identifier, or a mark. Here, the identifier is a character string unique to each piece of paper. The mark is a mark to be printed on a piece of paper in order to designate a region from which a feature value used for identifying the piece of paper is extracted.

  The transmission image acquisition unit 801 and the reflection image acquisition unit 802 can be realized as an image input terminal for the paper piece registration / identification device, and for example, a commercially available optical scanner can be used.

  The computing unit 803 can be realized as a general-purpose MPU (Micro Processor Unit), and the feature extraction unit 811, the feature quantity statistic calculation unit 819, and the similarity calculation unit 814 execute necessary calculations.

  Processing for recognizing the identifier and detecting the mark is stored in the storage unit 808 as a program. The storage unit 807 includes an identifier recognition unit 808, a mark detection unit 809, a perturbation processing unit 810, a feature extraction unit 811, an authenticity determination unit 812, an identifier decoding unit 813, a similarity calculation unit 814, an identifier issuance unit 815, and a feature quantity. A program that describes the processing procedures of the statistic decoding means 816, the feature quantity decoding means 817, the paper piece registration means 818, and the feature quantity statistic calculation means 819 is stored.

  Each element of the printing unit 820 can be realized by using a commercially available printer or printer module. The mark printing unit 821 is a unit that prints a mark on a piece of paper, the identifier printing unit 822 is a unit that prints an identifier on a piece of paper, and the feature amount printing unit 823 is a unit that prints a feature amount on a piece of paper. The quantity printing unit 824 is a unit that prints the statistical quantity of the feature quantity on a piece of paper. The time calculation means is means for obtaining the current time.

  FIG. 9 is a diagram showing a more specific configuration and flow of each processing program of the second embodiment. As shown in the figure, it is roughly divided into a paper piece registration process 902 and a paper piece identification process 903.

  The paper piece registration process 902 is a process of extracting a characteristic amount unique to a paper piece from the transmission image of the paper piece 901, quantifying the characteristic quantity, and printing it on the paper piece. The transmission image acquisition unit 903 is a unit for obtaining a transmission image of the paper piece 901. The feature extraction processing unit 904 is a unit that extracts a feature amount from the transmission image. The feature quantity statistic calculating means 905 is a process for calculating the feature quantity statistic obtained by the feature extraction processing means 904. For example, the variance of the histogram of the pixel value of the paper piece image or the variance of the histogram of the hue value of the paper piece image is used as the statistical amount of the feature amount. Details will be described later with reference to FIGS. A clock 906 is a means for obtaining the current time. In FIG. 9, the clock 906 is built in the paper piece registration means 902, but it may be obtained from an external server via a network. The feature quantity / statistical quantity / mark / time printing means 907 is a means for printing the feature quantity, the statistical quantity of the feature quantity, the mark, and the time on a piece of paper. As a form of printing, a numerical value may be printed as a character string, or may be printed with a code such as a barcode or a two-dimensional code. The identifier printing unit 909 is a unit that prints the identifier on a piece of paper. The identifier generation unit 908 may generate a character string unique to the paper piece inside the paper piece registration process 902, or may cause the identifier management server 205 to generate a character string unique to the paper piece via the network 105.

  The paper piece identification process 911 is a process for determining the authenticity of a paper piece by extracting a characteristic quantity unique to the paper piece from the transmission image of the paper piece, digitizing the characteristic quantity, and comparing it with the characteristic quantity printed on the paper piece. is there. The reflected image acquisition unit 912 is a unit that obtains a reflected image of the paper piece 910. The statistic / time decoding unit 913 is a unit that decodes the statistic amount and time of the feature amount included in the reflected image. When the numerical value is printed as a character string as the feature amount / statistical amount / mark / time printing unit 909, a known character recognition technique can be used. Even when printing is performed using a code such as a barcode or a two-dimensional code, a known barcode or two-dimensional code reading technique can be used. The feature amount decoding unit 914 is a unit that decodes the standard feature amount from the reflection image of the piece of paper obtained by the reflection image acquisition unit 912. The mark detection unit 916 is a unit that detects a mark from the reflection image obtained by the reflection image acquisition unit 912.

  The feature amount extraction coordinates specified by the mark may be misaligned during the mark printing process or when the reflected image is acquired. Therefore, the coordinates for feature extraction are shifted little by little by the perturbation processing unit 917, and the feature quantity is extracted by the feature extraction processing unit 918 from the shifted region. In order to shift, at least one method is previously stored in the paper piece identifying means 911. Then, the feature amount correcting unit 922 corrects the feature amount from the statistical amount of the feature amount obtained by the statistic / time decoding unit 913, the time when the piece of paper is registered, and the current time obtained by the clock 919. A specific correction method will be described later in the procedure of the paper piece authenticity determination process. Then, the similarity calculation means 920 calculates the degree of similarity (similarity) between the corrected feature quantity and the standard feature quantity obtained by the feature quantity encoding means 914. As a method of calculating the similarity, a correlation function or a cosine similarity can be used. It can also be the reciprocal of the Euclidean distance or Mahalanobis distance. In the authenticity determination process 921, if the similarity obtained by any of the perturbation processes is equal to or greater than the threshold θ, the piece of paper 910 is regarded as genuine and the result is output. Conversely, if it is less than the threshold value θ, the paper piece 910 is regarded as a fake and the result is output.

  FIG. 10 is a diagram for explaining the statistical amount of the feature amount. Specifically, it is a diagram showing a histogram of pixel values of a transparent image and a relationship between the histogram and elapsed time. A graph 1001 is a graph representing the distribution of pixel values of a paper piece image at a certain time t1. The horizontal axis represents the pixel value, and the vertical axis represents the frequency. A graph 1003 represents the distribution of pixel values of the same piece of paper image as the graph 1001 at a certain time t2 (> t1). The horizontal axis represents the pixel value, and the vertical axis represents the frequency. The composition structure inside the paper is less likely to change with time compared to the composition structure on the paper surface, but the internal composition structure changes slightly due to external friction, bending, dirt, etc. Thus, the histogram shape of the transmission image is Change. For example, as is clear from the comparison between the distribution 1002 and the distribution 1004, the dispersion of pixel values decreases with time. If these variances are graphed over time, they can be expressed as a graph 1005. Immediately after papermaking, the dispersion is large and the dispersion decreases with time. The feature quantity statistic calculation means 905 obtains this variance. Specifically, first, a histogram of pixel values such as a graph 1001 is calculated from the transmission image, and the variance of the distribution 1002 is obtained. A known statistical calculation algorithm can be used for the calculation of the variance. With respect to the average value of the pixels, a change with time can be graphed like a graph 1005 from the feature amount of a large amount of paper pieces.

  FIG. 11 is a diagram for explaining the statistical amount of the feature amount. Specifically, it is a diagram showing a histogram of hue values of a transmission image and a relationship between the histogram and elapsed time. A graph 1101 is a diagram showing the relationship between the distribution of hue values of the paper piece image and the elapsed time at a certain time t1. The horizontal axis represents the pixel value, and the vertical axis represents the frequency. A graph 1103 is a graph showing the relationship between the distribution of hue values and the elapsed time of the same piece of paper image as the graph 1101 at a certain time t2 (> t1). The horizontal axis represents the pixel value, and the vertical axis represents the frequency. As time goes on, the histogram shape of the hue image changes in this way due to a change in ink composition printed on one side of the paper, stains, and the like. For example, as is apparent from a comparison between the distribution 1102 and the distribution 1104, the variance of the hue value decreases with time. If these variances are graphed over time, they can be expressed as a graph 1105. Immediately after papermaking, the dispersion is large and the dispersion decreases with time. The feature quantity statistic calculation means 905 obtains this variance. Specifically, first, a color RGB transmission image is acquired to obtain a hue image. In order to obtain a hue image from a color RGB image, a known algorithm in the image processing field can be used. Next, a histogram of pixel values such as a graph 1101 is calculated from the hue image, and the variance of the distribution 1102 is obtained. A known statistical calculation algorithm can be used for the calculation of the variance. With respect to the average value of hue, a change with time can be graphed as shown in a graph 1005 from the feature amount of a large amount of paper pieces.

  Here, the statistical value of the feature value is shown by the variance value of the histogram of the pixel value of the transparent image and the variance of the histogram of the pixel value of the hue image obtained from the transparent image, but is not limited to these two examples. Any statistic obtained from a paper piece transmission image that changes over time can be used. Further, it is possible to estimate an image change accompanying a change with time more accurately by using both the dispersion value of the histogram of the transmission image and the dispersion value of the histogram of the hue image.

  FIG. 12 is a diagram showing a procedure for calculating the statistical amount of the feature amount of the piece of paper in the second embodiment. In the feature extraction process, first, a transmission image is acquired (1202). Next, based on the specified feature extraction region (1203), a partial image is cut out from the transparent image (1204). Subsequently, normalization is performed (1205). At this time, the image is subjected to projective transformation in accordance with a rectangle of a predetermined size and cut out. Next, a low pass filter (LPF) is applied to the image (1206), and the image in the transmission image is removed. Further, histogram normalization is performed on the obtained image (1207). This process is a process of obtaining a histogram of the input image and converting the pixel value so that the histogram becomes flat. By this processing, it becomes possible to correct the difference in characteristics between the transmission image captured at the time of paper piece registration and the transmission image captured at the time of paper piece identification. Finally, the variance of the feature amount is calculated from the image obtained as a result of the histogram normalization (1207) (1208). The method described with reference to FIG. 10 can be used as a calculation method of the variance of the feature amount. Finally, the obtained variance is output as a feature quantity statistic (1209).

  FIG. 13 is a diagram showing another procedure for calculating the statistical amount of the feature amount of the piece of paper in the second embodiment. In the feature extraction process, first, a transparent image is acquired (1302). Next, based on the specified feature extraction region (1303), a partial image is cut out from the transparent image (1304). Subsequently, normalization is performed (1305). At this time, the image is subjected to projective transformation in accordance with a rectangle of a predetermined size and cut out. Next, a low-pass filter (LPF) is applied to the image (1306), and the image in the transmission image is removed. Further, histogram normalization is performed on the obtained image (1307). This process is a process of obtaining a histogram of the input image and converting the pixel value so that the histogram becomes flat. By this processing, it becomes possible to correct the difference in characteristics between the transmission image captured at the time of paper piece registration and the transmission image captured at the time of paper piece identification. Further, HSV conversion is performed from the RGB values (1308), and a histogram of hue values (H values) is calculated (1309). Then, the variance of the feature amount is calculated from the histogram of hue values (1310). The method described with reference to FIG. 10 can be used as a calculation method of the variance of the feature amount. Finally, the obtained variance is output as a statistical quantity of the feature quantity (1311).

  FIG. 14 is a diagram showing the procedure of a paper piece authenticity determination process. First, in the reflection image acquisition processing 1402, a reflection image of a piece of paper is acquired using the transmission image acquisition means 802. Next, in the transmission image acquisition process 1403, a reflection image acquisition unit 802 is used to acquire a transmission image of a piece of paper. Subsequently, the identifier is recognized from the reflected image or the transmitted image using the identifier recognizing means 808 (1404). Subsequently, the feature amount decoding means 817 is used to decode the feature amount from the obtained reflection image or transmission image (1405). Next, the feature quantity statistic decoding means 816 is used to decode the feature quantity statistic (1406). Next, the mark detection unit 809 is used to recognize the mark from the reflected image or the transmitted image (1407). Next, the following processing is repeated in the loop 1408 for the number of perturbation methods. First, by using the perturbation processing means 810, the coordinates x of the region where feature extraction is performed are slightly shifted by the i-th perturbation method (1409). Then, the feature amount f of the shifted feature area coordinates is extracted using the feature extraction unit 811 (1410), and the current time is calculated using the time calculation unit 825 (1411). Based on the calculated current time The feature quantity f is corrected using the feature quantity correction means 825 to generate the feature quantity g (1412). Then, using the similarity calculation means 814, the degree of similarity θ between the decoded feature quantity and the feature quantity g is calculated (1413). Then, the authenticity determination means 812 is used to determine whether the degree of similarity θ is greater than or equal to the threshold th1 (1414), a genuine note is output as a determination result (1415), and the process is terminated (1416). In any perturbation method, if the similarity θ is less than the threshold value, a counterfeit note is output as a determination result (1417), and the process is terminated (1418).

  A feature quantity statistic is used for correction of the feature quantity in the processing 1412. In the following, as the statistic, the variance value σ ^ 2 and the average value μ of the feature quantity and information on how these values change with time are used. These can be prepared in advance by calculating feature values from a large number of pieces of paper and calculating feature values of the same piece of paper over time. When a pixel value is used as the feature amount, the feature amount can be corrected using the feature amount statistics as follows. Assume that the value of each pixel before correction is x, the value of each pixel after correction is y, the distribution of pixel value distribution before correction is σ1 ^ 2, and the average distribution of pixel values before correction is μ1. First, the normalized feature quantity z is calculated as z = (x−μ1) / σ1 (Expression 1). Next, the current time calculation 1411 calculates the elapsed time from the time of registering the piece of paper, and from the graph 1005 showing how the average and variance of the feature amount change with the passage of time, the current time is calculated. The variance value σ2 and the average value μ2 are estimated. Then, by performing inverse transformation of (Equation 1), that is, z = (x′−μ 2) / σ 2, the pixel value x ′ at the current time can be estimated. That is, according to this procedure, the pixel value x at the time of registering the piece of paper can be corrected to the pixel value x ′ at the current time. Further, instead of pixel values, color histogram information can be used as shown in FIG.

  Alternatively, in addition to the threshold value processing of the similarity θ in the processing 1414, the distribution of the pixels that are the objects of the authenticity determination is obtained, and if the distribution is equal to or smaller than the variance at the current position that can be read from the graph 1005, it is determined to be genuine. Authentication can also be determined.

  Hereinafter, the flow of the paper piece registration / identification process according to the third embodiment of the present invention will be described. Only differences from the first embodiment will be described in detail. In the third embodiment, a method is disclosed in which a piece of paper can be identified with high accuracy even when characters are printed or written on the piece of paper.

  FIG. 15 is a diagram showing the procedure of the paper piece feature extraction process. In the feature extraction process, first, a transparent image is acquired (1502). Next, based on the designated feature extraction region (1503), a partial image is cut out from the transparent image (1504). Subsequently, normalization is performed (1505). At this time, the image is subjected to projective transformation in accordance with a rectangle of a predetermined size and cut out. Next, a low-pass filter (LPF) is applied to the image (1506), and the image in the transmission image is removed. Further, histogram normalization is performed on the obtained image (1507). This process is a process of obtaining a histogram of the input image and converting the pixel value so that the histogram becomes flat. By this processing, it becomes possible to correct the difference in characteristics between the transmission image captured at the time of paper piece registration and the transmission image captured at the time of paper piece identification.

  Subsequently, the character area is removed (1508). An image binarization processing algorithm can be applied to the removal of the character area. Using the fact that a character area generally has a smaller pixel value than a plain area where no characters are printed, for each image of the input image, a pixel having a pixel value larger than a predetermined threshold is assigned to the character area. A pixel having a pixel value equal to or greater than the region and the threshold can be used for feature extraction as a background region. Also, as described in Noriyuki Otsu, “Automatic threshold selection method based on discriminant and least-square criterion”, IEICE Transactions Journal D, Vol. 63, No. 4, pp. 349-356, 1980 Alternatively, a technique for setting a threshold may be applied.

  Finally, numerical features are extracted (1509). For example, the value of each pixel is used as the numerical feature quantity. In this case, if the output of image segmentation / normalization is a rectangular RGB color image with horizontal w pixels and vertical h pixels, a feature quantity having a dimension number of 3 × w × h can be obtained. In the case of a gray image, a feature amount having a dimension number of w × h is obtained. In addition, an image filter can be convolved with the image that has been cut out and normalized, and the output value can be used as a feature amount. If the feature extraction area is not rectangular, projective transformation is used as necessary.

  According to the above-described procedure, it is possible to automatically exclude a character area from a transparent image and perform feature extraction, and to determine the authenticity of a piece of paper with high accuracy.

  In addition, when information such as what character is written at which position on the piece of paper and what kind of photograph is printed can be obtained in advance, the information is obtained and the area of the character or photograph is removed. Can also be realized.

  Hereinafter, the flow of the paper piece registration / identification process according to the fourth embodiment of the present invention will be described. Only differences from the first embodiment will be described in detail. In the fourth embodiment, there are fluctuations in feature amounts due to variations in the composition of paper pieces, fluctuations in feature amounts due to differences in papermaking conditions, fluctuations in feature amounts due to individual differences among scanners installed in the paper piece registration / identification device, etc. A method for identifying a piece of paper with high accuracy even when it exists is described.

  FIG. 16 shows the procedure of feature extraction processing in the fourth embodiment. This is an extended procedure of the feature extraction process described in FIG. Specifically, the “feature amount extraction” described in FIG. 16 corresponds to the processing from 602 to 608, and the processing from 1602 to 1608 is added. The procedure other than the feature extraction process is the same as in the first embodiment. The procedure will be described in detail below. First, the feature quantity of the feature extraction area coordinate x is extracted (1602). Let the extracted feature quantity be the feature quantity f. Next, the feature amount of the region y adjacent to the feature extraction region coordinate x is extracted (1603). Let the extracted feature quantity be the feature quantity g. Next, the following loop is executed while the similarity between the feature quantity f and the feature quantity g is greater than or equal to the threshold value θ, the feature extraction area coordinate x is in the paper piece, and the coordinate y is in the paper piece. First, the feature extraction region x is expanded by one pixel vertically and horizontally (1605). Next, the feature extraction area y is expanded by one pixel vertically and horizontally (1606). Furthermore, the feature quantity of the extended feature extraction area coordinate x is extracted, and this feature quantity is re-defined as the feature quantity f. Further, the feature amount of the extended feature amount extraction region coordinate y is extracted, and this feature amount is again referred to as a feature amount g.

  By performing such processing, it is possible to set the feature extraction range so that the similarity between feature amounts for different feature extraction regions is less than a threshold value. Since different feature extraction regions can be regarded as separate pieces of paper, it is possible to improve the accuracy with which a false ticket is erroneously determined to be a genuine note.

  In the above, the processing procedure was described for the method of fixing the threshold θ and gradually extending the feature extraction range, but conversely fixing the feature extraction area and gradually reducing the threshold θ to falsely insert a counterfeit note. Therefore, it is possible to improve the accuracy of determining a genuine note.

  INDUSTRIAL APPLICABILITY The present invention is useful as a system that can perform authenticity determination at high speed and high accuracy in a system for determining authenticity of a piece of paper.

DESCRIPTION OF SYMBOLS 101 ... Transmission image acquisition means, 102 ... Reflection image acquisition means, 103 ... Calculation means, 104 ... Communication means, 105 ... Network, 106 ... Bus, 107 ... Storage means, 108 ... Identifier recognition means, 109 ... Mark detection means, 110 ... perturbation processing means, 111 ... feature extraction processing means, 112 ... parameter estimation means, 113 ... identifier decoding means, 114 ... similarity calculation means, 115 ... identifier issuing means, 116 ... feature amount decoding means, 117 ... parameter decoding 118 ... paper piece registration means, 119 ... authentication determination means, 120 ... printing means, 121 ... mark printing means, 122 ... identifier printing means, 123 ... feature amount printing means, 124 ... parameter printing means, 201 ... paper piece registration / Identification device, 202... Paper piece registration / identification device, 203... Paper piece registration / identification device, 204. DESCRIPTION OF SYMBOLS 5 ... Paper piece identifier management server, 206 ... Paper piece identifier management DB, 301 ... Paper piece, 302 ... Paper piece registration means, 303 ... Transparent image acquisition means, 304 ... Feature extraction processing means, 305 ... Parameter estimation means, 306 ... Feature quantity / parameter Mark printing means, 307 ... identifier issuing means, 308 ... identifier printing means, 309 ... paper piece, 310 ... paper piece identification means, 311 ... reflection image acquisition means, 312 ... parameter decoding means, 313 ... feature amount decoding means, 314 ... transparent image acquisition means, 315 ... mark detection means, 316 ... perturbation processing means, 317 ... feature extraction processing means, 318 ... similarity calculation means, 319 ... authentication determination means, 401 ... transparent image of paper, 501 ... start, 502 ... Transparent image imaging processing, 503... Feature amount calculation processing, 504... Parameter calculation processing, 505. ... feature quantity / parameter / mark printing process, 507 ... identifier printing process, 508 ... end, 601 ... start, 602 ... transparent image acquisition process, 603 ... coordinate acquisition process, 604 ... image cutout process, 605 ... normalization process, 606 ... LPF processing, 607 ... histogram normalization processing, 608 ... feature value digitization processing, 609 ... end, 701 ... start, 702 ... reflected image acquisition processing, 703 ... transmission image acquisition processing, 704 ... identifier recognition processing, 705 ... features. Quantity decoding process, 706 ... parameter decoding process, 707 ... mark recognition process, 708 ... loop process, 709 ... perturbation process, 710 ... feature extraction process, 711 ... perturbation process, 712 ... feature extraction process, 713 ... similarity calculation Processing, 714 ... similarity calculation processing, 715 ... threshold processing, 716 ... threshold processing, 717 ... determination result output processing, 718 ... end 19 ... judgment result output processing, 720 ... end, 801 ... transmission image acquisition means, 802 ... reflection image acquisition means, 803 ... calculation means, 804 ... communication means, 805 ... network, 806 ... bus, 807 ... storage means, 808 ... Identifier recognition means, 809 ... mark detection means, 810 ... perturbation processing means, 811 ... feature extraction means, 812 ... authenticity determination means, 813 ... identifier decoding means, 814 ... similarity calculation means, 815 ... identifier issue means, 816 ... Feature quantity statistic decoding means, 817 ... Feature quantity decoding means, 818 ... Paper piece registration means, 819 ... Feature quantity statistic calculation means, 820 ... Printing means, 821 ... Mark printing means, 822 ... Identifier printing means, 823 ... Feature quantity printing means, 824 ... feature quantity statistics printing means, 825 ... time calculation means, 826 ... feature quantity correction means, 901 ... paper piece, 902 Paper piece registration means, 903... Transparent image acquisition means, 904... Feature extraction processing means, 905... Feature quantity calculation means, 906... Clock, 907 ... feature quantity / statistics / mark / time printing means, 908. Means 909 ... identifier printing means, 910 ... paper piece, 911 ... paper piece identification means, 912 ... reflection image acquisition means, 913 ... statistic / time decoding means for feature quantity, 914 ... feature quantity decoding means, 915 ... transparent image Acquisition means, 916... Mark detection means, 917 ... perturbation processing means, 918 ... feature extraction processing means, 919 ... clock, 920 ... similarity calculation means, 921 ... authenticity determination means, 922 ... feature amount correction means, 1001 ... transparent image 1002... Histogram of pixel values, 1002... Histogram, 1003. 1101 ... Histogram of hue value of transmission image, 1102 ... Histogram, 1103 ... Histogram of hue value of transmission image, 1201 ... Start, 1202 ... Transmission image acquisition processing, 1203 ... Coordinate acquisition Processing, 1204 ... Image clipping processing, 1205 ... Normalization processing, 1206 ... LPF processing, 1207 ... Histogram normalization processing, 1208 ... Variance calculation processing, 1209 ... Feature amount output processing, 1210 ... End, 1301 ... Start, 1302 ... Transmission Image acquisition processing, 1303 ... Coordinate acquisition processing, 1304 ... Image cut-out processing, 1305 ... Normalization processing, 1306 ... LPF processing, 1307 ... Histogram normalization processing, 1308 ... HSV conversion processing, 1309 ... H-value histogram calculation processing, 1310 ... dispersion calculation processing, 1311 ... feature amount output processing, 131 2 ... End, 1401 ... Start, 1402 ... Reflected image acquisition process, 1403 ... Transparent image acquisition process, 1404 ... Identifier recognition process, 1405 ... Feature quantity code decoding process, 1406 ... Feature quantity statistic decoding process, 1407 ... Mark recognition processing, 1408 ... perturbation loop processing, 1409 ... perturbation processing, 1410 ... feature extraction processing, 1411 ... time calculation processing, 1412 ... feature value correction processing, 1413 ... similarity calculation processing, 1414 ... threshold processing, 1415 ... determination result Output processing, 1416 ... End, 1417 ... Determination result output processing, 1418 ... End, 1501 ... Start, 1502 ... Transparent image acquisition processing, 1503 ... Coordinate acquisition processing, 1504 ... Image clipping processing, 1505 ... Normalization processing, 1506 ... LPF Processing, 1507 ... Histogram normalization processing, 1508 ... Character area removal processing, 1509 ... Features Digitization processing, 1510 ... End, 1601 ... Start, 1602 ... Feature extraction processing, 1603 ... Feature extraction processing, 1604 ... Determination processing, 1605 ... Feature extraction region expansion processing, 1606 ... Feature extraction processing expansion processing, 1607 ... Feature extraction processing , 1608 ... feature extraction processing, 1609 ... end

Claims (15)

In a paper piece identification system comprising a paper piece registration device and a paper piece identification device,
The paper registration device, a printing unit that prints a first feature amount Extraction process detection section for extracting a feature value from a piece of paper, information about the feature amount extracted by the feature extraction processing unit of the first to the paper And
The paper identification unit comprises a feature amount acquisition unit that acquires the second feature amount Extraction process detection section for extracting a feature value from a piece of paper, information about the feature amount printed on the paper piece by the printing unit from the paper, A feature amount decoding unit that extracts a feature amount from information related to the feature amount, and a similarity between the feature amount extracted by the feature amount decoding unit and the feature amount extracted by the second feature amount extraction processing unit a paper identification system characterized by having a determining similarity calculation unit,
The paper piece identification device has an authenticity determination unit that determines the authenticity of a paper piece based on the similarity determined by the similarity calculation unit,
A parameter estimation unit that estimates one or more of the unique characteristics of the paper piece registration device, the unique characteristics of the paper piece identification device, and the unique characteristics of the paper piece;
The authenticity determination unit determines the authenticity of a piece of paper based on the similarity and the characteristic estimated by the parameter estimation unit. In the paper piece identification system according to claim 1,
The paper piece registration device and the paper piece identification device include a transparent image acquisition unit that acquires a transparent image of a paper piece, and extracts the feature amount of the paper piece from the transparent image acquired by the transparent image acquisition unit. Paper piece identification system. In the paper piece identification system according to claim 1,
The printing unit prints a mark indicating a position of a piece of paper from which the first feature amount is extracted;
The paper piece identification device includes a mark detection unit that detects the mark printed on a paper piece, and the second feature amount extraction processing unit is configured to use the feature amount based on the mark detected by the mark detection unit. A paper piece identification system characterized by determining a region from which to extract. In the paper piece identification system according to claim 3,
The paper piece identification device includes a perturbation processing unit that perturbs a region from which the second feature amount extraction processing unit extracts a feature amount based on the detected mark,
The paper piece identification system, wherein the second feature amount extraction processing unit extracts feature amounts in a plurality of regions based on perturbation by the perturbation processing unit. In the paper piece identification system according to claim 1,
The paper piece registration apparatus includes a statistic calculation unit that calculates a statistic of the feature amount from the feature amount extracted by the first feature amount extraction processing unit, and the printing unit is calculated by the statistic calculation unit. Print information about the statistics taken on a piece of paper,
The paper piece identification device is extracted by a statistic decoding unit that extracts a statistic from information relating to a statistic printed on the paper piece, and the second feature quantity extraction processing unit based on the decoded statistic A feature amount correction unit that corrects the feature amount, and the similarity calculation unit is configured to obtain a similarity between the feature amount extracted by the feature amount decoding unit and the feature amount corrected by the feature amount correction unit. A paper piece identification system characterized by determining the degree. In the paper piece identification system according to claim 1,
The paper piece registration device and the paper piece identification device have a transmission image acquisition unit that acquires a transmission image of a paper piece, and the first and second feature amount extraction processing units are the transmission images acquired by the transmission image acquisition unit. A paper piece identification system, wherein a character area is removed from an image, and a feature amount of the paper piece is extracted from a transparent image from which the character area is removed. A feature quantity extraction processing unit for extracting feature quantities from a piece of paper;
A feature amount acquisition unit that acquires information about the feature amount printed on the piece of paper;
A feature amount decoding unit that extracts a feature amount from information on the feature amount;
A paper piece identification device comprising: a similarity calculation unit that determines a similarity between the feature amount extracted by the feature amount decoding unit and the feature amount extracted by the feature amount extraction unit ;
An authenticity determination unit that determines the authenticity of a piece of paper based on the similarity determined by the similarity calculation unit;
A parameter estimation unit that estimates one or more of the unique characteristics of the paper piece identification device and the unique characteristics of the paper piece;
The authenticity determination unit determines the authenticity of a piece of paper based on the similarity and the characteristic estimated by the parameter estimation unit. In the paper piece identification device according to claim 7,
An apparatus for identifying a piece of paper, comprising: a printing unit that prints information on a feature amount extracted from a piece of paper by the feature amount extraction processing unit on the piece of paper. In the paper piece identification device according to claim 7,
A paper piece identification apparatus, comprising: a transmission image acquisition unit that acquires a transmission image of a piece of paper, wherein the feature amount extraction processing unit extracts the feature amount of the piece of paper from the transmission image acquired by the transmission image acquisition unit . In the paper piece identification device according to claim 7,
A mark detection unit for detecting a mark printed on a piece of paper, wherein the feature amount extraction processing unit determines a region for extracting the feature amount based on the mark detected by the mark detection unit; Paper piece identification device. In the paper piece identification device according to claim 10,
The feature amount extraction processing unit has a perturbation processing unit for perturbing a region from which a feature amount is extracted based on the detected mark;
The paper piece identifying apparatus, wherein the feature amount extraction processing unit extracts feature amounts in a plurality of regions based on perturbation by the perturbation processing unit. In the paper piece identification device according to claim 7,
A statistic decoding unit that extracts a statistic from information relating to the statistic of the feature quantity printed on the piece of paper, and a feature quantity extracted by the feature quantity extraction unit based on the decoded statistic A feature amount correction unit, wherein the similarity calculation unit determines a similarity between the feature amount extracted by the feature amount decoding unit and the feature amount corrected by the feature amount correction unit. Paper piece identification device . In the paper piece identification device according to claim 7,
A transparent image acquisition unit that acquires a transparent image of a piece of paper, wherein the feature amount extraction processing unit removes the character area from the transparent image acquired by the transparent image acquisition unit, and the transparent image from which the character area has been removed A paper piece identification device that extracts a feature amount of a paper piece. In the paper piece identification device according to claim 8,
The printing unit prints a mark indicating the position of the piece of paper from which the first feature amount is extracted,
A mark detection unit for detecting the mark printed on a piece of paper;
The paper piece identifying apparatus, wherein the feature amount extraction processing unit determines a region from which the feature amount is extracted based on the mark detected by the mark detection unit. In the paper piece identification device according to claim 8,
A statistic calculator that calculates a statistic of the feature amount from the feature amount extracted by the feature amount extraction processing unit;
The printing unit prints information on the statistic calculated by the statistic calculating unit on a piece of paper,
A statistic decoding unit that extracts a statistic from information on the statistic printed on the piece of paper;
A feature amount correction unit that corrects the feature amount extracted by the feature amount extraction unit based on the decoded statistics,
The paper piece identifying apparatus, wherein the similarity calculation unit determines a similarity between the feature amount extracted by the feature amount decoding unit and the feature amount corrected by the feature amount correction unit.