JP2014174727A - Paper money discrimination method, software for paper money discrimination, and paper money discrimination device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a paper money discrimination device, software for paper money discrimination and a paper money discrimination method allowing the visually impaired in particular to easily discriminate a denomination of paper money.SOLUTION: The paper money discrimination method includes the steps of: acquiring moving image data of paper money subject to denomination discrimination; extracting feature amounts at a plurality of local feature positions using the moving image data to create a plurality of local discrimination feature amount data; discriminating similarity by comparing and collating the plurality of local discrimination feature amount data with a plurality of local reference feature amount data obtained by extracting feature amounts at a plurality of local feature positions for each preset denomination and creating notification data representing a discrimination result; and outputting notification data and notifying of the discrimination result, thereby discriminating the denomination of the paper money.

Description

  The present invention relates to a bill discriminating method, bill discriminating software, and a bill discriminating device, and more particularly, to a bill discriminating method, bill discriminating software, and bill discriminating device that allow a visually impaired person to easily discriminate bill types.

  In recent years, it has been performed to discriminate a ticket type using a discriminator for valuable printed matter such as banknotes, cash vouchers, gift certificates, admission tickets and the like.

  As a banknote identification device used for vending machines, ATMs, money changers, etc., for example, Patent Document 1 discloses a magnetic sensor and a magnetic sensor in a specific section of a banknote set in advance from the output of the magnetic sensor. Disclosed is a bill discriminating apparatus comprising means for detecting the maximum width of a non-existing portion, and determination means for comparing the width detected by the detecting means with a preset allowable range to determine the authenticity of the bill. ing.

  In addition, portable ticket type discriminating devices that can be carried are proposed so that visually impaired persons can discriminate bill types regardless of location.

  For example, Patent Document 2 includes a detection unit that detects a magnetic pattern and an optical pattern of a bill as a portable discrimination device, and a transport unit that manually feeds the bill through a through hole. Whether the magnetic pattern and the optical pattern of the banknote detected by the transport means and the detection means are within the range between the upper limit value and the lower limit value of the reference values stored in advance. A portable banknote discriminator for visually handicapped persons that identifies which banknote corresponds to or does not correspond by performing a comparison operation is disclosed.

  Further, Patent Document 3 includes an information reading unit that reads information for authenticity determination from a bill, a comparison determination unit that compares the information read by the information reading unit with reference information, and determines the authenticity of the bill. A portable terminal device including means for outputting an authentication result is disclosed.

Japanese Patent No. 28006267 Japanese Patent Application Laid-Open No. 07-175955 JP 2007-293405 A

  However, the bill recognition device used in the vending machine, ATM, money changer and the like disclosed in Patent Document 1 includes a large-scale transport device and reading device and cannot be carried around. For this reason, there existed a problem that a visually handicapped person could not perform the banknote type discrimination easily regardless of a place.

  Since the portable banknote discriminator for visually impaired persons disclosed in Patent Document 2 is portable, it can discriminate the banknote type regardless of location. However, when performing ticket type discrimination, it is necessary to align the bills accurately at a predetermined location of the transport means. For this reason, there was a problem that a visually impaired person cannot distinguish ticket types easily.

  In addition, in order to determine the bill type of a bill, it is necessary to photograph the bill with a camera and acquire image data. Conventionally, when acquiring image data, a disturbing background image appears around the banknote, the shooting distance, variation in shooting angle, wrinkles on part of the banknote, dirt, shadows, uneven illumination, image deformation, etc. There is a problem that the ticket type cannot be accurately determined. In the portable terminal device disclosed in Patent Document 3, it is necessary to accurately align the camera and the bill in order to acquire the image data of the bill. When a visually handicapped person performs ticket type determination, it is difficult to perform such accurate positioning, and there is a problem that the ticket type cannot be easily determined.

  Moreover, in any of Patent Documents 1 to 3, there is no description about discriminating a banknote folded in two or four.

  In view of the above circumstances, the present invention does not require accurate alignment between the camera and banknotes, and wrinkles, dirt, shadows, uneven illumination, image deformation, variations in shooting distances and shooting angles, etc., on some banknotes. To provide a banknote discriminating method, banknote discriminating software, and a banknote discriminating device that allow a visually impaired person to easily and accurately discriminate the type of banknote even when it occurs or when a banknote is folded. With the goal.

The bill discriminating method of the present invention is
Acquiring movie data of bills subject to ticket type discrimination;
Using the moving image data, extracting features at a plurality of local feature positions to create a plurality of locally discriminating feature data;
The plurality of local discriminating feature amount data is similar to a plurality of local reference feature amount data serving as a reference from which features at a plurality of local feature positions for each preset ticket type are extracted and compared. Determining and creating notification data indicating the determination result;
Outputting the notification data to notify the determination result;
It is characterized by providing.

Banknote discriminating software for causing a bill discriminating apparatus including an image capturing unit, a calculation unit, a storage unit, and an output unit to execute the method for discriminating the bill type of the present invention is as follows.
Step of acquiring movie data of bills subject to ticket type discrimination by the image capturing unit;
A step of extracting a feature at a plurality of local feature positions by using the moving image data and creating a plurality of locally discriminating feature amount data by the arithmetic unit;
A plurality of local reference features serving as a reference from which the plurality of local discriminating feature amount data and features at a plurality of local feature positions for each ticket type stored in advance in the storage unit are extracted by the arithmetic unit. Comparing and comparing quantity data to determine similarity, and creating notification data indicating the determination result;
Outputting the notification data to notify the determination result by the output unit;
The bill discriminating apparatus is caused to execute a bill discriminating method.

The bill discriminating apparatus of the present invention is
An image capture unit that acquires movie data of bills subject to ticket type discrimination;
A storage unit that stores in advance a plurality of local reference feature amount data serving as a reference from which features at a plurality of local feature positions for each ticket type are extracted;
Using the moving image data, features at a plurality of local feature positions are extracted to create a plurality of local discrimination feature amount data, and the plurality of local discrimination feature amount data and stored in the storage unit An arithmetic unit for comparing and collating the plurality of local reference feature data for each ticket type to determine similarity, and creating notification data indicating a determination result;
An output unit for outputting the notification data to notify the determination result;
It is characterized by providing.

  According to the bill discriminating method, bill discriminating software, and bill discriminating apparatus of the present invention, accurate alignment between the camera and the bill is unnecessary, and wrinkles, dirt, shadows, illumination unevenness, image unevenness of a part of the bill The visually impaired person can easily and accurately discriminate the bill type when the deformation, variation in photographing distance, photographing angle, or the like occurs, or when the bill is folded.

It is the schematic which showed the external appearance of the banknote discrimination | determination apparatus by embodiment of this invention. It is the block diagram which showed the structure of the banknote discrimination | determination apparatus by the embodiment. In the same embodiment, it is the flowchart which showed the procedure which acquires beforehand and stores several local reference | standard feature-value data for every ticket type. In the same embodiment, it is the flowchart which showed the procedure which discriminate | determines the banknote type of the banknote of discrimination | determination object using the some local reference | standard feature-value data for every banknote type. It is explanatory drawing which shows typically a part of banknote by several local discrimination image data. It is explanatory drawing which shows the method of obtaining a discrimination | determination feature image data by setting an attention pixel and calculating | requiring a feature position. It is explanatory drawing which shows the method of calculating | requiring the gravity center around a feature position. It is explanatory drawing which shows the method of carrying out rotation correction | amendment so that the gravity center of a feature position periphery may face a predetermined direction, and the method of extracting the feature in a feature position. It is explanatory drawing which shows the method of producing | generating the local discrimination | determination feature-value data which show the feature in the extracted feature position. It is the flowchart which showed the procedure which compares a local discrimination | determination feature-value data with the local reference | standard feature-value data for every ticket type, and discriminates a ticket type.

  Hereinafter, a bill discriminating device, bill discriminating software, and a bill discriminating method according to embodiments of the present invention will be described with reference to the drawings.

  In the present embodiment, the bill types of the banknotes to be discriminated are assumed to be a thousand yen ticket, a thousand yen ticket, a five thousand yen ticket, and a 10,000 yen ticket.

  FIG. 1 shows an external view of a portable banknote discriminating apparatus 1 that stores banknote discrimination software as a banknote discriminating apparatus according to the present embodiment. The bill discriminating method according to the present embodiment can be implemented using the bill discriminating apparatus 1. Moreover, the banknote discrimination software according to the present embodiment executes banknote discrimination processing by being installed in a terminal device, a computer device or the like having an intelligent function like the portable banknote discrimination device 1 according to the first embodiment. can do.

  The banknote discriminating apparatus 1 includes an image capturing unit 2, a built-in calculation unit 3 and storage unit 4 surrounded by a dotted line, an image display unit 5 a, an audio output unit 5 b, and an operation unit 6 as an output unit 5. Here, the operation unit 6 is configured by a touch panel, and is also used as the image display unit 5a. The bill discriminating apparatus 1 incorporates a portable battery (not shown) as a power source, and can be operated by turning on the power source.

  FIG. 2 shows a block configuration of the image capturing unit 2, the calculation unit 3, the storage unit 4, the output unit 5 including the image display unit 5a and the audio output unit 5b, and the operation unit 6.

  The image capturing unit 2 acquires the moving image data of the banknote that is the target of ticket type discrimination, and outputs it to the calculation unit 3. The image capturing unit 2 includes a lighting unit, a light receiving unit, an amplifier, and an analog / digital conversion circuit which are not shown. As a result, the illumination light is irradiated to the banknote to be identified by the bill type, the visible light is received by the light receiving unit, the output signal received by the amplifier is amplified, and converted into digital data by the analog / digital conversion circuit. To get video data.

  The illumination unit is not an essential component. However, by providing the illumination unit, it is possible to obtain accurate moving image data in determining the ticket type, and it is possible to determine the ticket type in a place where there is no illumination light. An incandescent bulb, a light emitting diode, a halogen lamp, sunlight, an HID lamp, an LED lamp, or the like can be used for the illumination unit, and is not limited. The light receiving unit is not limited as long as moving image data can be acquired. For example, a CCD area sensor camera or the like can be used.

  In the storage unit 4, for each banknote type of banknote to be discriminated in advance, that is, in this embodiment, for each thousand yen ticket, two thousand yen ticket, five thousand yen ticket, 10,000 yen ticket, each local In addition, local reference feature amount data representing typical features, and a luminance reference value, a continuous reference value, an accumulated reference value, and a vote reference value, which will be described later, are stored.

  The calculation unit 3 generates a plurality of pieces of local discrimination feature amount data in the discrimination target banknote using the given moving image data. Then, by comparing and collating the obtained plurality of local discriminating feature amount data and the plurality of local reference feature amount data for each bill type stored in advance in the storage unit 5, it is possible to determine the similarity of the bill type. Do. Based on the determination result, notification data is generated and output to the image display unit 5a and the audio output unit 5b. The calculation unit 3 controls operations of the image capturing unit 2, the storage unit 4, and the output unit 5.

  The image display unit 5a performs image display for notifying which ticket type is in accordance with the notification data.

  The sound output unit 5b performs sound output for notifying which ticket type is in accordance with the notification data. Moreover, the audio | voice output part 5b may generate | occur | produce the vibration etc. which differ for every ticket type instead of an audio | voice.

  As described above, the operation unit 6 is configured to input an image displayed on the image display unit 5a as a touch panel. Operation data input by the operator from the operation unit 6 is given to the calculation unit 3, the image capturing unit 2 captures an image according to the operation data, the operation unit 3 performs various calculations, and the storage unit 4 stores data. Various operations such as an operation for storing or reading the image, an operation for the image display unit 5a to display an image, and an operation for the sound output unit 5b to output sound or vibration are controlled.

  The operation unit 6 includes instructions for starting and ending the discrimination operation, luminance reference values related to the discrimination criteria, continuous reference values, cumulative reference values, setting changes for the vote reference values, setting changes relating to the output of notification data, for example, images Settings such as brightness, hue, saturation, contrast of contrast (contrast), volume output, and vibration pattern in display can be changed. Here, the operation unit 6 may be configured by a keyboard or the like independently of the image display unit 5a.

  A method for discriminating the bill type of the bill using the bill discriminating apparatus 1 having such a configuration will be described.

  The flowchart of FIG. 3 shows a procedure for generating and storing in advance a plurality of local reference feature data for each ticket type.

  As step S <b> 11, moving image data or still image data of banknotes is captured using the image capturing unit 2.

  In step S12, the calculation unit 3 acquires local reference image data indicating a plurality of local images from the captured moving image data, and generates a reference feature image data by detecting a plurality of feature positions for each ticket type. To do. For this processing, for example, a known FAST can be used.

  Here, the author's name, article name, journal name, and website address on the Internet in which information related to FAST is disclosed are described.

E. Rosten and T. Drummond, Machine learning for high-speed corner detection, European Conference on Computer Vision, volume 1, 2006
http://www.edwardrosten.com/work/rosten_2006_machine.pdf

  There are two desirable requirements for the feature position. (1) The same position can be stably extracted from the image data regardless of slight fluctuations in the shooting angle, shooting distance, etc. (2) Different shapes for each type of ticket It is a position where features often appear.

  By satisfying the requirement (1), there is a high possibility that the feature position for each ticket type stored in advance and the feature position extracted from the image data to be discriminated are extracted from the same position of the object. The possibility that it can be determined increases.

  Regarding requirement (2), for example, even if a feature is extracted from a black solid portion in an object, such a portion is highly likely to exist in other ticket types. Such a position where the feature of the shape for each ticket type is difficult to appear is not desirable in determining the ticket type.

  In step S13, the calculation unit 3 extracts feature information at a plurality of detected feature positions, and creates a plurality of local reference feature data. For this processing, for example, a known ORB can be used.

  Here, the author's name, thesis name, journal name, and website address on the Internet in which information related to the ORB is disclosed are described.

E. Rublee, V. Rabaud, K. Konolige, and G. Bradski, ORB: an efficient alternative to SIFT or SURF, Proceeding of IEEE International Conference on Computer Vision, 2011
https://willowgarage.com/sites/default/files/orb_final.pdf

  As step S14, the local reference feature data is stored in the storage unit 4.

  In addition, about step S11, S12, and S13, it is not necessary to necessarily implement using the banknote discrimination device 1. FIG. In that case, for example, a commercially available optical scanner may be used instead of the image capturing unit 2 in step S11, and a commercially available personal computer CPU may be used instead of the arithmetic unit 3 in steps S12 and S13. It is possible to create local reference feature data. However, in any case, the storage location of the local reference feature data in step S14 is always the storage unit 4 provided in the banknote discriminating apparatus 1.

  The flowchart of FIG. 4 shows a procedure for discriminating the bill type of the bill to be discriminated using a plurality of local reference feature data for each bill type.

  In step S21, the image capturing unit 2 is used to capture the moving image data of the banknote to be determined.

  In step S22, the calculation unit 3 acquires local discrimination image data indicating a plurality of local images from the captured moving image data, detects a plurality of feature positions for each ticket type, and determines discrimination feature image data. create.

  In step S23, the calculation unit 3 extracts feature information at a plurality of detected feature positions, and creates a plurality of pieces of local discrimination feature amount data.

  As step S24, a plurality of local discriminating feature amount data in the discrimination target banknote and a plurality of local reference feature amount data for each bill type stored in the storage unit 4 are collated to create notification data.

  Here, as a method for comparing the local discriminating feature amount data and the local reference feature amount data, for example, in order to search for an approximately similar portion between the two feature amount data, for example, kd-tree, LSH (Locality Sensitive) A known algorithm such as Hashing) may be used.

  Here, the author's name, thesis name, journal name, and website address on the Internet in which information related to kd-tree is disclosed are described.

JLBentley, Multidimensional Binary Search Trees in Database Applications, IEEE Transactions On Software Engineering, Vol.SE-5, No.4, July 1979
http://www.cse.usf.edu/~ytu/Literature/KDtree-SE79.pdf

  In addition, the author's name, article name, journal name, and website address on the Internet, in which information related to LSH is disclosed, are described.

A.Gionis, P.Indyk, and R.Motwani, Similarity Search in High Dimensions via Hashing, VLDB '99 Proceedings of the 25th International Conference on Very Large Data Bases
http://i.stanford.edu/~cgzhu/QualPaper/Similarity/Gionis_99_LSH.pdf

  In step S25, the notification data is output from the output unit 5.

  Here, as step S22, using the captured moving image data, the calculation unit 3 detects a plurality of local feature positions for each ticket type and creates discrimination feature image data. As step S23, the calculation unit 3 A method for extracting each feature at a plurality of detected local feature positions and creating a plurality of local discrimination feature amount data will be described in detail. In this method, as step 12 in FIG. 3, using the captured moving image data or still image data, the calculation unit 3 detects a plurality of local feature positions for each ticket type and creates reference feature image data. Then, as step S13, the method is the same as the method of extracting each feature at a plurality of local feature positions detected by the calculation unit 3 and creating a plurality of local reference feature data.

  As step S22, when performing the process which detects a some feature position, the local discrimination | determination image data which show the some local image in a banknote are first acquired from the moving image data which image | photographed and acquired the banknote first.

  5A to 5F show a plurality of local schematic diagrams in one banknote indicated by the local discrimination image data. FIG. 5A shows a partial pattern area in the 10,000 yen ticket, and FIG. 5B shows a Chinese character “壱” area in the 10,000 yen ticket. Further, FIGS. 5C to 5F show a numerical area of a fee amount in a 10,000 yen ticket.

  The plurality of local discrimination image data may partially overlap as shown in FIGS. 5D and 5E, or as shown in FIGS. 5E and 5F. Thus, they may be separated without overlapping.

  Then, the calculation unit 3 extracts a plurality of feature positions in such local discrimination image data.

  In this embodiment, an angle of an object (hereinafter referred to as “corner”) is used as a kind of image processing algorithm as a position that is highly likely to satisfy the two requirements (1) and (2) that are desirable as the characteristic position as described above. Is automatically detected by a corner detection algorithm, and the position is set as a feature position.

  When extracting a feature position, a luminance reference value and a continuous reference value stored in advance in the storage unit 4 are used.

  The luminance reference value is used when comparing the luminance value of the pixel of interest, which is determined sequentially one pixel at a time, with the luminance value of each pixel on the circumference around the pixel of interest in the local discrimination image data. Is a reference value for.

  The continuous reference value compares the luminance value of the pixel of interest with the luminance value of the pixel on the circumference around the pixel of interest, and when the pixels that are brighter or darker than the luminance reference value continue, This is the reference value.

  First, in one local image, one pixel at a time is set as a target pixel.

  The luminance reference value stored in advance in the storage unit 4 is compared with the luminance value of the pixel on the circumference around the target pixel with respect to the luminance value of the target pixel. When a pixel brighter than the luminance reference value or a pixel darker than the luminance reference value continues on the circumference by a number equal to or greater than the continuous reference value, the pixel of interest is detected as a “corner”. Such processing is repeated by sequentially setting all pixels as the target pixel. Then, a plurality of discriminating feature image data is generated by regarding the position of the target pixel detected as “corner” as the feature position.

  A procedure for acquiring a plurality of discriminating feature image data will be described with reference to FIG.

  FIG. 6A shows an example of one local image indicated by the local discrimination image data. As shown in FIG. 6B, one pixel included in this image is set as the target pixel P.

  For example, as shown in FIG. 6D, when the luminance value of the target pixel P is x, the upper limit value of pixels brighter than the luminance reference value is ya, the lower limit value of pixels darker than the luminance reference value is yb, and continuous. The reference value is set to “9”.

  A method of arranging 16 pixels from “0” to “15” as shown in FIG. 6C substantially along a circumference of 7 pixels in diameter is defined in the above FAST. Yes. If the predetermined number or more of pixels set in advance as the continuous reference value among the 16 pixels are continuously brighter than the upper limit value ya or lower than the lower limit value yb for the luminance value x of the target pixel P, The target pixel P is detected as a “corner”, that is, a feature position.

  As described above, in the present embodiment, the pixels on the circumference of the target pixel P having the luminance value x have a luminance equal to or higher than the upper limit value ya or lower than the lower limit value yb, and are continuously 9 on the circumference. Discrimination feature image data is generated by detecting a pixel that continues more than one as a feature position.

  The process of extracting the “corner” as described above is repeated by sequentially setting all pixels included in each image as the target pixel. For example, when one image is composed of 640 × 480 pixels, it is determined whether the “corner” is 307,200 times corresponding to the total number of images.

  A procedure for calculating a plurality of pieces of local discriminating feature amount data from pixel information at peripheral positions of the plurality of discriminating feature image data based on the plurality of obtained discriminating feature image data will be described with reference to FIG.

  FIG. 7A shows the feature position P indicated by the discrimination feature image data.

  As shown in FIG. 7B, the center of gravity position around the feature position indicated by the feature position P is calculated. Specifically, the center of gravity of the brightness of the pixels in the area surrounded by the line 11 is calculated.

  First, when obtaining the center of gravity position of the brightness in the gray scale image shown in FIG. 7B, the physical center of gravity is obtained by considering the brightness of each pixel in the area surrounded by the line 11 as the weight. Perform the calculation. As a method of considering the brightness of each pixel as a weight, the brightness value of each pixel, for example, “white” is 255, black is “0”, and the gray portion between them is “1” depending on the brightness. Any value between “254” is used as it is as the weight value. As coordinates, ordinate and abscissa in orthogonal coordinates in the image are used.

A mass point of weight (Wi) is placed at an arbitrary position (Xi, Yi) on Cartesian coordinates, and the center of gravity (X, Y) when there are a plurality of such mass points is
X = Σ (Xi × Wi) / ΣWi, Y = Σ (Yi × Wi) / ΣWi
It becomes.
The position of the center of gravity can be obtained by substituting the horizontal axis position, vertical axis position, and brightness value of each pixel into Xi, Yi, and Wi.

  Then, as shown in FIG. 8A, rotation correction is performed so that the center of gravity faces a predetermined direction set in advance. By performing the rotation correction, the influence caused by the difference in the orientation of the images of the respective discriminating feature image data obtained in the moving image data acquired at any time during shooting is eliminated.

  Next, a process for obtaining the feature amount at the feature position is performed by determining the brightness of the two pixels around the feature position.

  First, processing for obtaining a feature value at an arbitrary position in an image, which is defined in the ORB, will be described with reference to FIG.

  From the reference position, here the feature position 12, for example, the brightness of the pixel 13 at the position of 8 pixels to the right and 3 pixels at the bottom, and the 9 pixels from the feature position 12 to the right and 5 pixels upward. The brightness of the pixel 14 at the minute position is compared. If the pixel 12 is brighter, for example, “1” is substituted as the first value in the first digit of the feature amount, and “0” is assigned as the second value in the first digit of the feature amount otherwise. substitute.

  Similarly, the brightness of the first pixel (not shown) located 4 pixels rightward from the feature position 12 and 2 pixels upward, and 7 pixels rightward from the feature position 12 downward. The brightness of the second pixel (not shown) at the position corresponding to 12 pixels is compared. If the first pixel is brighter, “1” is assigned to the second digit of the feature value, and “0” is assigned to the second digit of the feature value otherwise. In this way, the brightness comparison of 256 sets of two points of pixels set in advance in the ORB is performed as shown in FIG. As a result, 256 comparison results, ie, 256-digit numbers consisting of “0” or “1” are obtained. A 256-digit number composed of “0” or “1” is used as a feature amount.

  The feature quantity obtained in this way has a property affected by the brightness of two pixels to be compared, that is, the brightness of each part in the image, and such a property is defined as a feature quantity. This feature amount is a list of 256 light / dark comparison results arranged side by side, and which of the 256 light / dark comparison results is assigned to which digit is preset.

  Then, according to the procedure shown in FIG. 10, the calculation unit 3 compares and collates the local discrimination feature value data in the obtained banknote to be discriminated with the local reference feature value data stored in the storage unit 4. To create notification data.

  In step S31, the plurality of local discrimination feature value data and the plurality of local reference feature value data for each ticket type stored in advance in the storage unit 4 are collated by similarity discrimination.

  As described above, each digit number in the feature amount obtained according to the ORB indicates the result of comparing the brightness of two pixels with the feature position as a reference. Therefore, when comparing the local discriminating feature amount data and the local reference feature amount data, the numbers of the same digits of the two feature amounts are compared, and the number of digits that do not match the numbers is counted.

  If the number of digits that do not match is the distance between the two feature values, the greater the distance between the two feature values, that is, the greater the number of mismatches in the light / dark comparison results, the more the two feature values are relatively similar. Defined as a state. Conversely, the smaller the distance between the two feature amounts, that is, the smaller the number of mismatches between the light and dark comparison results, the two feature amounts are defined to be relatively similar.

  For example, it is assumed that the local discrimination feature value data in the discrimination target banknote is “001110101”, and “10110001”, “00100101”, and “11111101” exist as the local reference feature value data for each bill type.

  There are two differences between the local discriminating feature data “001110101” and the local reference feature data “101100001”.

  There is one difference between the local discriminating feature amount data “00110101” and the local reference feature amount data “001001001”.

  There are three differences between the local discriminating feature data “001110101” and the local reference feature data “11111101”.

  In such a case, the local reference feature amount data most similar to the local discriminating feature amount data “001110101” is “00100101”.

  Here, a technique called approximate nearest neighbor search is used in the above kd-tree and LSH. For example, local reference feature data stored in advance in the storage unit 4 is divided into several groups of similar data. At the time of comparison and collation, first, it is determined which group the local determination feature value data is most similar to. Next, by collating the local reference feature data in the most similar group with the local discriminating feature data in a step-by-step comparison, matching with all local reference feature data Can be eliminated, and the number of verifications can be reduced to improve efficiency.

  In step S32, the local discriminating feature value data subjected to the similarity discrimination vote for the closest reference standard feature value data. The number of votes obtained for each ticket type is accumulated and generated as ticket type accumulated data.

  In step S33, the ticket type cumulative data for each ticket type is compared with the cumulative reference value stored in advance in the storage unit 4. Here, the cumulative reference value is a value used as a reference when comparing and verifying the ticket type cumulative data in order to generate the ticket type cumulative data and the notification data.

  In step S34, the ticket type accumulated data for each ticket type is compared with the accumulated reference value. If at least one ticket type accumulated data is equal to or greater than the accumulated reference value, the process proceeds to step S35.

  If the ticket type cumulative data of any of the ticket types is less than the cumulative reference value, the process proceeds to step S37 and notification data indicating that the ticket type is not determined is generated.

  In step S35, it is determined whether or not the ratio of the number of votes between the ticket type cumulative data of the ticket type with the highest number of votes and the ticket type cumulative data of the other ticket types with the second highest number of votes is equal to or higher than the vote reference value. Here, the vote reference value is a value used as a reference when comparing the ticket type accumulated data of the ticket type with the highest number of votes with the ticket type accumulated data of other ticket types in order to generate notification data.

  In step S36, when the ratio of the number of votes is equal to or more than the vote reference value, it is determined that the banknote to be discriminated is the ticket type of the ticket type accumulated data of the ticket type with the highest number of votes, and the notification data is generated.

  If the difference in the number of votes between the ticket type cumulative data of the ticket type with the highest number of votes and the ticket type cumulative data of other ticket types is less than the vote standard value, the process proceeds to step S37 and the ticket type is not determined. The notification data is generated.

  The output unit 5 performs notification by receiving the notification data generated by the calculation unit 3. As described above, the output unit 5 includes the screen display unit 5a and the audio output unit 5b. When the image display unit 5a performs screen display, in order to increase the difference between light and dark so that a visually impaired person can easily see, for example, display is performed by a combination of a cold color and a warm color or a combination of a primary color and a pastel color. May be.

  The bill discriminating method, the bill discriminating software, and the bill discriminating apparatus may have a call function of a mobile phone terminal. Further, the local reference feature data of the present invention can be received by a server (not shown) to access the update local reference feature data and update the data.

  As described above, according to the above-described embodiment of the present invention, a part of banknotes to be photographed cannot be collated due to the influence of wrinkles, dirt, shadows, uneven illumination, image deformation, reflection of a photographer's finger or the like. In this case, it is possible to discriminate the bill types of bill types of bills such as thousand yen, two thousand yen, five thousand yen and 10,000 yen from the remaining plurality of local discrimination feature data. In addition, even if the shooting distance or shooting angle varies, it is possible to accurately determine the ticket type. In the comparison and collation between the local discriminant feature quantity data and the local reference feature quantity data, the accuracy of collating each piece of data does not have to be high, and a plurality of local discriminant feature quantity data and a plurality of local reference feature data It is possible to increase the accuracy of ticket type discrimination stochastically by collating with the quantity data.

  In particular, for a visually impaired person, accurate bill type discrimination is possible without accurately aligning the position of the camera and the bill in order to acquire the image data of the bill. In addition, it is possible to accurately discriminate the bill type in a folded state without spreading a bill folded in a wallet or the like, which was impossible in the past.

  Although the embodiment of the present invention has been described, the above embodiment is presented as an example, and is not intended to limit the technical scope of the invention. The novel embodiment can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the spirit of the invention. The above-described embodiments and modifications thereof are included in the technical scope and gist of the invention, and are included in the invention described in the claims and the equivalents thereof.

DESCRIPTION OF SYMBOLS 1 Banknote discrimination apparatus 2 Image taking-in part 3 Calculation part 4 Storage part 5 Output part 5a Image display part 5b Audio | voice output part 6 Operation part P Attention pixel x Luminance value ya Upper limit value yb Lower limit value

The luminance value of the target pixel is compared with the luminance value of the pixel on the circumference around the target pixel. When the comparison result is that pixels that are brighter than the range of the brightness reference value stored in advance in the storage unit 4 or pixels that are darker than the brightness reference value continue continuously on the circumference by a number equal to or greater than the continuous reference value. The pixel of interest is detected as a “corner”. Such processing is repeated by sequentially setting all pixels as the target pixel. Then, a plurality of discriminating feature image data is generated by regarding the position of the target pixel detected as “corner” as the feature position.

Claims (15)

Acquiring movie data of bills subject to ticket type discrimination;
Using the moving image data, extracting features at a plurality of local feature positions to create a plurality of locally discriminating feature data;
The plurality of local discriminating feature amount data is similar to a plurality of local reference feature amount data serving as a reference from which features at a plurality of local feature positions for each preset ticket type are extracted and compared. Determining and creating notification data indicating the determination result;
Outputting the notification data to notify the determination result;
A bill discriminating method comprising: The step of creating the plurality of local reference feature data includes:
Acquiring local reference image data indicating a local image from moving image data or still image data acquired for each ticket type; and
The luminance reference value set in advance for the luminance value of the target pixel set one pixel at a time in the local reference image data is compared with the luminance value of the pixel on the circumference around the target pixel, and the luminance The process of detecting the pixel of interest as a corner when pixels that are brighter or darker than the reference value continue for a predetermined continuous reference value or more, and sequentially sets all the pixels of the local reference image data as pixels of interest. Obtaining a plurality of reference feature image data indicating the position of the target pixel detected as the corner; and
The center-of-gravity data of the pixel information indicated by the plurality of reference feature image data is calculated, and the plurality of reference feature image data is rotationally corrected using the center-of-gravity data so that the plurality of reference feature image data are directed in a predetermined direction. The correction reference feature image data is generated, the brightness values of a plurality of sets of two pixels in the plurality of correction reference determination feature image data are respectively compared, and one of the two pixels is set in advance. 2. The bill discriminating method according to claim 1, further comprising the step of creating the plurality of local reference feature data having a first value when bright and a second value otherwise. The step of creating the plurality of locally discriminating feature amount data includes:
Obtaining local discrimination image data indicating a local image from the moving image data;
The luminance reference value set in advance for the luminance value of the target pixel set one pixel at a time in the local discrimination image data is compared with the luminance value of the pixel on the circumference around the target pixel, and the luminance When pixels brighter or darker than the reference value continue for a predetermined continuous reference value or more, the process of detecting the pixel of interest as a corner is sequentially set with all the pixels of the local discrimination image data as pixels of interest. Obtaining a plurality of discriminating feature image data indicating the position of the target pixel detected as the corner; and
The center of gravity data of the pixel information indicated by the plurality of discriminant feature image data is calculated, and the plurality of discriminant feature image data is rotationally corrected so as to face a predetermined direction by using the barycentric data, and a plurality of Correction discrimination feature image data is generated, the brightness values of a plurality of sets of two points in each of the plurality of correction discrimination feature image data are respectively compared, and one of the two points is brighter. The bill discriminating method according to claim 1, further comprising a step of creating the plurality of local discriminating feature amount data having a first value in a case and a second value in other cases. . The step of creating the notification data includes:
The plurality of local discrimination feature quantity data and the plurality of local reference feature quantity data are compared and collated, and the ticket type of the local reference feature quantity data that most closely matches the local discrimination feature quantity data is voted , A step of accumulating the number of votes obtained for each ticket type to generate ticket type accumulated data;
The ticket type cumulative data for each ticket type is compared with the preset vote standard value, and the ticket type is not discriminated when the ticket type cumulative data of any ticket type is less than the vote standard value. Generating the notification data of:
When the ticket type cumulative data of at least one ticket type is greater than or equal to the vote standard value, the ticket type cumulative data of the ticket type with the first number of votes and the ticket type cumulative data of the other ticket types with the second number of votes. And the ratio is equal to or greater than a preset vote reference value, the notification data is generated as a ticket type to be notified of the ticket type of the ticket type accumulated data with the first number of votes, and the ratio is the vote The bill discriminating method according to any one of claims 1 to 3, further comprising a step of generating the notification data indicating that the bill type is not discriminated when it is less than a reference value.   5. The bill discriminating method according to claim 1, wherein the notification data is output by outputting at least one of screen display, vibration, and voice. In the bill discriminating software for causing a bill discriminating apparatus including an image capturing unit, a calculation unit, a storage unit, and an output unit to execute a method for discriminating a bill type of banknotes,
Step of acquiring movie data of bills subject to ticket type discrimination by the image capturing unit;
A step of extracting a feature at a plurality of local feature positions by using the moving image data and creating a plurality of locally discriminating feature amount data by the arithmetic unit;
A plurality of local reference features serving as a reference from which the plurality of local discriminating feature amount data and features at a plurality of local feature positions for each ticket type stored in advance in the storage unit are extracted by the arithmetic unit. Comparing and comparing quantity data to determine similarity, and creating notification data indicating the determination result;
Outputting the notification data to notify the determination result by the output unit;
A bill discriminating software that causes a bill discriminating apparatus to execute a bill discriminating method. The step of creating the plurality of local reference feature data includes:
Step of acquiring local reference image data indicating a local image from moving image data or still image data acquired for each ticket type by the calculation unit;
A luminance reference value set in advance for the luminance value of the target pixel set one pixel at a time in the local reference image data by the arithmetic unit, and a luminance value of the pixel on the circumference around the target pixel In comparison, when a pixel brighter or darker than the luminance reference value continues for a predetermined continuous reference value or more, the process of detecting the pixel of interest as a corner is focused on all the pixels of the local reference image data. Sequentially setting as pixels, obtaining a plurality of reference feature image data indicating the position of the target pixel detected as the corner;
The calculation unit calculates centroid data of pixel information indicated by the plurality of reference feature image data, and uses the centroid data to rotate the plurality of reference feature image data so as to face a predetermined direction set in advance. To generate a plurality of correction reference feature image data, compare the lightness of a plurality of sets of two pixels in the plurality of correction reference determination feature image data, respectively, and set one of the two points in advance. And a step of creating a plurality of local reference feature data having a first value when the pixel is brighter and a second value otherwise. Software. The step of creating the plurality of locally discriminating feature amount data includes:
Step of acquiring local discrimination image data indicating a local image from the moving image data by the arithmetic unit;
A luminance reference value set in advance for the luminance value of the target pixel set one pixel at a time in the local discrimination image data by the arithmetic unit, and the luminance value of the pixel on the circumference around the target pixel In comparison, when a pixel brighter or darker than the luminance reference value continues for a predetermined continuous reference value or more, the process of detecting the pixel of interest as a corner is focused on all the pixels of the local discrimination image data. Sequentially setting as pixels, and obtaining a plurality of discriminating feature image data indicating the position of the target pixel detected as the corner;
The calculation unit calculates centroid data of pixel information indicated by the plurality of discriminant feature image data, and uses the centroid data to rotate the plural discriminant feature image data so as to face a predetermined direction. To generate a plurality of correction discriminating feature image data, compare the brightness of a plurality of sets of two points of pixels in the plurality of correction discriminating feature image data, respectively, and set one of the two points of pixels set in advance The step of creating the plurality of local discriminating feature amount data having a first value when a pixel is brighter and a second value in other cases is provided. The banknote discrimination software described. The step of creating the notification data includes:
The arithmetic unit compares and collates the plurality of local discrimination feature quantity data with the plurality of local reference feature quantity data, and the local reference feature quantity data that most closely approximates the local discrimination feature quantity data. Voting on ticket types, accumulating the number of votes for each ticket type to generate ticket type cumulative data;
The calculation unit compares the ticket type accumulated data for each ticket type with a preset vote standard value, and the ticket type accumulated data for any ticket type is less than the vote standard value. Generating the notification data not to determine the species;
When the accumulated data of at least one ticket type is equal to or greater than the vote standard value by the calculation unit, the accumulated ticket type data of the ticket type with the first number of votes and the other ticket type with the second number of votes. When the ratio of the ticket type accumulated data is equal to or higher than a preset vote reference value, the notification data is generated as the ticket type to be notified of the ticket type of the ticket type accumulated data with the first number of votes, The bill discriminating software according to any one of claims 6 to 8, further comprising a step of generating the notification data indicating that the bill type is not discriminated when the ratio is less than the vote reference value. .   10. The bill discriminating software according to claim 6, wherein the notification data is output by the output unit by at least one of screen display, vibration, and voice. An image capture unit that acquires movie data of bills subject to ticket type discrimination;
A storage unit that stores in advance a plurality of local reference feature amount data serving as a reference from which features at a plurality of local feature positions for each ticket type are extracted;
Using the moving image data, features at a plurality of local feature positions are extracted to create a plurality of local discrimination feature amount data, and the plurality of local discrimination feature amount data and stored in the storage unit An arithmetic unit for comparing and collating the plurality of local reference feature data for each ticket type to determine similarity, and creating notification data indicating a determination result;
An output unit for outputting the notification data to notify the determination result;
A bill discriminating apparatus comprising: When the calculation unit creates the plurality of local reference feature data,
Obtain local reference image data indicating a local image from moving image data or still image data obtained for each ticket type,
The luminance reference value set in advance for the luminance value of the target pixel set one pixel at a time in the local reference image data is compared with the luminance value of the pixel on the circumference around the target pixel, and the luminance The process of detecting the pixel of interest as a corner when pixels that are brighter or darker than the reference value continue for a predetermined continuous reference value or more, and sequentially sets all the pixels of the local reference image data as pixels of interest. Obtaining a plurality of reference feature image data indicating the position of the target pixel detected as the corner,
The center-of-gravity data of the pixel information indicated by the plurality of reference feature image data is calculated, and the plurality of reference feature image data is rotationally corrected using the center-of-gravity data so that the plurality of reference feature image data are directed in a predetermined direction. The correction reference feature image data is generated, the brightness values of a plurality of sets of two pixels in the plurality of correction reference determination feature image data are respectively compared, and one of the two pixels is set in advance. The banknote discriminating device according to claim 11, wherein a plurality of local reference feature amount data having a first value when bright and a second value otherwise are created. When the calculation unit creates the plurality of local discrimination feature data,
Obtaining local discrimination image data indicating a local image from the moving image data,
The luminance reference value set in advance for the luminance value of the target pixel set one pixel at a time in the local discrimination image data is compared with the luminance value of the pixel on the circumference around the target pixel, and the luminance When pixels brighter or darker than the reference value continue for a predetermined continuous reference value or more, the process of detecting the pixel of interest as a corner is sequentially set with all the pixels of the local discrimination image data as pixels of interest. Obtaining a plurality of discriminating feature image data indicating the position of the target pixel detected as the corner,
The center of gravity data of the pixel information indicated by the plurality of discriminant feature image data is calculated, and the plurality of discriminant feature image data is rotationally corrected so as to face a predetermined direction by using the barycentric data, and a plurality of Correction discrimination feature image data is generated, the brightness values of a plurality of sets of two points in each of the plurality of correction discrimination feature image data are respectively compared, and one of the two points is brighter. The banknote discriminating device according to claim 11 or 12, wherein the plurality of local discriminating feature amount data having a first value in a case and a second value in other cases is created. When the calculation unit creates the notification data,
The plurality of local discrimination feature quantity data and the plurality of local reference feature quantity data are compared and collated, and the ticket type of the local reference feature quantity data that most closely matches the local discrimination feature quantity data is voted , Accumulate the number of votes for each ticket type to generate ticket type cumulative data,
The ticket type cumulative data for each ticket type is compared with the preset vote standard value, and the ticket type is not discriminated when the ticket type cumulative data of any ticket type is less than the vote standard value. Generating the notification data of
When the ticket type cumulative data of at least one ticket type is greater than or equal to the vote standard value, the ticket type cumulative data of the ticket type with the first number of votes and the ticket type cumulative data of the other ticket types with the second number of votes. And the ratio is equal to or greater than a preset vote reference value, the notification data is generated as a ticket type to be notified of the ticket type of the ticket type accumulated data with the first number of votes, and the ratio is the vote The bill discriminating device according to any one of claims 11 to 13, wherein the notification data indicating that the bill type is not discriminated is generated when the value is less than a reference value.   The bill output device according to any one of claims 11 to 14, wherein the output unit outputs the notification data by at least one of screen display, vibration, and voice.

Images