JP2018037020A - Information processing device and information processing program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an information processing device for reducing a calculation amount in collation processing in performing parallel processing.SOLUTION: Reference data storage means of the information processing device subtracts the average value of values of elements in reference data composed of a plurality of elements from values of the elements, and stores a plurality of reference data obtained by normalizing the length of a multidimensional vector composed of the elements, normalization means subtracts the average value of values of elements in inspection data composed of a plurality of elements from values of the elements, and normalizes the length of the multidimensional vector composed of the elements, collation means collates each normalized reference data stored in the reference data storage means with normalized inspection data by the normalization means, and determination means compares a collation result by the collation means with a predetermined threshold, consolidates collation results of a plurality of reference data and the inspection data to determine the authenticity of the inspection data, and performs the normalization means and the collation means with parallel processing.SELECTED DRAWING: Figure 1

Description

  The present invention relates to an information processing apparatus and an information processing program.

  Patent Document 1 has an object to easily and accurately determine the authenticity of a solid, and the entanglement of the fibrous material forming the paper has randomness that cannot be controlled at the time of manufacture. The transparency of the paper also has a random change unique to each paper. By using this, the reference area on the paper (for example, a non-printed portion of a size of about 2 × 2 mm) is optically read and registered as reference data. At the time of authenticity determination, a collation area including the reference area and larger than the reference area (for example, about 4 × 4 mm) is read by a scanner, and a partial area having the same size as the reference area is obtained from the collation data obtained by the reading. The calculation of the correlation value with the reference data by the normalized correlation method is repeated while shifting the position of the partial area in the collation area, and the maximum correlation value and the no It is disclosed that true / false judgment is performed by comparing each of the score with a threshold value.

  In Patent Document 2, it is an object to easily and accurately determine the authenticity of a solid. A reference area on a real paper is optically read from two different directions and registered as a reference image. For the paper to be falsely determined, the collation area including the reference area and having a size larger than the reference area is read in two different directions by the scanner, and the partial area data having the same size as the reference area is obtained from the respective collation data obtained by the reading. Extracting and optically reading the reference image from the same direction and the collation image, calculating the correlation value with the reference image by the normalized correlation method, while shifting the position of the partial area in the collation area Repeatedly, authenticity is determined by comparing the maximum correlation value and the normalized score of the maximum value with each threshold value, and both are determined to be “true” in the authenticity determination in each group. The first-time users of authenticity determination target of the paper it is disclosed that determines "true".

  In Patent Document 3, imaging attribute information such as an imaging device, a part, and an imaging direction is automatically extracted from a medical image by image processing and added to management information to enable efficient management of a huge amount of medical images. The template image storage unit stores a template image related to a category to be identified for each classification, and the image classification unit classifies the input image from the management information and the image size of the medical image, The image selection unit selects a template image of a plurality of categories that are candidates for identification using the classification result of the input image from the template image storage unit, and the image identification unit compares the selected template image with the input image. The category of the template image that best matches is determined, and the management information adding unit manages the imaging attribute information of the template image of the identified category It is disclosed to be added to the distribution.

  In Patent Document 4, even when a large amount of images are stored in a database, the memory capacity required for the clustering process is small, and the database can be updated sequentially. An object of the present invention is to provide a database device. A database device for image search uses a memory for storing a range of subspaces in a vector space and a feature vector of a plurality of images for use in clustering processing. In this database device, a local feature amount of each of a plurality of feature points included in an image is represented as a feature amount vector, and the vector includes a plurality of feature amount vectors. Clustering is performed in the space, and the vector space is divided into multiple subspaces. The search table indicating a correspondence relationship between each image and while is generated is disclosed.

  In Patent Document 5, it is an object to provide an information search technique that enables a user to search his / her intended information from a large amount of information with a simple operation. Presenting, inputting user's suitability judgment information for it, creating search key information by taking, for example, the average value of the feature amount from the information that has been matched, and the created search key information and sample information The similarity between the search key information and the matched information is close to 1, for example, and the similarity between the search key information and the non-matched information is close to 0, for example. Then, create a conversion rule from the distance defined for each feature quantity to similarity, determine the integrated form of similarity with the higher level of the information that is relevant, and from the integrated form and the created conversion rule , Accumulated in the storage means To search for information that is similar to the search key information from the broadcast is disclosed.

JP 2005-038389 A JP 2006-053736 A JP 2002-253539 A JP 2010-009332 A Japanese Patent Application Laid-Open No. 2003-077617

There is a technique for determining the authenticity of a solid. Specifically, authenticity determination is performed by matching processing between reference data obtained by photographing a true object and examination data obtained by photographing an object that is an inspection target (that is, whether the authenticity is unknown at that time).
However, in the case of authenticity confirmation processing as this authenticity determination processing, it is necessary to perform authenticity determination processing between a large number of information on genuine products and target product information, In order to cope with the displacement due to position, rotation, etc., a large amount of calculation is required. In order to cope with any deviation, the reference data becomes enormous, and simply performing parallel processing takes processing time.
It is an object of the present invention to provide an information processing apparatus and an information processing program that reduce the amount of calculation in collation processing when performing parallel processing.

The gist of the present invention for achieving the object lies in the inventions of the following items.
The invention according to claim 1 normalizes the length of a multidimensional vector constituted by the elements by subtracting the average value of the elements in the reference data constituted by a plurality of elements from the value of the elements A plurality of reference data storage means for storing a plurality of reference data and an average value of element values in inspection data constituted by a plurality of elements are subtracted from the value of the element, Normalization means for normalizing the length of the dimension vector; collation means for collating each reference data after normalization stored in the reference data storage means and inspection data after normalization by the normalization means; A determination means for comparing the verification result by the verification means with a predetermined threshold value, and aggregating the verification results of the plurality of reference data and the inspection data to determine the authenticity of the inspection data. The above Performing the comparison means and-normalized means in a parallel processing, an information processing apparatus.

  According to a second aspect of the present invention, the normalized reference data and the normalized inspection data are vectors, and a plurality of the normalized reference data are combined into a matrix, and image processing is performed from the inspection data. 2. The matrix according to claim 1, wherein a plurality of pieces of data converted and normalized according to claim 1 are combined into a matrix, and the collating unit collectively calculates correlation values with the inspection data subjected to the image processing with respect to a plurality of reference data. Information processing apparatus.

  A third aspect of the present invention is the information processing apparatus according to the second aspect, wherein the processing by the determination means is performed by parallel calculation.

  According to a fourth aspect of the present invention, there is provided the information processing apparatus according to the third aspect, wherein the processing by the determination means is performed by a mapper or a reducer in parallel calculation, or by an executor or a driver.

  The invention according to claim 5 further comprises image processing means for performing image processing including affine transformation on the inspection data, wherein the reference data and the inspection data are images, and the normalization means includes the image processing 5. The information processing apparatus according to claim 2, wherein normalization processing is performed on inspection data image-processed by the means.

  The invention of claim 6 stores, as the affine transformation, a conversion matrix for performing conversion processing and partial data extraction processing by rotation at a predetermined angle, enlargement / reduction at a predetermined magnification, or a combination thereof. The information processing apparatus according to claim 5, wherein matrix information is executed to create calculation data for inspection data.

  The invention of claim 7 further normalizes each of a plurality of pieces of data created by converting from inspection data by image processing, sets the normalized individual data as a vector, and forms a matrix by combining a plurality of pieces of data. Management means for managing position information in the matrix, storage means for storing information associated with the original information from the position in the matrix, and original information from the position in the matrix using the correspondence The information processing apparatus according to claim 2, further comprising an extraction unit configured to extract.

  The invention of claim 8 further comprises second storage means for storing a determination rule corresponding to the position of the matrix based on the correspondence, and the determination means determines authenticity using the determination rule. An information processing apparatus according to claim 7.

  According to the ninth aspect of the present invention, when a plurality of inspection data is simultaneously input to the determination unit, information for identifying the original inspection data is notified to the determination unit, or before the processing of the determination unit The information processing apparatus according to claim 8, further comprising a processing unit that divides the collation result according to the original inspection data.

  The invention according to claim 10 is the information processing apparatus according to any one of claims 1 to 9, wherein the normalizing means sets a value that is an out-of-target element to zero.

  The invention according to claim 11 is the reference data storage means for storing the average value of the elements in the reference data constituted by a plurality of elements, a plurality of reference data as the values of the elements, and the plurality of elements Generating means for generating an average value of the element values in the inspection data configured as inspection data, and collation for comparing each reference data stored in the reference data storage means with the inspection data by the generating means A determination means for comparing the result of the verification by the verification means and a predetermined threshold value, and aggregating the verification results of the plurality of reference data and the inspection data to determine the authenticity of the inspection data And an information processing apparatus that performs the normalization means and the collation means in parallel processing.

  In the invention of claim 12, the computer subtracts the average value of the element values in the reference data constituted by a plurality of elements from the value of the elements, and the length of the multidimensional vector constituted by the elements The reference data storage means for storing a plurality of reference data normalized, and the average value of the element values in the inspection data constituted by a plurality of elements is subtracted from the value of the elements, and constituted by the elements Normalization means for normalizing the length of the multidimensional vector to be performed, each reference data after normalization stored in the reference data storage means and the inspection data after normalization by the normalization means are collated Judgment of the verification data by comparing the verification result by the verification unit with a predetermined threshold value and aggregating the verification results of the plurality of reference data and the inspection data To function as a step, it performs the verification means and the normalizing means in a parallel processing, an information processing program.

  In the invention of claim 13, the computer comprises a reference data storage means for storing an average value of element values in the reference data constituted by a plurality of elements, and a plurality of reference data which are the values of the elements; A generation unit that generates, as inspection data, an average value of elements in inspection data constituted by a plurality of elements, each reference data stored in the reference data storage unit, and inspection data by the generation unit The collation means for collation and the collation result by the collation means are compared with a predetermined threshold value, and the collation results of the plurality of reference data and the inspection data are aggregated to determine the authenticity of the inspection data An information processing program that functions as a determination unit and performs the normalization unit and the collation unit in parallel processing.

  According to the information processing apparatus of the first aspect, when performing parallel processing, the amount of calculation in the collation processing is reduced.

  According to the information processing apparatus of the second aspect, correlation values with the inspection data subjected to image processing are collectively calculated for a plurality of reference data.

  According to the information processing apparatus of the third aspect, the authenticity is determined by parallel calculation.

  According to the information processing apparatus of the fourth aspect, the mapper or reducer, or executor or driver in the parallel calculation makes a true / false determination.

  According to the information processing apparatus of the fifth aspect, the normalization process is performed after the image process is performed on the inspection data.

  According to the information processing apparatus of claim 6, as the affine transformation, a conversion matrix that performs a conversion process and a partial data cut-out process by rotation at a predetermined angle, enlargement / reduction at a predetermined magnification, or a combination thereof. Are stored, and a matrix operation is executed to generate data for calculation of inspection data.

  According to the information processing apparatus of the seventh aspect, the collation calculation considering the image processing can be collectively performed at high speed by the matrix calculation.

  According to the information processing apparatus of the eighth aspect, the authenticity is determined using the determination rule.

  According to the information processing apparatus of claim 9, when a plurality of inspection data is simultaneously input to the determination unit, information for identifying the original inspection data is notified to the determination unit, or the collation result is obtained before the determination process. It is divided according to the original inspection data.

  According to the information processing apparatus of the tenth aspect, normalization is performed so that a value that is a non-target element is zero.

  According to the information processing apparatus of the eleventh aspect, when performing parallel processing, the amount of calculation in the collation processing is reduced.

  According to the information processing program of the twelfth aspect, when performing parallel processing, the amount of calculation in the collation processing is reduced.

  According to the information processing program of the thirteenth aspect, when performing parallel processing, the amount of calculation in the collation processing is reduced.

It is a conceptual module block diagram about the structural example of this Embodiment. It is explanatory drawing which shows the authenticity determination processing example. It is explanatory drawing which shows the parallel processing example of authenticity determination. It is a flowchart which shows the process example by this Embodiment. It is a flowchart which shows the process example by this Embodiment. It is a flowchart which shows the process example by this Embodiment. It is a flowchart which shows the process example by this Embodiment. It is a flowchart which shows the process example by this Embodiment. It is a flowchart which shows the process example by this Embodiment. It is a flowchart which shows the process example by this Embodiment. It is explanatory drawing which shows the system configuration example of this Embodiment. It is a block diagram which shows the hardware structural example of the computer which implement | achieves this Embodiment.

First, before describing the present embodiment, the premise or an information processing apparatus using the present embodiment will be described. This description is intended to facilitate understanding of the present embodiment.
There is a technique for determining the authenticity of inspection data by comparing reference data and inspection data. For example, in Patent Document 1, the authenticity of a solid in which readable and unique features having randomness are distributed along the surface is determined. The reference data is data representing features distributed on the true solid obtained by reading the features of the true solid in advance, and is optically read, for example. More specifically, it is image data obtained by irradiating a true solid with light and reading reflected light or transmitted light. The inspection data is data representing features distributed on the determination target solid by reading the characteristics of the determination target solid, for example, optically read. More specifically, it is image data obtained by irradiating a solid to be determined with light and reading reflected light or transmitted light.
Then, based on the reference data and the inspection data, data representing characteristics distributed in a first region of a predetermined size on one of the true solid and the solid to be determined, and the first region on the other solid The calculation of the correlation value with the data representing the feature distributed in the second area of the same size is repeated while moving the position of the second area on the other solid in the area larger than the predetermined size. Then, the maximum value of the plurality of correlation values obtained by the calculation is equal to or greater than the first predetermined value, and a value obtained by subtracting the average value of the correlation values from the maximum correlation value is divided by the standard deviation of the correlation values. Based on whether or not the normalized score of the maximum correlation value obtained in this way is greater than or equal to the second predetermined value, the authenticity of the object to be determined is determined.
More specifically, the same processing is repeated by shifting one dot at a time with respect to the positional deviation from the reference data. Then, the rotation data and the size shift are dealt with by rotating and enlarging / reducing the inspection data (inspection image). In this case, it is not always necessary to register the data rotated or enlarged on the reference data side. Since the registration data (registered image) has an enormous amount of data, if it is not necessary to duplicate and store these, a large efficiency effect can be obtained both in terms of storage capacity and data transfer. However, the reference data may be rotated and enlarged / reduced. Although the processing load for one calculation is relatively light, the actual processing load is heavy because it is possible to cope with deviations in position, rotation, etc. by repeated calculation. More specifically, when collating one piece of inspection data, the large number of reference data to be collated is the main cause of the heavy processing load. The reference data side can easily exceed 100 million, but even if the verification process between one inspection data and one reference data is in the order of ms, it will finish processing for all the reference data exceeding 100 million. Is a calculation that takes days.

Hereinafter, an example of a preferred embodiment for realizing the present invention will be described with reference to the drawings.
FIG. 1 shows a conceptual module configuration diagram of a configuration example of the present embodiment.
The module generally refers to components such as software (computer program) and hardware that can be logically separated. Therefore, the module in the present embodiment indicates not only a module in a computer program but also a module in a hardware configuration. Therefore, the present embodiment is a computer program for causing these modules to function (a program for causing a computer to execute each procedure, a program for causing a computer to function as each means, and a function for each computer. This also serves as an explanation of the program and system and method for realizing the above. However, for the sake of explanation, the words “store”, “store”, and equivalents thereof are used. However, when the embodiment is a computer program, these words are stored in a storage device or stored in memory. This means that control is performed so as to be stored in the apparatus. Modules may correspond to functions one-to-one, but in mounting, one module may be configured by one program, or a plurality of modules may be configured by one program, and conversely, one module May be composed of a plurality of programs. Note that one module may include other modules. Hereinafter, “connection” is used not only for physical connection but also for logical connection (data exchange, instruction, reference relationship between data, etc.). “Predetermined” means that the process is determined before the target process, and not only before the process according to this embodiment starts but also after the process according to this embodiment starts. Also, if it is before the target processing, it is used in accordance with the situation / status at that time or with the intention to be decided according to the status / status up to that point. When there are a plurality of “predetermined values”, they may be different values, or two or more values (of course, including all values) may be the same. In addition, the description of “do B when A” is used to mean “determine whether or not A and do B when A”. However, the case where it is not necessary to determine whether or not A is excluded.
In addition, the system or device is configured by connecting a plurality of computers, hardware, devices, and the like by communication means such as a network (including one-to-one correspondence communication connection), etc., and one computer, hardware, device. The case where it implement | achieves by etc. is also included. “Apparatus” and “system” are used as synonymous terms. Of course, the “system” does not include a social “mechanism” (social system) that is an artificial arrangement.
In addition, when performing a plurality of processes in each module or in each module, the target information is read from the storage device for each process, and the processing result is written to the storage device after performing the processing. is there. Therefore, description of reading from the storage device before processing and writing to the storage device after processing may be omitted. Here, the storage device may include a hard disk, a RAM (Random Access Memory), an external storage medium, a storage device via a communication line, a register in a CPU (Central Processing Unit), and the like.

The information processing apparatus 100 according to the present embodiment determines the authenticity of the inspection data. As illustrated in the example of FIG. 1, the reference data processing module 105, the reference data database 125, and the inspection data processing module 140 are used. have.
The information processing apparatus 100 performs parallel processing (also referred to as distributed processing or parallel distributed processing). For example, there are Spark, Storm, Hadoop, etc., and when processing an image, a GPU (Graphics Processing Unit) may be used. Computation is implemented in a numerical matrix calculation that can realize high-speed parallel computing with a parallel processing mechanism.
In addition, data that is cut out in consideration of movement, rotation, enlargement / reduction, distortion, etc. is prepared for the inspection data. When tensor calculation is possible, this cut-out can also be performed at high speed using high-speed tensor calculation processing (a library such as CuDNN).
This will be described in more detail.
One inspection data becomes one vector when pixels are arranged one-dimensionally. By multiplying this vector by a matrix indicating which pixel is extracted, a vector from which a partial image is extracted can be created. Similarly, scaling and rotation can also be realized as matrix calculation by preparing a matrix corresponding to the designated conversion. If tensor calculation can be realized at high speed by using GPU or the like, high speed can be expected by calculating these matrices together as a tensor.
Partial segmentation (corresponding to shifting the position and collating), rotation, and enlargement / reduction can be combined. If the total number of conversion processes is N, a vector obtained by converting the inspection data can be created corresponding to each. That is, N inspection images for collation calculation are generated from one inspection image. These inspection images for verification calculation are treated as vectors, and normalization is performed for each vector. Then, by arranging the normalized vectors to form a matrix on the inspection image side, and simply multiplying the matrix on the registered reference data side, a calculation for correlating a large amount of reference data is performed. Can be done together. These processes can be speeded up by parallel processing on the GPU together with distributed processing. The above-described “perform normalization for each vector” will be described in detail. In order to strictly execute Equation 1 described later, normalization for each vector is necessary. However, when the accuracy may be somewhat reduced (when the performance is high enough to determine authenticity), an average value for each of a plurality of vectors may be used.

The reference data processing module 105 includes a registration data receiving module 110, an optimization processing (reference) module 115, and a registration module 120. The reference data processing module 105 creates reference data. The processing of the reference data processing module 105 will be described later using the flowchart shown in the example of FIG.
The registration data receiving module 110 is connected to the optimization processing (reference) module 115. The registration data receiving module 110 receives the reference data and passes the reference data to the optimization processing (reference) module 115. The reference data may be an image. Accepting an image means, for example, reading an image with a scanner, a camera, etc., receiving an image from an external device via a communication line by fax, etc., a hard disk (in addition to what is built in a computer, via a network) And the like, and the like read out the images stored in the device etc.). The image may be a binary image or a multi-value image (including a color image). One image may be received or a plurality of images may be received. Further, as a solid photographed in the image, any solid material having random and readable unique features distributed along the surface may be used, for example, various paper documents (in addition to banknotes and securities, For example, there are passports, various rights documents, resident's cards, birth certificates, insurance certificates, guarantees, confidential documents, etc.), product surfaces (for example, drugs (especially tablets)), and the like.

The optimization processing (reference) module 115 is connected to the registration data receiving module 110 and the registration module 120. The optimization (reference) module 115 subtracts the average value of the element values in the reference data constituted by a plurality of elements from the value of the element, and performs multidimensional vector normalization constituted by the elements. Do. The normalization here is normalization in which the length of the multidimensional vector is set to a predetermined numerical value. This “predetermined numerical value” may be 1, for example. Even if it is not necessarily 1, it can be calculated in the same way even if it is multiplied by a constant, for example.
The normalized reference data may be a matrix.
When the reference data is an image, the optimization processing (reference) module 115 may perform image processing including affine transformation on the reference data.
Then, the optimization processing (reference) module 115 may perform normalization processing on the reference data subjected to image processing.
Further, as the affine transformation, a conversion matrix for performing a conversion process and a partial image cut-out process by rotation at a predetermined angle, enlargement / reduction at a predetermined magnification, or a combination thereof is stored, and the conversion matrix is The calculation data of the inspection data may be generated by executing matrix calculation using the data.
In addition, image processing (for example, partial cutout, rotation, enlargement / reduction, other affine transformations, distortion correction, pixel value change processing, etc. is included (in the description of this embodiment, enlargement etc. This is shown as an example of image processing))), and normalize multiple data created by converting from inspection data, and each normalized data is used as a vector, and the data is combined into a matrix However, position information in the matrix may be managed. Then, information associated with the original information from the position in the matrix may be stored, and the original information may be extracted from the position within the matrix using the correspondence. Here, the position in the matrix includes a row number or a column number, and the original information includes information on a cutout position, a rotation angle, a scaling factor, and the like.
Further, a determination rule corresponding to the position of the matrix may be stored based on the correspondence, and the collation module 160 may determine the authenticity using the determination rule. For example, when an image converted from an inspection image is prepared at a small angle (for example, 72 ways in increments of 5 degrees) in order to cope with a shift in rotation or enlargement / reduction, a plurality of close conversion angles may be used depending on circumstances. There may be a case where the collation calculation result exceeds a threshold value for a thing. In order to be able to execute a process that distinguishes a case where the threshold value has been exceeded multiple times between greatly different rotation angles in this case, a determination rule according to the conversion method when the inspection image is converted is constructed. It becomes desirable.
In addition, when a plurality of inspection data is simultaneously input to the collation module 160, information for identifying the original inspection data is notified to the collation module 160, or the collation result is sent to the original inspection image before the processing of the collation module 160. You may make it divide | segment according to.
The process of the optimization process (reference) module 115 will be described later using the flowchart shown in the example of FIG.
The registration module 120 is connected to the optimization processing (reference) module 115 and the reference data database 125. The registration module 120 forms the reference data processed by the optimization processing (reference) module 115 into a matrix and registers it in the reference data database 125. The processing of the registration module 120 will be described later using the flowchart shown in the example of FIG.
Note that, as described above, the optimization process (reference) module 115 does not perform the normalization process, and calculates the average value of the element values in the reference data constituted by a plurality of elements as the value of the element. A plurality of pieces of reference data may be generated.

The reference data database 125 includes a target part information storage module 130 and a reference data storage module 135, a registration module 120 of the reference data processing module 105, a collation module 160 of the inspection data processing module 140, and an optimization process (inspection). A module 155 is connected. The reference data database 125 stores processing results (reference data) by the reference data processing module 105. For example, the reference data database 125 is desirably a DB (DataBase) in a distributed environment.
The target part information storage module 130 stores the inspection data processed by the optimization processing (inspection) module 155. The collation module 160 may collate the reference data in the reference data storage module 135 with the inspection data in the target part information storage module 130.
The reference data storage module 135 subtracts the average value of the element values in the reference data constituted by a plurality of elements from the value of the element, and normalizes the length of the multidimensional vector constituted by the elements A plurality of reference data obtained by performing is stored. Specifically, a plurality of reference data of processing results by the optimization processing (reference) module 115 and the registration module 120 are stored.
When the reference data is subjected to the collation processing, the distribution processing is performed in a matrix in which a plurality of reference data are collected. There are cases where the data are stored together in this matrix, and when the data is distributed to the distributed / parallel calculation, the data is arranged in a matrix.
In both cases, information indicating which position (row, column) of the matrix is the original reference data needs to be held. That is, it is necessary to display which reference data matches the data when the collation is confirmed.
This information may be stored, for example, in the reference data storage module 135 and referenced when reporting results. In this case, since the matrix ID (identifier) and the position in the matrix (information on the number of rows and columns) are used to make an inquiry, it is necessary to distribute the matrix with an ID. There is. Further, instead of directly assigning information corresponding to the ID to the matrix, the information corresponding to the ID may be calculated from the ID of the job (Job, processing) and the number information of the matrix handled in the job.
On the other hand, the correspondence information of the identifier of the reference data with the position in each matrix may be distributed together during distributed parallel calculation. In this case, when the collation is successful, the information on the identifier of the corresponding reference data can be selected by the apparatus performing distributed parallel calculation and sent to the subsequent processing.

The inspection data processing module 140 includes an inspection data receiving module 145, a molding module 150, an optimization processing (inspection) module 155, a collation module 160, and an output module 165.
In this embodiment, the collation module 160 operates in a distributed parallel manner to increase the speed. There is no necessity that there are a plurality of output modules 165, inspection data receiving modules 145, molding modules 150, and optimization processing (inspection) modules 155 for the verification module 160. That is, although there are a plurality of collation modules 160, there may be one other module (output module 165, inspection data receiving module 145, molding module 150, optimization processing (inspection) module 155).
The inspection data processing module 140 performs processing on the inspection data, compares it with the reference data, and determines whether the inspection data is true or false. The processing of the inspection data processing module 140 will be described later using the flowchart shown in the example of FIG.
The inspection data reception module 145 is connected to the molding module 150. The inspection data reception module 145 receives inspection data and passes the inspection data to the molding module 150. The inspection data may be an image. To accept an image, for example, reading an image with a scanner, a camera, etc., receiving an image from an external device via a communication line with a fax, etc. In addition to what is built in the computer, it includes reading out an image stored in a computer etc.). The image may be a binary image or a multi-value image (including a color image). One image may be received or a plurality of images may be received. In the case of a plurality of sheets, information that can be associated with the processing result of the verification module 160 and the original inspection data is stored. Or you may make it pass to the collation module 160 (or output module 165 not via the collation module 160) with the test data and the data which shape | molded and normalized the test data.
Further, as a solid imaged in the image, any solid material that is an object to be inspected and in which readable unique features having randomness are distributed along the surface may be used. For example, various paper documents (banknotes) In addition to securities and securities, there are, for example, passports, various rights documents, resident's cards, birth certificates, insurance certificates, guarantees, confidential documents, etc.) and product surfaces (for example, drugs (especially tablets)). If the image to be inspected corresponds to the reference image, it is determined to be true, and if different from any reference image, it is determined to be false. Note that “the image to be inspected corresponds to the reference image” indicates that the same solid is captured.

The molding module 150 is connected to the inspection data receiving module 145 and the optimization processing (inspection) module 155. The forming module 150 forms the inspection data for verification.
When the inspection data is an image, the molding module 150 may perform image processing including affine transformation on the inspection data.
The processing of the molding module 150 will be described later using the flowchart shown in the example of FIG.

The optimization processing (inspection) module 155 is connected to the reference data database 125, the molding module 150, and the verification module 160. The optimization processing (inspection) module 155 subtracts the average value of the element values in the inspection data constituted by a plurality of elements from the value of the element, and the length of the multidimensional vector constituted by the elements Perform normalization. The normalization here is normalization in which the length of the multidimensional vector is set to a predetermined numerical value. This “predetermined numerical value” may be 1, for example. Even if it is not necessarily 1, it can be calculated in the same way even if it is multiplied by a constant, for example.
Note that the inspection data after normalization may be a matrix. Regarding the matrix processing, specifically, an image cut out from the inspection data while shifting the position, a rotated image, an enlarged or reduced image, and other affine transformations, distortion correction, and pixel value change What is necessary is just to put together the normalized image data of image processing such as processing and the result of combining these processings into a matrix. At this time, it is desirable to form a matrix by a predetermined number, and when the data to be matrixed becomes half-finished, a vector whose values are all 0 may be added to the missing part. This makes it possible to perform processing that is simple and does not cause adverse effects.
Further, the optimization process (inspection) module 155 may perform the normalization process on the inspection data image-processed by the molding module 150.
The process by the optimization process (inspection) module 155 is the same process as the optimization process (reference) module 115 described above, and the reference data may be replaced with the inspection data in the description. It should be noted that, depending on where the function of collecting the normalized vector data into a matrix is provided, whether the reference data is normalized or the inspection data is normalized differs.
In this optimization processing module 155, collation module 160, or target part information storage module 130, a plurality of data (or data obtained by optimizing this data) created by the molding module 150 based on the inspection data are collected. Processing to form a matrix may be performed, and correspondence information between each matrix and position information in the matrix may be stored. This is because it is necessary when outputting information on how to form and match the registered reference image with success.
As described above, the optimization process (inspection) module 155 does not perform the normalization process, and calculates the average value of the elements in the inspection data constituted by a plurality of elements as the value of the element. You may make it produce | generate as a certain test | inspection data.

ただし、Ｆは基準画像（基準データの集合の一例）、ｆ_ｉは基準画像の個々の画素の画素値（白黒画像の明度値、特定色の明度値や彩度等を含む）、Ｌ（Ｌ＝Ｍ×Ｎ）は基準画像の総画素数、Ｇは検査画像（検査データの一例）の部分領域、ｇ_ｉは検査画像の部分領域の個々の画素の明度値、ｆ_ＡＶＥは基準画像の個々の画素の明度値の平均値、ｇ_ＡＶＥは検査画像の部分領域の個々の画素の明度値の平均値である。複数の照合画像を演算対象の検査画像として上記の演算を各々行うことで、基準画像のドット数をｍ×ｎ、照合画像のドット数をＭ×Ｎとすると、単一の検査画像当たり（Ｍ―ｍ＋１）×（Ｎ−ｎ＋１）個の相関値が得られる。 The verification module 160 is connected to the reference data database 125, the optimization processing (inspection) module 155, and the output module 165. The collation module 160 collates each reference data after normalization stored in the reference data storage module 135 with the inspection data after normalization by the optimization processing (inspection) module 155.
In addition, the normalized reference data and the normalized inspection data are each a vector, and a plurality of normalized reference data are collected into a matrix and converted from the inspection data by image processing and normalized. Is a matrix in which a plurality of are collected.
Then, the collation module 160 collectively calculates correlation values between the plurality of reference data and the image processed inspection image. Note that the collation module 160 obtains a result equivalent to the calculation result of Expression 1 without executing the calculation of Expression 1 itself. That is, since it has been normalized first, the calculated value of Equation 1 can be obtained simply by calculating the inner product of the vectors. For this reason, it is possible to collectively perform collation calculation between a plurality of pieces of data image-processed from a plurality of reference data and inspection data only by performing multiplication between matrices. That is, the calculation of Expression 1 is realized by normalization processing by the optimization processing (inspection) module 155 and computation (more specifically, multiplication) between the matrices performed by the matching module 160.

Here, F is a reference image (an example of a set of reference data), f _i is a pixel value of each pixel of the reference image (including a lightness value of a black and white image, a lightness value and a saturation of a specific color), and L (L = M × N) is the total number of pixels of the reference image, G is a partial area of the inspection image (an example of inspection data), g _i is the brightness value of each pixel of the partial area of the inspection image, and f _AVE is an individual of the reference image G _AVE is the average value of the brightness values of the individual pixels in the partial area of the inspection image. By performing each of the above calculations using a plurality of verification images as inspection images to be calculated, assuming that the number of dots in the reference image is m × n and the number of dots in the verification image is M × N, (M −m + 1) × (N−n + 1) correlation values are obtained.

Then, the collation module 160 compares the collation result with a predetermined threshold value, aggregates the collation results of the plurality of reference data and the inspection data, and determines the authenticity of the inspection data.
At least the processing by the verification module 160 is performed in parallel processing. In addition, when the load of image processing increases, it is desirable that the processing of the optimization processing (inspection) module 155 be parallel processing.
Further, the processing by the collation module 160 may be performed by a mapper in parallel calculation. Specifically, the correlation value calculation itself may be performed by a mapper in the case of Hadoop. Further, it may be performed by both a mapper and a reducer (a combiner may be inserted between the mapper and the reducer). Specifically, both of them may be performed step by step. As a stepwise process, for example, a mapper performs a process of determining matching based on the calculated correlation value, information when matching is successful (information indicating which reference image is successfully matched, The process of generating information indicating which conversion has been applied and the correct answer for collation may be performed by the reducer. In addition, when using Spark as a distributed processing system, an executor (executor (or a worker in terms of expression)) to be processed in parallel and a portion to be collected (a portion that is executed by a reducer) (Driver).
The above-mentioned “information indicating which reference image is successfully collated” will be described in detail. When a plurality of inspection-side images are collectively input into the information processing apparatus 100, the processing is processing for generating information indicating information indicating which inspection-side image and which reference image are successfully collated. Further, as auxiliary information, it is desirable to generate information indicating how the collation is successful when the image on the inspection side is converted.
Further, as “information when collation is successful”, “information indicating which image on the inspection side is successfully collated” may be included. Here, when a plurality of inspection-side images are input to the information processing apparatus 100, which inspection-side image among them has been successfully verified.
In addition, as “information when the collation is successful”, there are “inspection data created by image processing (rotation, movement, etc.) on the image on the inspection side, and which inspection data after the image processing and the reference side "Information indicating whether or not the collation has succeeded with the other data" may be included. Therefore, the information is also information indicating what kind of image processing (rotation angle, moving direction, moving distance, etc.) has been performed on the image on the inspection side.
The processing of the collation module 160 will be described later using the flowchart shown in the example of FIG.

  The output module 165 is connected to the verification module 160. The output module 165 outputs the determination result by the collation module 160. To output the determination result is, for example, printing on a printing device such as a printer, displaying on a display device such as a display, writing to a storage device such as a determination result database, or storing it in a storage medium such as a memory card. , Passing to another information processing apparatus, and the like.

FIG. 2 is an explanatory diagram illustrating an example of true / false determination processing. Whether the inspection image 210 is true (the reference image 200 is included in the inspection image 210) or false (the reference image 200 is not included in the inspection image 210) with respect to the reference image 200 Judgment. Note that “included” in “the reference image 200 is included in the inspection image 210” does not need to be completely identical. That is, strictly speaking, since the shooting is different, it is almost impossible to include the pixel value sequence of the reference image. This means that a portion corresponding to the reference image is included.
It is assumed that the processing result by the reference data processing module 105 is registered in the reference data storage module 135 in advance for the reference image 200 shown in the example of FIG. The processing result of the reference image processing module 105 for the reference image 200 is a vector (one-dimensional vector) in which pixel values (in the example of FIG. 2, “pixel value−average value of pixel values”) are arranged in one column.
The inspection data reception module 145 receives the inspection image 210 shown in the example of FIG. Then, a cutout image 220 shown in the example of FIG. 2 (b2) is cut out from the inspection image 210. Note that this cut-out process is one of processing examples of the molding module 150. In addition to the above, the processing described above is applied to processing for cutting out a rotated object, processing for extracting from an enlarged / reduced object, processing for extracting from a rotated / enlarged / reduced object, processing for adding noise to a pixel value, etc. There are various molding processes, such as the process of performing. Then, the optimization process (search) module 155 performs a process of subtracting the average value from the formed pixel vector. Moreover, you may perform the process which optimizes length. As a result of processing by the shaping module 150 and the optimization processing (inspection) module 155 of the cutout image 220, a pixel value (in the example of FIG. 2, “pixel value−average value of pixel values”) is a vector (primary). Original vector).
The collation module 160 calculates a correlation coefficient between the reference image 200 and the cutout image 220.
The example of FIG. 2C shows a case where the position where the partial image is cut out by the molding module 150 is changed in the X and Y directions. Then, for example, if the value of the highest correlation coefficient (vertex 230 in the example of FIG. 2) is higher than a predetermined threshold value, it is determined to be true, and otherwise, it is determined to be false (贋).
At this time, if the difference between the highest correlation coefficient value and other values, for example, the one with the second highest value (which can be determined by the difference or ratio) does not satisfy the predetermined condition, the determination is true. It may not be.
When determining the conditions here, the determination may be made with reference to information specified at the time of molding (for example, the rotation angle and the ratio of enlargement / reduction). For example, if the rotation angle (rotation increment) is made fine, or the enlargement / reduction ratio (magnification) is made fine, the collation may be successful multiple times near the setting. Even in this case, the designation of deformation at the time of molding may be referred to in the verification result determination process so that the verification can be determined to be successful.

FIG. 3 is an explanatory diagram illustrating an example of parallel processing for authenticity determination.
In the registered image DB 300, a plurality of reference images 200 (four in the example of FIG. 2) are stored. Then, the authenticity of the inspection image 210 is determined.
In the example of FIG. 3, the collation module 160 is further subjected to distributed parallel processing on the part that performs correlation calculation. At this time, the inspection data receiving module 145, the molding module 150, etc. need not be plural. In the plurality of collation modules 160, the inspection image 210 and the reference image 200a are collated, the inspection image 210 and the reference image 200b are collated, the inspection image 210 and the reference image 200c are collated, and the inspection image 210 and the reference image 200d are collated. Is done.

However, the inspection image 210 needs to be cut out. In other words, in the example of FIG. 3, the reference image 200a, the reference image 200b, and the reference image 200c are registered information about different objects. For example, in the inspection data processing module 140a, the cut image 220a and the reference image 200a are collated, the cut image 220b and the reference image 200a are collated, and the cut image 220c and the reference image 200a are collated. . That is, correlation coefficient calculation 310a, correlation coefficient calculation 310b, and correlation coefficient calculation 310c are performed. Then, a correlation coefficient 320 of pixel values is generated from the correlation coefficient calculation 310a, the correlation coefficient calculation 310b, and the correlation coefficient calculation 310c, and a maximum value (Max) 330, an average value (Mean) 340, a standard deviation ( σ) 350 is calculated, and the following determination is made.
(“Maximum value (Max) 330” ≧ threshold A (ncc_threshold))) and (((“maximum value (Max) 330” − “mean value (Mean) 340”) / “standard deviation (σ) 350”)) If it is ≧ threshold value B (nscore_threshold)), it is determined as a determination result (true) 390a, and in other cases, it is determined as a determination result (贋) 390b.
The determination result of each inspection data processing module 140 is acquired, and the reference image 200 that has become the determination result (true) 390a is compared with the inspection image 210 (true). At this time, when conversion such as rotation and enlargement / reduction is also performed, the same determination may be made. Further, the determination may be made with reference to the correlation value, the type of conversion, and the parameter value.

FIG. 4 is a flowchart showing an example of processing according to this embodiment.
In step S402, the registration data reception module 110 reads registration data 450 that is reference data.
In step S404, the optimization process (reference) module 115 performs the optimization process. Detailed processing in step S404 will be described later using the flowchart shown in the example of FIG.
In step S406, the registration module 120 registers the reference data information in the target part information storage module 130 and the reference data storage module 135. Detailed processing in step S406 will be described later using the flowchart shown in the example of FIG.
In step S408, it is determined whether or not the processing for all registered data 450 has been completed. If completed, the processing is terminated (step S499). Otherwise, the process returns to step Sstep S402.

FIG. 5 is a flowchart showing an example of processing according to the present embodiment.
In step S502, target elements are arranged. Here, an example of an image will be described.
Information in an area of a predetermined size (for example, 32 × 32 pixels, 64 × 64 pixels, etc.) is cut out.
Then, information of each element (pixel) is stored in a predetermined value range. For example, pixels outside the range may be excluded from the processing target. That is, information outside the range may be deleted.
Also, for example, conversion is performed so that the value of each element (pixel) falls within the range. The matching algorithm used in the matching module 160 is a process that matches the case where the pixel value changes gently. More specifically, a sigmoid function or a hyperbolic function tanh may be used.

  In step S504, normalization is performed. In this normalization, the average value of the target elements is obtained, the average value is subtracted from the value of each element, and the length of the vector is set to 1 as if it is a multidimensional vector in which the values of each element are arranged. Convert the value to A general normalization process may be used. At this time, the value of the pixel serving as the non-target element may be set to zero. This is a corresponding method for efficiently executing the calculation of Equation 1 as normalization of vectors and inner product calculation between normalized vectors.

FIG. 6 is a flowchart showing a processing example (reference data aggregation processing example) according to this embodiment.
In step S602, reference data is received.
In step S604, an aggregation process is performed. That is, the received reference data is a vector, and the reference data is grouped into a predetermined number (the “number” here may be plural or variable). A matrix is formed (formation processing in units of reference data processing). In the case of a configuration using non-target element information, the non-target element information is also formed in a matrix.
In step S606, the processed reference data is stored.

FIG. 7 is a flowchart showing an example of processing according to the present embodiment.
In step S702, the inspection data reception module 145 receives inspection data.
In step S704, a molding process corresponding to translation, rotation, enlargement / reduction, or the like is performed on the inspection data. Detailed processing in step S704 will be described later using the flowchart shown in the example of FIG.
In step S706, optimization processing is performed on each processing result obtained in step S704. The detailed processing in step S706 has been described above using the flowchart shown in the example of FIG.

In step S708, inspection data processing unit formation processing is performed. Specifically, using each optimization processing result in step S706 as a vector, matrix information obtained by collecting them is formed (formation processing in units of inspection data processing).
That is, the image information after the parallel movement, rotation, and enlargement / reduction after the optimization process is set as each vector information, and is grouped into a matrix for each predetermined number to be an inspection data processing unit. There may be a plurality of inspection data processing units for one inspection data. A group of a plurality of inspection data processing units corresponding to one inspection data is defined as an inspection data processing unit column.

In step S710, collation processing is performed. Detailed processing in step S710 will be described later using the flowchart shown in the example of FIG.
In step S712, the result is output. For example, the collation processing result in step S710 is output to a display device such as a display or a printer, an information storage device such as a DB, and the processing is terminated.

FIG. 8 is a flowchart showing an example of processing according to this embodiment.
In step S802, a translation image is cut out.
In step S804, a distortion image is cut out.
In step S806, the enlarged / reduced image is cut out.
In step S808, the rotated image is cut out.
In step S810, a defocused image is cut out. Here, “cutting out” includes making a processed image and cutting it out, or cutting out a blurred image while calculating the original pixel value.
Note that any one of the processes in steps S802 to S810 may be performed, or a plurality of processes may be performed.

The information (partial image) to be subjected to the collation process is cut out from the inspection data (for example, an image). The cut-out position, etc., depends on a combination of predetermined processing such as translation, rotation, enlargement / reduction, and parameters. Thus, the cut out information (partial image) is generated.
This process may be a combination of processes. Therefore, although the processing load may increase, a high-speed calculation mechanism (sparse matrix) is prepared in advance by setting a matrix (sparse matrix) that designates a position to be cut out from the current image as a vector. GPU or the like) can also be used, and the processing flow corresponding to each output image is highly independent, so that this part and the next optimization processing can be combined to form distributed parallel processing. However, as described above, there may be a case where the load of this portion does not need to be subjected to distributed parallel processing, so distributed processing is not necessarily required.
In step S812, the result is output. That is, it passes to the process of step S706.

FIG. 9 is a flowchart showing a processing example (collation processing example) according to this embodiment. Here, it is an example when enormous reference data is registered, and a processing flow in distributed processing will be described.
In step S902, an inspection data processing unit (column) is received.
In step S904, information on the reference data processing unit group (for example, an ID string of the reference data processing unit group) is acquired.
In step S906, a set of inspection data processing units and reference data processing units is configured (a job for distributed processing or the like is designed). That is, a job sequence that is a combination of both (inspection data processing unit and reference data processing unit) is created.

  In step S908, collation job execution processing is performed. That is, the collation job is executed by putting the job columns together or by submitting them to the distributed processing for each partial column. Detailed processing in step S908 will be described later using the flowchart shown in the example of FIG.

In step S910, an aggregation process (collation job result aggregation process) is performed on the processing results of each verification job.
A collation job output corresponding to one inspection data is collected, and a collation job output element whose collation calculation value exceeds a predetermined threshold (the collation job output is a vector or a matrix, and the collation job output element is Element or matrix element value).
When there are a plurality of inspection data processing units, it is necessary to perform a process for collecting collation job outputs for each reference data.
The following processing is performed for each reference data.
If no threshold is exceeded, information indicating that is stored in correspondence with the identifier of the reference data.
If there is one that exceeds the threshold, information indicating that is stored in correspondence with the identifier of the reference data.
For example, if a plurality of reference data exceeds a threshold value, and a collation calculation value corresponding to parallel movement that periodically exceeds the threshold value is output, it is output that a match is not recognized. Also good.
If a plurality of pieces of reference data exceed a threshold value, the correct collation (true) may be made if the parallel movement distance is within a predetermined threshold value.
If a plurality of reference data exceed a threshold value, the correct collation (true) may be made if the difference between the rotation and enlargement / reduction parameters is within a predetermined threshold value.

The following processing is performed for all reference data.
If there is only one item that has been successfully verified, the identifier of the corresponding reference data is output as a successful verification.
-If there is no successful verification, information indicating that there is no successful verification is output.
If the verification is successful for a plurality of reference data, the corresponding information is output by a predetermined method. For example, as output information, there may be various variations such as an error or output of a plurality of verification success information.
In step S912, the result is output. That is, it passes to the process of step S712.

FIG. 10 is a flowchart showing an example of processing (example of collation job execution processing) according to this embodiment.
In step S1002, job information is received.
In step S1004, the reference data processing unit and the inspection data processing unit are received.
In step S1006, a collation value is calculated (matrix multiplication processing, tensor multiplication processing).
In step S1008, the result is output. That is, it passes to the process of step S910.

The computer that has received the job information (Mapper in the case of Hadoop, Executor in the case of Spark) obtains the reference data processing unit (when it is bundled with the job information or based on the ID etc., the storage location of the DB, etc. And the inspection data processing unit is obtained (may be included in the job information or obtained from a predetermined storage location).
Since the obtained reference data processing unit and inspection data processing unit are formed into matrix information that can be calculated at high speed by a GPU or the like, matrix multiplication calculation is directly executed by the GPU or the like. Note that when the reference data processing unit and the inspection data processing unit are grouped as columns, tensor calculation is performed.
A processing result is obtained as vector information or matrix information, which is output. In the case of Hadoop, transmission is directed to the reducer. In the case of Spark, the transmission is directed to the Driver program.

FIG. 11 is an explanatory diagram showing a system configuration example according to the present embodiment. An arrangement example of each processing apparatus for the collation job result aggregation process will be described. Here, an example of Hadoop 1120 is used as a parallel processing environment.
An HDFS (Hadoop Distributed File System) 1110 stores a reference image in advance.
The client 1100 transmits the inspection image 210 to the HDFS 1110. Here, when the inspection image is transmitted, each mapper is configured to perform translation, normalization, and the like. If the client 1100 converts and registers the inspection data processing units (columns), there is no need to perform multiple inspection image parallel movement processing on the Mapper side. Further, the step of creating the inspection data processing unit (column) itself may be separately performed by Mapper by distributed processing.
Each Mapper 1130 in the Hadoop 1120 uses each processing device 1140 to collate the inspection image 210 with each reference image 200. Specifically, the correlation coefficient is calculated by performing matrix multiplication processing with the GPU 1142 in each processing device 1140. That is, the inner product calculation of the normalized vector is performed, and the linear algebra calculation library on the GPU 1142 is used.
Then, the collation job result aggregation process is executed by the Reducer 1150 and the process result is transmitted to the client 1100.

Further, the collation job result aggregation process may be partially arranged in the Maper 1130.
In this case, for each unit of job processed by each Mapper 1130, the “for each reference data” portion of the collation job result aggregation processing is executed for each reference data in the distributed job, and the processing result is transmitted to the Reducer 1150. .
Further, the collation job result aggregation process may be newly configured as a distributed processing job, and a part of “every reference data” or “all reference data” may be distributed and processed by the Mapper 1130. For most reference images, matching should fail. Therefore, the result data may not be transmitted when collation fails, or information in which the collation failure results are aggregated may be transmitted. As a result, the load of the collation job result aggregation process can be reduced as compared with the case where this process is not performed.

  A hardware configuration example of the information processing apparatus according to the present embodiment will be described with reference to FIG. The configuration shown in FIG. 12 is configured by a personal computer (PC), for example, and shows a hardware configuration example including a data reading unit 1217 such as a scanner and a data output unit 1218 such as a printer.

  The CPU (Central Processing Unit) 1201 (or GPU) includes various modules described in the above-described embodiments, that is, the reference data processing module 105, the registration data receiving module 110, the optimization processing (reference) module 115, and the registration module. 120, an inspection data processing module 140, an inspection data receiving module 145, a molding module 150, an optimization processing (inspection) module 155, a collation module 160, an output module 165, and the like. It is a control part which performs a process.

  A ROM (Read Only Memory) 1202 stores programs used by the CPU 1201, calculation parameters, and the like. A RAM (Random Access Memory) 1203 stores programs used in the execution of the CPU 1201, parameters that change as appropriate during the execution, and the like. These are connected to each other by a host bus 1204 configured by a CPU bus or the like.

  The host bus 1204 is connected to an external bus 1206 such as a PCI (Peripheral Component Interconnect / Interface) bus via a bridge 1205.

  A keyboard 1208 and a pointing device 1209 such as a mouse are devices operated by an operator. The display 1210 includes a liquid crystal display device or a CRT (Cathode Ray Tube), and displays various types of information as text or image information. Further, a touch screen or the like having both functions of the pointing device 1209 and the display 1210 may be used.

  An HDD (Hard Disk Drive) 1211 includes a hard disk (may be a flash memory or the like), drives the hard disk, and records or reproduces a program executed by the CPU 1201 and information. The hard disk realizes functions as the target part information storage module 130, the reference data storage module 135, and the like. Furthermore, other various data (including inspection data, verification results, etc.), various computer programs, and the like are stored.

  The drive 1212 reads data or a program recorded in a removable recording medium 1213 such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, and the data or program is read out to the interface 1207 and the external bus 1206. , The bridge 1205, and the RAM 1203 connected via the host bus 1204. Note that the removable recording medium 1213 can also be used as a data recording area.

  The connection port 1214 is a port for connecting the external connection device 1215 and has a connection unit such as USB, IEEE1394. The connection port 1214 is connected to the CPU 1201 and the like via the interface 1207, the external bus 1206, the bridge 1205, the host bus 1204, and the like. The communication unit 1216 is connected to a communication line and executes data communication processing with the outside. The data reading unit 1217 is a scanner, for example, and executes document reading processing. The data output unit 1218 is a printer, for example, and executes document data output processing.

  Note that the hardware configuration of the information processing apparatus illustrated in FIG. 12 illustrates one configuration example, and the present embodiment is not limited to the configuration illustrated in FIG. 12, and the modules described in the present embodiment are executed. Any configuration is possible. For example, some modules may be configured with dedicated hardware (for example, Application Specific Integrated Circuit (ASIC), etc.), and some modules are in an external system and connected via a communication line Further, a plurality of systems shown in FIG. 12 may be connected to each other via communication lines so as to cooperate with each other. In particular, in addition to personal computers, portable information communication devices (including mobile phones, smartphones, mobile devices, wearable computers, etc.), information appliances, robots, copiers, fax machines, scanners, printers, multifunction devices (scanners, printers, An image processing apparatus having two or more functions such as a copying machine and a fax machine) may be incorporated.

The program described above may be provided by being stored in a recording medium, or the program may be provided by communication means. In that case, for example, the above-described program may be regarded as an invention of a “computer-readable recording medium recording the program”.
The “computer-readable recording medium on which a program is recorded” refers to a computer-readable recording medium on which a program is recorded, which is used for program installation, execution, program distribution, and the like.
The recording medium is, for example, a digital versatile disc (DVD), which is a standard established by the DVD Forum, such as “DVD-R, DVD-RW, DVD-RAM,” and DVD + RW. Standard “DVD + R, DVD + RW, etc.”, compact disc (CD), read-only memory (CD-ROM), CD recordable (CD-R), CD rewritable (CD-RW), Blu-ray disc ( Blu-ray (registered trademark) Disc), magneto-optical disk (MO), flexible disk (FD), magnetic tape, hard disk, read-only memory (ROM), electrically erasable and rewritable read-only memory (EEPROM (registered trademark)) )), Flash memory, Random access memory (RAM) SD (Secure Digital) memory card and the like.
Then, the whole or a part of the program may be recorded on the recording medium for storage or distribution. Also, by communication, for example, a local area network (LAN), a metropolitan area network (MAN), a wide area network (WAN), a wired network used for the Internet, an intranet, an extranet, or a wireless communication It may be transmitted using a transmission medium such as a network or a combination of these, or may be carried on a carrier wave.
Furthermore, the program may be a part or all of another program, or may be recorded on a recording medium together with a separate program. Moreover, it may be divided and recorded on a plurality of recording media. Further, it may be recorded in any manner as long as it can be restored, such as compression or encryption.

DESCRIPTION OF SYMBOLS 100 ... Information processing apparatus 105 ... Reference data processing module 110 ... Registration data reception module 115 ... Optimization process (reference) module 120 ... Registration module 125 ... Reference data database 130 ... Target part information storage module 135 ... Reference data storage module 140 ... Inspection data processing module 145 ... Inspection data receiving module 150 ... Molding module 155 ... Optimization processing (inspection) module 160 ... Verification module 165 ... Output module 1100 ... Client 1110 ... HDFS
1120 ... Hadoop
1130 ... Mapper
1140 ... Processing device 1142 ... GPU etc. 1150 ... Reducer

Claims (13)

The average value of the element values in the reference data constituted by a plurality of elements is subtracted from the value of the element, and a plurality of reference data obtained by normalizing the length of the multidimensional vector constituted by the elements are stored. A reference data storage means,
Normalization means for subtracting an average value of element values in the inspection data constituted by a plurality of elements from the value of the elements, and normalizing the length of a multidimensional vector constituted by the elements;
Collating means for collating each reference data after normalization stored in the reference data storage means and inspection data after normalization by the normalizing means,
Comparing a comparison result by the verification unit with a predetermined threshold, and a determination unit for determining the authenticity of the inspection data by aggregating the verification results of the plurality of reference data and the inspection data,
Performing the normalization means and the collation means in parallel processing;
Information processing device. The normalized reference data and the normalized inspection data are each a vector, and a plurality of normalized reference data are combined into a matrix and converted from the inspection data by image processing and normalized. Is a matrix in which a plurality of
The collating means collectively calculates a correlation value with the image-processed inspection data for a plurality of reference data;
The information processing apparatus according to claim 1. The processing by the determination means is performed by parallel calculation.
The information processing apparatus according to claim 2. The processing by the determination means is performed by a mapper or reducer in parallel calculation, or by an executor or driver,
The information processing apparatus according to claim 3. The reference data and the inspection data are images,
Image processing means for performing image processing including affine transformation on the inspection data;
The normalization means performs a normalization process on the inspection data image-processed by the image processing means;
The information processing apparatus according to any one of claims 2 to 4. As the affine transformation, a conversion matrix for performing conversion processing and partial data extraction processing by rotation at a predetermined angle, enlargement / reduction at a predetermined magnification, or a combination thereof, and executing matrix calculation Create data for calculation of inspection data,
The information processing apparatus according to claim 5. further,
A plurality of data created by converting from inspection data by image processing is normalized, and the normalized individual data is used as a vector to form a matrix and manage position information in the matrix. Management means to
Storage means for storing information associated with the original information from a position in the matrix;
The information processing apparatus according to claim 2, further comprising an extraction unit that extracts original information from a position in the matrix using the correspondence. further,
Second storage means for storing a determination rule according to the position of the matrix based on the correspondence;
The determination means determines authenticity using the determination rule.
The information processing apparatus according to claim 7. further,
In the case where a plurality of inspection data is simultaneously input to the determination unit, information for identifying the original inspection data is notified to the determination unit, or the collation result is determined according to the original inspection data before the processing of the determination unit. The information processing apparatus according to claim 8, further comprising a processing unit that divides the information. The normalization means sets a value that is not a target to 0,
The information processing apparatus according to any one of claims 1 to 9. Reference data storage means for storing an average value of element values in reference data constituted by a plurality of elements, and a plurality of reference data as values of the elements;
Generating means for generating, as inspection data, an average value of element values in inspection data constituted by a plurality of elements;
Collation means for collating each reference data stored in the reference data storage means with inspection data by the generation means;
Comparing a comparison result by the verification unit with a predetermined threshold, and a determination unit for determining the authenticity of the inspection data by aggregating the verification results of the plurality of reference data and the inspection data,
Performing the normalization means and the collation means in parallel processing;
Information processing device. Computer
The average value of the element values in the reference data constituted by a plurality of elements is subtracted from the value of the element, and a plurality of reference data obtained by normalizing the length of the multidimensional vector constituted by the elements are stored. A reference data storage means,
Normalization means for subtracting an average value of element values in the inspection data constituted by a plurality of elements from the value of the elements, and normalizing the length of a multidimensional vector constituted by the elements;
Collating means for collating each reference data after normalization stored in the reference data storage means and inspection data after normalization by the normalizing means,
Comparing the collation result by the collation unit with a predetermined threshold value, causing the collation result of the plurality of reference data and the inspection data to function as a determination unit that determines the authenticity of the inspection data,
Performing the normalization means and the collation means in parallel processing;
Information processing program. Computer
Reference data storage means for storing an average value of element values in reference data constituted by a plurality of elements, and a plurality of reference data as values of the elements;
Generating means for generating, as inspection data, an average value of element values in inspection data constituted by a plurality of elements;
Collation means for collating each reference data stored in the reference data storage means with inspection data by the generation means;
Comparing the collation result by the collation unit with a predetermined threshold value, causing the collation result of the plurality of reference data and the inspection data to function as a determination unit that determines the authenticity of the inspection data,
Performing the normalization means and the collation means in parallel processing;
Information processing program.

Images