JP2021174442A - Individual identification system - Google Patents

Abstract

To improve identification accuracy.SOLUTION: An individual identification system 1 includes an image acquisition processing unit 212, a feature point extraction processing unit 221, a local feature quantity calculation processing unit 222, and a specification processing unit 2612. The individual identification system 1 includes at least two of a visualization apparatus 33 for superimposing feature points on an image to be displayed on a display device, an imaging condition adjustment apparatus 31 having an imaging object adjustment processing unit 312 for adjusting an imaging object condition and an imaging device condition adjustment processing unit 311 for adjusting an imaging device condition, and a cross-check identification adjustment apparatus 32 having a collation result display processing unit 333 configured to execute collation result display processing for collectively displaying collation results of collation processing on the display device 12 using a parameter set.SELECTED DRAWING: Figure 1

Description

An embodiment of the present invention relates to an individual identification system capable of identifying an individual based on an individual pattern appearing on the surface of the individual, a so-called individual fingerprint.

In recent years, for example, industrial products are required to have high traceability not only in the distribution process after production but also in the production process. For example, even parts of the same type produced in the same process and in the same lot may have slight differences in their performance to the extent that there is no problem in normal use. This is because the performance of the product can be improved if the product can be manufactured in the optimum combination in consideration of such a slight difference in the characteristics of each part in the production process of the product. ..

Further, in the conventional production process, since the processing conditions of parts and the inspection results are not linked, it is difficult to reflect the inspection results in the processing conditions and the like. On the other hand, if it is possible to track each part in the production process, the inspection result can be appropriately reflected in the processing conditions and the like, and as a result, defects can be reduced and quality / performance can be further improved. ..

Conventionally, as a method for tracking and managing products and the like, for example, there is a method using a two-dimensional code or RFID. However, two-dimensional codes and RFID are expensive to print and paste. Therefore, it is economically difficult to assign a two-dimensional code or RFID to each of a large number of parts constituting a product.

Therefore, attention is being paid to a technique for identifying an individual by using the surface shape unique to each individual. With such a surface shape or pattern of an object, it is not necessary to add a two-dimensional code, RFID, or the like, so tracking management of each part can be realized at low cost.

Japanese Unexamined Patent Publication No. 2012-103758 Japanese Unexamined Patent Publication No. 2012-181566

However, identifying an individual using the surface shape or pattern of an object has the following multiple problems. First, as the first issue, the parts that make up an industrial product are affected by heat or change over time in the manufacturing process, for example, and the pattern on the surface of the parts to be identified is displayed at the time of registration and at the time of identification. Color etc. may change. Then, the characteristics of the surface pattern extracted at the time of registering the image and at the time of identification may be significantly changed, and as a result, the identification accuracy is lowered.

Next, as a second issue, in the factory, dust, lighting in the factory, natural light such as outside light, ambient noise sources, etc. become factors of disturbance, and the registered image is taken and the identification image is taken. The environment can change significantly. Then, the extracted feature points, local feature amounts, and the like may change, and the discrimination performance may deteriorate.

As a third problem, unlike human faces and fingerprints, the characteristics of patterns appearing on the surface of industrial product parts vary depending on the type, material, and processing method of the object to be collated and identified. Therefore, it is necessary to adjust the parameters of the algorithm used for collation identification to appropriate ones depending on the type of the object to be identified. However, there are many parameter items, and it takes a lot of time and effort for the user to manually search for the optimum combination from all the combinations.

The present invention has been made in view of the above problems, and an object of the present invention is to provide an individual identification system capable of solving at least two of the first to third problems at the same time.

The individual identification system (1) of the embodiment includes an image acquisition processing unit (212) capable of performing an image acquisition process for acquiring an image of a subject photographed by a photographing device as an identification image, and the identification acquired by the image acquisition process. A feature point extraction processing unit (221) capable of executing a feature point extraction process for extracting feature points included in an image, and a local feature amount calculation for calculating a local feature amount of the feature points extracted by the feature point extraction process. The local feature amount calculation processing unit (222) capable of performing the process compares the local feature amount related to the plurality of registered images registered in advance with the local feature amount of the identification image, and compares the local feature amount of the plurality of registered images. It is provided with an individual identification device having a specific processing unit (2612) capable of executing a specific process for specifying a registered image having the local feature amount having the highest correlation with the local feature amount of the identification image. This is an individual identification system (1).

In the individual identification system (1), in addition to the individual identification device, a feature point display process capable of executing a feature point display process of superimposing the feature points extracted by the feature point extraction process on the image and displaying them on the display device. The visualization device (33) having the unit (331), or the local feature amount in the registered image obtained by photographing the same subject among a plurality of registered images registered in advance and the identification image is compared, and the feature points of the feature points are compared. It is possible to cause the individual identification device to execute a collation process of acquiring the corresponding points and collating the registered image with the identification image, and to execute a shooting target condition adjustment process for adjusting the shooting target condition which is a condition related to the shooting target. A shooting condition adjusting device having a shooting target condition adjustment processing unit (312) and a shooting device condition adjustment processing unit (311) capable of executing a shooting device condition adjustment process for adjusting the shooting device condition which is a condition related to the shooting device. (31) Or, with respect to the parameter for setting the condition of the collation process, a parameter for generating a plurality of parameter sets by combining each of the parameters set in the set value within the setting range defined for each of the parameters. It has a parameter set generation processing unit (324) capable of executing a set generation process, and a parameter set setting processing unit (325) capable of executing a parameter set setting process for setting the parameter set based on a user operation. , The collation processing unit is made to execute the process of executing the collation process using each of the parameter sets generated in the parameter set generation process, and the collation result of the collation process is summarized using each of the parameter sets. The collation identification condition adjusting device (32) further having a collation result display processing unit (333) capable of executing the collation result display process to be displayed on the display device (12) is provided with at least two devices. ..

A block diagram showing an example of a schematic configuration of an individual identification system according to an embodiment. A block diagram showing an example of a schematic configuration of an imaging device included in an individual identification system according to an embodiment. A block diagram showing an example of a schematic configuration of a feature extraction device included in an individual identification system according to an embodiment. A block diagram showing an example of a schematic configuration of a registration processing device included in an individual identification system according to an embodiment. A block diagram showing an example of a schematic configuration of a recording device included in an individual identification system according to an embodiment. A block diagram showing an example of a schematic configuration of a dictionary device included in an individual identification system according to an embodiment. A block diagram showing an example of a schematic configuration of an image collation identification device included in an individual identification system according to an embodiment. A block diagram showing an example of a schematic configuration of an imaging condition adjusting device included in an individual identification system according to an embodiment. A block diagram showing an example of a schematic configuration of a collation identification condition adjusting device included in an individual identification system according to an embodiment. A block diagram showing an example of a schematic configuration of a visualization device included in an individual identification system according to an embodiment. The figure which shows an example of the execution result of the feature point extraction process in the individual identification system by one Embodiment (the 1). The figure which shows an example of the execution result of the feature point extraction process in the individual identification system by one Embodiment (the 2). A diagram conceptually showing an example of a method of calculating a global feature from a local feature in an individual identification system according to an embodiment. A diagram conceptually showing an example of a method for generating a global feature dictionary in an individual identification system according to an embodiment. A diagram conceptually showing an example of narrowing down processing using global features in an individual identification system according to an embodiment. A diagram conceptually showing an example of specific processing using local features in an individual identification system according to an embodiment. The figure which shows an example of the parameter set in the individual identification system by one Embodiment A diagram conceptually showing an example of a positive correspondence score and a negative correspondence score in an individual identification system according to an embodiment. The figure which shows an example of the combination of the conditions generated by the brute force condition generation process in the individual identification system by one Embodiment. The figure which shows an example of the graph generated by the optimum condition presentation processing in the individual identification system by one Embodiment. The figure which shows an example of the display content displayed on the display device by the feature point display process in the individual identification system by one Embodiment. The figure which shows an example of the display content displayed on the display device by the collation result display processing in the individual identification system by one Embodiment (the 1). The figure which shows an example of the display content displayed on the display device by the collation result display processing in the individual identification system by one Embodiment (the 2).

Hereinafter, one embodiment will be described with reference to the drawings. The arrows in each block diagram indicate the data flow. Each processing unit, that is, the functional block described in the present embodiment may be configured by different hardware, or may be configured to share a plurality of devices or processing units by one hardware. Although details are not shown, the hardware constituting each device and each processing unit is mainly composed of a microcomputer having a storage area such as a CPU, a ROM, a RAM, and a rewritable flash memory, respectively. be able to. The storage area stores an individual identification program for realizing each device and each processing unit. Then, by executing the individual identification program in the CPU, each process in each device and each processing unit can be realized.

That is, each processing unit described in the present embodiment executes the computer program stored in the non-transitional substantive storage medium such as the storage area described above by the CPU to execute the processing corresponding to the computer program. It is realized, that is, it is realized by software. In addition, at least a part of each processing unit may be realized by hardware.

The individual identification system 1 of the present embodiment compares a plurality of registered images registered in advance with an identification image taken by a photographing device, and identifies a registered image in which the subject of the identification image is shown from among the plurality of registered images. It is a system that can do this. This individual identification system 1 can be used, for example, for tracing parts in factories and the like, but is not limited thereto.

As shown in FIG. 1, the individual identification system 1 includes a photographing device 10, an input device 11, a display device 12, an individual identification device 20, and an adjustment device 30. In the case of the present embodiment, the individual identification system 1 includes a registration-side photographing device 10 and an identification-side photographing device 10. In the following description, when distinguishing between them, the registration-side photographing device 10 may be referred to as a registration-side photographing device 10a, and the identification-side photographing device 10 may be referred to as an identification-side photographing device 10b. In the present embodiment, the registration side photographing device 10a acquires an image for registration in the individual identification device 20. Further, the identification side photographing device 10b is for photographing a subject to be identified. Then, in the present embodiment, the registered image mainly means an image taken by the photographing device 10a on the registration side and registered or registered in the individual identification device 20. Further, the identification image mainly means an image taken by the photographing device 10b on the identification side.

The photographing device 10 includes a camera 101, an illumination 102, and a stage 103. The camera 101 has a function of photographing a subject such as a work to be photographed. The camera 101 is configured so that the shutter speed, the sensitivity of the image sensor, the resolution, the shooting range, the focus, and the like can be changed by an instruction from the outside. The illumination 102 is for illuminating a subject, and is configured so that the brightness, luminosity, color, and the like can be changed by an instruction from the outside. The stage 103 supports the camera 101 and the illumination 102, and the subject is placed on the stage 103. The stage 103 is configured so that the distance, angle, that is, the posture of the camera 101 or the illumination 102 and the subject can be changed.

The input device 11 is an interface that accepts user operations, such as a keyboard and a mouse. The user can operate the input device 11 to make various settings for the adjusting device 30 and the individual identification device 20. The display device 12 is a display capable of displaying characters, images, and the like, and can display information received from the individual identification device 20 and the adjustment device 30. By looking at the information displayed on the display device 12, the user can confirm the result of processing by the individual identification device 20 and the adjustment device 30, the setting contents, and the like.

The individual identification device 20 includes a photographing device 21, a feature extraction device 22, a registration processing device 23, a recording device 24, a dictionary device 25, an image matching identification device 26, a photographing condition imparting device 27, and a feature extraction condition imparting device 28. ing. In the case of the present embodiment, the photographing device 21, the feature extraction device 22, the registration processing device 23, the recording device 24, the photographing condition imparting device 27, and the feature extraction condition imparting device 28 are provided on the registration side and the identification side, respectively. There is. In this case, the device surrounded by the broken line a is the device on the registration side, and the device surrounded by the broken line b is the device on the identification side. The device a on the registration side and the device b on the identification side may be configured with different hardware or may be configured with the same hardware. Further, the individual identification device 20 and the adjustment device 30 may be installed on a cloud server connected to a telecommunication line such as the Internet.

The photographing device 21 has a function of controlling the drive of the photographing device 10 and photographing and acquiring an image of the subject. As shown in FIG. 2, the photographing device 21 has a photographing processing unit 211 and an image acquisition processing unit 212. The shooting processing unit 211 drives the shooting device 10 under predetermined conditions to shoot a subject. In this case, the predetermined conditions for photographing the photographing device 10 are referred to as photographing conditions. The shooting conditions include, for example, a shooting device condition which is a condition related to the shooting device 10 and a shooting target condition which is a condition related to a subject to be shot.

The shooting device conditions are mainly related to optical devices, for example, the shutter speed of the camera 101 of the shooting device 10, the sensitivity of the image sensor, the resolution, the shooting range, the focus, the magnification, the aperture, the type and brightness of the illumination 102, and the like. There are brightness, color, etc. The shooting target condition is a condition relating to a shooting target, that is, a subject, and includes, for example, the posture of the subject, that is, the angle of the subject with respect to the camera 101 and the illumination 102, the distance between the camera 101 and the illumination 102, and the subject.

The shooting processing unit 211 can drive the shooting device 10 by changing the shooting device conditions and the shooting target conditions based on the instruction content received from the shooting condition imparting device 27, and can shoot the subject.

The image acquisition processing unit 212 can execute the image acquisition process. The image acquisition process includes a process of acquiring an image of a subject photographed by the photographing device 10 and passing it to the feature extraction device 22. In the case of the present embodiment, the image acquisition processing unit 212 on the registration side a passes the image taken by the photographing device 10a as a registration image to the feature extraction device 22 on the registration side a. Further, the image acquisition processing unit 212 on the identification side b passes the image captured by the photographing device 10b as an identification image to the feature extraction device 22 on the identification side b. In the present embodiment, among the image acquisition processes, the process of acquiring the registered image performed by the registration side a is referred to as the registration image acquisition process, and the process of acquiring the identification image performed by the identification side b is referred to as the identification image acquisition process. Sometimes. The image acquisition process may include a process of acquiring an image from a means other than the photographing device 10, such as acquiring an image via the Internet or a storage medium.

The feature extraction device 22 has a function of extracting features included in the image acquired from the photographing device 21. The feature extraction device 22 has a function of calculating feature points, local feature amounts, and global feature amounts as image features. As shown in FIG. 3, the feature extraction device 22 includes a feature point extraction processing unit 221, a local feature amount calculation processing unit 222, a local feature amount group classification processing unit 223, and a global feature amount calculation processing unit 224. .. The feature extraction device 22 receives the feature extraction conditions from the feature extraction condition imparting device 28 and executes the processes in the respective processing units 221 to 224. The feature extraction conditions include, for example, the number of stages of the pyramid image when extracting feature points and calculating the local feature amount, and a threshold value for determining corners and curves.

The feature point extraction processing unit 221 can execute the feature point extraction process. As shown in FIG. 11, the feature point extraction process includes a process of extracting one or more feature points P appearing on the surface of the subject S from the image acquired by the image acquisition processing unit 212. In the case of the present embodiment, the feature point extraction process extracts, for example, thousands of feature points from one image. In this case, the feature point extraction processing unit 221 on the registration side a can extract the feature points of the registered image taken by the photographing device 10a on the registration side. Further, the feature point extraction processing unit 221 on the identification side b can extract the feature points of the identification image taken by the photographing device 10b on the identification side.

Here, for example, as shown in FIG. 12, a non-geometric pattern such as a corner or a curve is extracted as a feature point P from the patterns appearing on the surface of the work S, and the pattern is manufactured through the same manufacturing process. It is considered that the individual can be identified even among the same type of parts. This is because, for example, geometric patterns such as straight lines and right angles that appear on the surface of parts are often patterns that appear in common to all parts of that type, and are a feature for identifying individuals among parts. It is hard to become. On the other hand, non-geometric patterns such as corners and curves are often formed by slight differences in conditions during the manufacturing process, and it is considered that many of them are unique to the individual. Therefore, the feature point extraction processing unit 221 of the present embodiment extracts a pattern of a corner or a curve that locally appears on the surface of the subject as the feature point P.

The feature point extraction processing unit 221 of the present embodiment uses the luminance gradient intensity distribution obtained from the image of the subject to extract corners and curves locally appearing on the surface of the subject as feature points. The image has a brightness gradient and its intensity between pixels. In the case of a uniform pattern, the spread of the luminance gradient intensity is small, and in the case of a linear pattern, the luminance gradient intensity spreads in a specific direction. On the other hand, in the case of a pattern with corners and curves, the luminance gradient intensity spreads in multiple directions. Therefore, the feature point extraction processing unit 221 can extract corners and curves that serve as feature points by searching for a pattern in which the luminance gradient intensity spreads in a plurality of directions.

The local feature amount calculation processing unit 222 can execute the local feature amount calculation process. The local feature amount calculation process includes a process of calculating the feature amount of the feature point as a local feature amount based on the brightness gradient of the peripheral region including the feature point extracted by the feature point extraction process. The local feature has the coordinates of the feature point and a 256-bit binary feature. SIFT of the existing technology can be used for the calculation of the feature amount. Further, if possible, another image feature calculation algorithm such as SURF may be used. In the feature point extraction process, a pyramid image obtained by reducing the image in multiple steps may be obtained, and feature points may be independently extracted for each scale image under the same conditions. Then, the local feature amount calculation processing unit may calculate the local feature amount of the feature points obtained in each pyramid image. By doing so, it is possible to obtain a local feature amount that has scale invariance and is not easily affected by the difference in scale.

The local feature group classification processing unit 223 can execute the local feature group classification process. The local feature group classification process is a process that is a preliminary preparation for calculating the global feature amount of the registered image and the identification image. In the local feature group classification process, for example, as shown in FIGS. 13A and 13B, a plurality of local feature amounts A obtained by executing a local feature amount calculation process for each image are used as the local feature amount. It includes a process of classifying into a predetermined number, for example, 64 to 4096 local feature amount groups B according to the value of A. In other words, the local feature group classification process includes a process of clustering a large number of local feature A obtained from each image by collecting those having similar values. Each element Bn constituting the local feature amount group B has at least one pair of a feature point obtained from an image and the local feature amount thereof. Note that n in Bn is an integer indicating the number of elements.

The local feature group classification process includes a process of classifying the local feature A obtained from each image based on the dictionary information d. The dictionary information d is a typical feature amount of the feature points obtained from the learning image acquired in advance, and is prepared before the execution of the local feature amount classification process. The local feature amount group classification processing unit 223 classifies the local feature amount A to be classified by applying it to the dictionary information d having a feature amount close to the local feature amount A.

The global feature amount calculation processing unit 224 can execute the global feature amount calculation processing. The global feature calculation process includes a process of calculating a global feature based on each local feature group B obtained by the local feature separation process. As shown in FIG. 13C, the global feature calculation process calculates the frequency distribution, that is, the histogram obtained by voting for the dictionary information d having a high correlation with each local feature A obtained from the image as the global feature. Including processing. That is, the global feature amount has the local feature amount group B on the horizontal axis and the number of votes voted for each local feature amount group B on the vertical axis. By using the global feature amount calculated in this way, one image can be represented by one feature amount. For the global feature amount calculation process, for example, VLAD or BAG-OF-WORDS can be adopted as the calculation method.

なお、（１）式のＮＮ（ｖ）＝ｄｋは、画像中に含まれる特徴量の中で、この場合６４個ある各辞書情報ｄと最も相関が高い局所特徴量ｖを意味する。 When VLAD is adopted as the calculation method of the global feature amount, the global feature amount calculation process is performed by, for example, calculating the vector V, which is the sum of the difference vectors from the representative value, which is the dictionary information having a high correlation with the local feature amount, as many as the number of representative values. The calculation is performed, and the vector connecting them is calculated as the vector V of the global features. Specifically, the global feature calculation processing unit 224 applies the local feature obtained from the image to the calculation of the following equation (1) to calculate the vector V of the global feature. In this case, the global feature vector V is composed of 64 128-dimensional vectors. In this case, the local feature amount included in each local feature amount group B is v, and the dictionary information is d. Further, assuming that the number of features included in each local feature group B is N, 1 ≦ i ≦ N.

Note that NN (v) = dk in the equation (1) means a local feature amount v having the highest correlation with each of the 64 dictionary information ds in this case among the feature amounts included in the image.

Further, when BAG-OF-WORDS is adopted as the global feature amount, the global feature amount dictionary generation process calculates the global feature amount as follows. In this case, the global feature quantity is a vector having the number of dimensions of the dictionary information, for example, a vector V having 64 dimensions. The global feature amount dictionary generation processing unit 251 searches for the closest dictionary information d for each feature amount using 64 representative dictionary information, and adds 1 to the elements of the vector V that are close to each other. As a result, the vector V becomes a histogram of the features, and the sum of the elements of the vector V has the same number of features. This vector V is used by dividing the whole by this total number and normalizing it.

The registration processing device 23 has a function of recording the data obtained in each processing in the recording device 24. As shown in FIG. 4, the registration processing device 23 includes a search target image registration processing unit 231, an evaluation image registration processing unit 232, a local feature amount registration processing unit 233, and a global feature amount registration processing unit 234. .. The search target image registration processing unit 231 can execute the search target image registration process. The search target image registration process includes a process of registering a plurality of search target images as registered images in advance in the search target image registration unit 241 of the recording device 24.

The evaluation image registration processing unit 232 can execute the evaluation image registration process. The evaluation image registration process includes a process of registering an image used for evaluating the collation process executed by the image collation identification device 26 as an evaluation image in the evaluation image registration unit 242 of the recording device 24. The evaluation image is composed of a registered image group including a plurality of registered images of different subjects, and an identification image of the same subject as one of the registered images and taken under different conditions. Has been done. In this case, there may be a plurality of identification images depending on the number of subjects and conditions. Further, the registered image obtained by photographing the same subject and the identification image are registered as an evaluation image of the recording device 24 in a state of being linked to each other by, for example, assigning the same or partially common identification number or the like. It is registered in the section 242.

The local feature amount registration processing unit 233 can execute the local feature amount registration process. The local feature amount registration process includes a process of registering the local feature amount of the feature point extracted from the registered image by the process of the feature extraction device 22 in the local feature amount registration unit 243 of the recording device 24. Further, the local feature amount registration process includes a process of registering the local feature amount of the feature point extracted from the evaluation image by the process in the feature extraction device 22 in the local feature amount registration unit 243 of the recording device 24. The local feature amount registration processing unit 233 registers the local feature amount obtained from each image in association with the image as the extraction source.

The global feature amount registration processing unit 234 can execute the global feature amount registration process. The global feature amount registration process includes a process of registering the global feature amount related to the registered image in the global feature amount registration unit 244 of the recording device 24. Further, the global feature amount registration process includes a process of registering the global feature amount related to the evaluation image in the global feature amount registration unit 244 of the recording device 24. The local feature amount registration processing unit 233 registers the global feature amount obtained from each image in association with the image as the extraction source.

The recording device 24 has a function of recording various data sent by each registration process of the registration processing device 23, and constitutes a database. The recording device 24 can also be configured by, for example, a computer recording device constituting the individual identification device 20, an external server connected via a network line, or the like. As shown in FIG. 5, the recording device 24 includes a search target image registration unit 241, an evaluation image registration unit 242, a local feature amount registration unit 243, a global feature amount registration unit 244, and a global feature amount dictionary registration unit 245. doing.

The search target image registration unit 241 is a recording area for recording the search target image sent from the search target image registration processing unit 231 of the registration processing device 23. The evaluation image registration unit 242 is a recording area for recording the evaluation image sent from the evaluation image registration processing unit 232 of the registration processing device 23. The local feature amount registration unit 243 is a recording area for recording the local feature amount sent from the local feature amount registration processing unit 233 of the registration processing device 23. The global feature amount registration unit 244 is a recording area for recording the global feature amount sent from the global feature amount registration processing unit 234 of the registration processing device 23.

Further, the global feature dictionary registration unit 245 is a recording area for recording the global feature dictionary sent from the global feature dictionary registration processing unit 252 of the dictionary device 25. Before executing the search process in the image matching identification device 26, a plurality of global feature amount dictionaries having a plurality of dictionary information calculated based on the local feature amount obtained from the learning image acquired in advance are registered. ing.

The dictionary device 25 has a function of generating a plurality of dictionary information based on feature points extracted from a plurality of images and their local feature amounts, and generating, registering, and setting a global feature amount dictionary based on the dictionary information. Has. As shown in FIG. 6, the dictionary device 25 has a global feature amount dictionary generation processing unit 251 and a global feature amount dictionary registration processing unit 252.

The global feature dictionary generation processing unit 251 can execute the global feature dictionary generation processing. The global feature dictionary generation process includes, for example, a process of generating a global feature dictionary having representative values of each of the local feature groups obtained from a plurality of learning images as dictionary information. In this case, the global feature dictionary generation process includes a process of setting the centroid of the local feature group as a representative value in the dictionary information. The dictionary information is an element that constitutes a global feature dictionary, and a plurality of dictionary information is collected to form one global feature dictionary.

In this case, for example, in the global feature dictionary generation process, the global feature dictionary generation processing unit 251 can generate the global feature dictionary as follows. That is, in the global feature dictionary generation process, the global feature dictionary generation processing unit 251 first, as shown in FIGS. 14A and 14B, for example, m images for learning calculated by the feature extraction device 22. Receives local features A of T1 to Tm. Next, as shown in FIG. 14C, the global feature dictionary generation processing unit 251 receives k local features A using, for example, the k-means method, and in this case, 64 clusters, that is, local features. It is classified into groups B1 to Bk, and centroids C1 to Ck are obtained for each of the local feature group B1 to Bk. Then, the global feature dictionary generation processing unit 251 uses the centroids C1 to Ck of the local feature groups B1 to Bk thus obtained as the representative values of the local feature groups B1 to Bk, and the dictionary information d1. Set to ~ dk to generate a global feature dictionary E that summarizes dictionary information d1 to dk. Then, the global feature dictionary generation processing unit 251 generates a plurality of global feature dictionaries E using different learning images T. The dictionary information d1 to dk is not limited to the centroid as long as it is a value representing each local feature group B1 to Bk.

The global feature dictionary registration processing unit 252 can execute the global feature dictionary registration process. The global feature dictionary registration process includes a process of recording a plurality of global feature dictionary Eic generated by the global feature dictionary generation process in the global feature dictionary registration unit 245 of the recording device 24. In this case, the global feature dictionary E includes a plurality of dictionary information d1 to dk constituting the global feature dictionary E.

The image collation identification device 26 has a function of collating the registered image recorded in the recording device 24 with the identification image output from the feature extraction device 22 on the identification side and discriminating the match / mismatch of the images. As shown in FIG. 7, the image collation identification device 26 includes a search processing unit 261, a collation processing unit 262, and a result notification processing unit 263.

The search processing unit 261 can execute the search processing. The search process is performed on the global feature amount related to the identification image sent from the feature extraction device 22 on the identification side among the global feature amounts related to the plurality of registered images recorded in the search target image registration unit 241 of the recording device 24. Includes processing to search for highly correlated items. In the case of the present embodiment, the search processing unit 261 has a narrowing down processing unit 2611 capable of executing the narrowing down process and a specific processing unit 2612 capable of executing the specific processing. That is, in the case of the present embodiment, the search process includes a narrowing process and a specific process.

As shown in FIG. 15, the narrowing down processing executed by the narrowing down processing unit 2611 includes a global feature amount of the identification image from a plurality of registered images registered in the search target image registration unit 241 of the recording device 24. It includes a process of narrowing down a predetermined number of registered images having a global feature amount having a high correlation with the candidates. When the number of registered images registered in the search target image registration unit 241 is, for example, several thousand to several hundred thousand, the candidates for the registered image are narrowed down to, for example, several to several tens by the narrowing process.

The specific processing executed by the specific processing unit 2612 can be performed by using an image matching method using a local feature amount and an algorithm for verifying the consistency of the geometric arrangement of the feature points. First, the specific processing unit 2612 compares the local feature amount of the candidate registered image extracted by the narrowing process with the local feature amount of the identification image, and pairs the feature points that minimize the difference in the feature amount from both images. Ask as. Then, the specific processing unit 2612 extracts the feature points whose relative positional relationships with the other feature points do not contradict each other as the corresponding points, and identifies the registered image having the largest number of corresponding points.

In this case, as shown in FIG. 15, since the narrowing-down process is a process of comparing the global feature amounts of the registered image and the identification image, it is more than a specific process of comparing all the local feature amounts included in the image. Although the accuracy is low, the amount of processing is small and high-speed processing is possible. On the other hand, as shown in FIG. 16, the specific process is a process of searching for corresponding points for all of the local feature quantities A of the registered image and the identification image, and thus is a narrowing-down process of comparing the global feature quantities. The amount of processing is larger and it takes more time, but the accuracy is better. As described above, the search process can identify the same registered image as the identification image from a plurality of registered images at high speed and with high accuracy by executing a two-step process of a narrowing process and a specific process.

The collation processing unit 262 can execute the collation process. The collation process is a process that does not involve searching for images, and is executed when the adjusting device 30 collates specific images in order to adjust various conditions, algorithms, parameters, and the like. The collation process is executed using the same algorithm as the algorithm for collating images with each other in the search process. Therefore, the collation processing unit 262 can be a part of the search processing unit 261. Various conditions, algorithms, parameters and the like adjusted by the adjusting device 30 are reflected in the search processing of the search processing unit 261.

In the collation process, for example, the local feature amounts in the specific registered image registered in the evaluation image registration unit 242 as the evaluation image and the identification image are compared, and the corresponding points of the feature points are acquired to identify the registered image. Includes processing to match images. In this case, the collation processing unit 262 can collate the registered image and the identification image of the same subject according to the instruction from the adjusting device 30. Further, the collation processing unit 262 can collate the registered image and the identification image in which different subjects are photographed according to the instruction from the adjusting device 30. The collation processing unit 262 outputs the corresponding points of the local feature amount between the registered image and the identification image.

Further, the collation processing unit 262 can execute the collation process using the parameter set received from the collation identification condition adjusting device 32. As shown in FIG. 17, for example, the parameter set is a combination of values that can be set for each of a plurality of parameters for setting conditions for search processing and collation processing. Condition 1, condition 2, condition 3 ... In FIG. 17 are each one parameter set. In the case of the example of FIG. 17, there are a total of 5400 parameter sets.

The result notification processing unit 263 can execute the result notification processing. The result notification process includes a process of outputting the result of the search process and the result of the collation process to the adjusting device 30 or an external device. In this case, the result of the search process includes information about the identified registered image. In addition, the result of the collation process includes the corresponding points.

The photographing condition imparting device 27 has a function of imparting the above-mentioned imaging conditions to the imaging device 21 in response to an instruction from the imaging condition adjusting device 31 of the adjusting device 30. The feature extraction condition imparting device 28 has a function of imparting a feature extraction condition to the feature extraction device 22 in response to an instruction from the collation identification condition adjusting device 32 of the adjusting device 30.

The adjusting device 30 has a function of visualizing the result obtained from the image matching and identifying device 26 and displaying it on the display device 12, and receives feedback from the image matching and identifying device 26 to receive the imaging conditions and feature extraction device 22 of the photographing device 10. It has a function of adjusting various conditions, algorithms, parameters and the like used in the individual identification device 20 such as the conditions and algorithms used in the above. As shown in FIG. 1, the adjusting device 30 includes a photographing condition adjusting device 31, a collation identification condition adjusting device 32, and a visualization device 33.

The shooting condition adjusting device 31 adjusts various conditions when shooting a shooting target using the shooting device 10 automatically by receiving feedback from the image matching identification device 26 or manually by receiving an operation from the user. It has a function of instructing the shooting condition imparting device 27 on the conditions. The user can adjust various conditions by processing the shooting condition adjusting device 31, for example, by operating the input device 11. Further, the photographing condition adjusting device 31 has a function of automatically adjusting various conditions according to the number of feature points obtained by the processing of the feature extracting device 22. As shown in FIG. 7, the shooting condition adjusting device 31 includes a shooting device condition adjusting processing unit 311, a shooting target condition adjusting processing unit 312, a brute force condition generation processing unit 313, and an optimum condition presentation processing unit 314. ..

The shooting device condition adjustment processing unit 311 can execute the shooting device condition adjustment process for adjusting the shooting device conditions. By executing the shooting device condition adjustment process, the conditions mainly related to the optical device, for example, the shutter speed of the camera 101 of the shooting device 10, the sensitivity, the resolution, the shooting range, the focus, the brightness and the luminous intensity of the illumination 102, the color, etc. of the image sensor, etc. Is adjusted. The photographing device condition adjustment processing unit 311 transmits the adjusted photographing device condition to the photographing condition imparting device 27. As a result, the photographing device 21 mainly operates the camera 101 and the illumination 102 of the photographing device 10 in accordance with the photographing device conditions adjusted by the photographing device condition adjusting processing unit 311 to photograph the photographing target.

The shooting target condition adjustment processing unit 312 can execute the shooting target condition adjustment process for adjusting the shooting target condition, which is a condition relating to the subject to be shot. By executing the shooting target condition adjustment process, for example, the posture of the subject, that is, the angle of the subject with respect to the camera 101 or the illumination 102, the distance between the camera 101 or the illumination 102 and the subject, and the like are adjusted. The shooting target condition adjustment processing unit 312 transmits the adjusted shooting target condition to the shooting condition imparting device 27. As a result, the photographing device 21 mainly operates the stage 103 of the photographing device 10 according to the photographing target conditions adjusted by the photographing target condition adjustment processing unit 312 to photograph the photographing target.

The brute force condition generation processing unit 313 can execute the brute force condition generation processing. The brute force condition generation process includes a process of generating a combination of a shooting target condition within a preset range and a shooting device condition within a preset range by brute force. In this case, the image acquisition process executed by the image acquisition processing unit 212 of the photographing apparatus 21 includes a process of acquiring an image captured under all the conditions generated by the brute force condition generation process. As a result, the image acquisition processing unit 212 can acquire a large number of images taken under different conditions, that is, a large number of images taken by waving each of the shooting target condition and the shooting device condition.

For example, in the example of FIG. 19, the illumination distance, which is the distance between the subject and the illumination 102, is set as the imaging target condition, and the light intensity setting value, which is the light intensity of the illumination 102, is set as the imaging equipment condition. In this case, the brute force condition generation processing unit 313 generates all possible combinations of the illumination distance and the light amount set value by brute force. Then, the image acquisition processing unit 212 automatically shoots under the condition of this brute force combination to acquire the registered image and the identification image.

The optimum condition presentation processing unit 314 can execute the optimum condition presentation processing. In the optimum condition presentation process, among the registered image and the identification image captured under all the conditions generated by the brute force condition generation process, the registered image and the identification image captured by the same subject are collated by the collation processing unit 262. Is executed, and the process of presenting the combination of the shooting target condition and the shooting device condition with a large number of corresponding points obtained by the collation process to the user as the optimum condition or automatically setting the image is included. In this case, the optimum condition presentation processing unit 314 may display the optimum condition on the display device 12. As a result, the user can easily find and set the optimum combination of the shooting device condition and the shooting target condition.

For example, when the registered image and the identification image are taken of the same subject under the conditions shown in FIG. 19, the optimum condition presentation processing unit 314 determines the graph shown in FIG. 20, for example, from the corresponding points of the respective images. In this case, the horizontal axis represents the light intensity set value, that is, Lux, and the vertical axis represents the corresponding points to generate a graph. Then, the optimum condition presentation processing unit 314 finds a combination of conditions having a large number of corresponding points between the registered image and the identification image and having high robustness from this graph. In this case, it can be seen that the light amount setting value at the apex of the graph of A1 having the illumination distance 1 has a large number of corresponding points and is the most robust. Therefore, in this case, the optimum condition presentation processing unit 314 presents the light amount setting values of the vertices of the graph of A1 at the illumination distance 1 and the illumination distance 1 as the optimum conditions for the combination of the imaging target condition and the imaging device condition. In this way, the optimum condition presentation processing unit 314 presents to the user the optimum combination of the shooting target condition and the shooting device condition. The optimal condition presentation processing unit 314 can also select some candidates for conditions having a large number of corresponding points and high robustness, and present these conditions to the user as recommended conditions. Further, the optimum condition presentation processing unit 314 may perform a process of displaying the graph shown in FIG. 20 on the display device 12.

In the optimum condition presentation process, a plurality of feature extraction conditions, which are conditions related to the feature point extraction process, are prepared in advance, and the same registered image and identification image taken under all the conditions generated by the brute force condition generation process are prepared in advance. It includes a process of presenting a combination of conditions that maximizes the number of corresponding points obtained by executing a collation process using the feature points obtained by executing the feature point extraction process under a plurality of feature extraction conditions. As a result, the user can easily find and set the optimum combination of the shooting device condition and the shooting target condition including the feature extraction condition.

The collation identification condition adjusting device 32 has a function of tuning the parameters of the collation identification algorithm used in the search processing unit 261 and the collation processing unit 262 of the image collation identification device 26. The collation identification condition adjusting device 32 receives feedback from the image collation identification device 26 and automatically or manually receives an operation from the user to manually use the conditions and parameters used in the processing of the feature extraction device 22 and the image collation identification device 26. Etc., and have a function of instructing the feature extraction device 22 and the image matching identification device 26 on the conditions and parameters thereof. In the case of the present embodiment, the collation identification condition adjusting device 32 performs collation by the collation processing unit 262 of the image collation identification device 26 using the evaluation image registered in the evaluation image registration unit 242 of the recording device 24. The collation result In this case, the conditions and parameters are adjusted according to the corresponding points.

The user can adjust the conditions and parameters used in the feature extraction device 22 and the image matching identification device 26 by, for example, operating the input device 11 to execute the processing in the matching identification condition adjusting device 32. As shown in FIG. 9, the collation identification condition adjusting device 32 includes a feature extraction condition adjusting processing unit 321, a global feature amount dictionary setting processing unit 322, a global feature amount correlation coefficient calculation processing unit 323, and a parameter set generation processing unit 324. It also has a parameter set setting processing unit 325.

The feature extraction condition adjustment processing unit 321 can execute the feature extraction condition adjustment processing. The feature extraction condition adjustment process includes a process of adjusting the feature extraction conditions used in the feature point extraction process in the feature point extraction processing unit 221 or the local feature amount calculation process in the local feature amount calculation processing unit 222. The user operates the input device 11 to adjust and set the feature extraction conditions such as the number of stages of the pyramid image when extracting the feature points and calculating the local feature amount, and the threshold value for determining the corner and the curve-likeness. be able to.

In the feature extraction condition adjustment processing, the ratio of the positive correspondence score to the negative correspondence score is the largest for the feature extraction condition by the local feature amount calculation processing based on the result of the matching processing in the matching processing unit 262 of the image matching identification device 26. Includes processing to adjust to. In the present embodiment, as shown in FIG. 18A, the positive correspondence point refers to the same subject as the subject of the identification image among the registered image group registered in the evaluation image registration unit 242 of the recording device 24. It means the corresponding point between the captured registered image and the identification image. And, the positive correspondence point means the number of positive correspondence points. In the example of FIG. 18A, the subject S1 shown in the registered image and the subject S1 shown in the identification image are the same.

Further, as shown in FIG. 18B, the negative correspondence point is a registered image obtained by photographing a subject different from the subject of the identification image among the registered image group registered in the evaluation image registration unit 242 of the recording device 24. Means the corresponding point with the identification image. The number of negative correspondence points means the number of negative correspondence points. In the example of FIG. 18B, the subject S2 shown in the registered image and the subject S1 shown in the identification image are different.

The global feature dictionary setting processing unit 322 can execute the global feature dictionary setting process. The global feature dictionary setting process calculates a score for each global feature calculation statistical information, and the one with a high score is used in the local feature group classification process of the local feature group classification processing unit 223. Includes processing that is automatically set as. The global feature calculation statistical information is statistical information obtained in the process of calculating the global feature by the global feature calculation process, that is, in the process of the local feature group classification process in the local feature group classification processing unit 223, and is local. It can be represented by a histogram obtained by voting for dictionary information having a high correlation with the feature amount or a score of a statistic of the correlation value. Dictionary information with a high score based on statistical information such as the number of times specific dictionary information is selected is considered to have high performance. According to the global feature dictionary setting process, it is possible to automatically select a global feature dictionary having such high-performance dictionary information.

The global feature dictionary setting process includes a registered image and an identification image of the same subject taken based on the global feature correlation coefficient among a plurality of global feature dictionaries registered in the global feature dictionary registration unit 245. Select a global feature amount dictionary that can judge that the registered images and the identification images of different subjects are not the same, and set it in the global feature amount dictionary used for the local feature amount group classification process. including.

The global feature correlation coefficient is an index showing the degree of coincidence of the global features between the registered image and the identification image, that is, the performance. The greater the correlation between the global features between the two images, the higher the possibility that the same subject is included in both images. Therefore, the fact that the correlation of the global features between the images of the same subject is large means that the degree of coincidence of the global features calculated using the global feature dictionary for the images of the same subject is high. Means. That is, this means that the performance of the global feature dictionary is high between images of the same subject.

On the other hand, the smaller the correlation between the global features between the two images, the higher the possibility that the subjects included in the two images will be different. Therefore, the fact that the correlation of the global features between the images of different subjects is small means that the degree of coincidence of the global features calculated using the global feature dictionary for the images of the different subjects is low. do. That is, this means that the performance of the global feature dictionary is high between images of different subjects.

In the global feature dictionary setting process, for example, the global feature correlation coefficient can be calculated as follows, and the performance of the global feature dictionary can be determined based on the global feature correlation coefficient. That is, the global feature dictionary setting process is the same for the values obtained by normalizing the absolute values of the two global features to 1 and taking the inner product when the two global features are vectors V, and the value of the inner product is close to 1. It is judged that the property is high, and the identity is low when the value of the inner product is close to -1.

In this case, for example, one of the two global features vector Va is defined as the following equation (2), the normalized global feature vector Vb is defined as the following equation (3), and d is defined as a natural number. do. In this case, the normalized global feature vector Vb is expressed by the following equation (4) using the global feature vector Va. Then, assuming that the global feature vector Va is the following equation (5), the absolute value of the global feature vector Va is the following equation (6).

Then, if the other global feature vector Wa of the two global features is the following equation (7), the value of the inner product is the following equation (8). Then, the global feature dictionary registration unit 245 has a high possibility that the value of the inner product obtained by Eq. (8), that is, the one whose global feature correlation coefficient is close to 1, has high identity, that is, the same subject. If the value of the inner product is close to -1, it is judged that the identity is low, that is, the possibility that the subject is the same is low.

Further, in the global feature dictionary setting process, for example, the global feature correlation coefficient can be calculated as follows, and the performance of the global feature dictionary can be determined based on the global feature correlation coefficient. That is, the global feature correlation coefficient is the value of the distance between the vectors of the two global features. Then, in the global feature dictionary setting process, if the global feature correlation coefficient, which is the value of the distance between the vectors of the two global features, is close to 0, it is determined that the identity is high, and if it is large, the identity is low. Judge.

In this case, for example, when one of the two global feature vector Va is given the above equations (2) and (5) and the other global feature vector Wa is given the above equation (7) and the following equation (9), The distance between the global feature vectors Va and Wa is expressed by the following equation (10).

Then, the global feature dictionary registration unit 245 states that if the distance between the vectors Va and Wa obtained by Eq. (12) is short, that is, close to 0, the same subject is high, that is, there is a high possibility that they are the same subject. Judgment is made, and it is judged that the ones having a large distance between the vectors Va and Wa have low identity, that is, the possibility that they are the same subject is low. At this time, the respective vectors Va and Wa do not necessarily have to be normalized.

The global feature amount correlation coefficient calculation processing unit 323 can execute the global feature amount correlation coefficient calculation processing. The global feature amount correlation coefficient calculation process includes a process of comparing the global feature amount of each registered image with the global feature amount of the identification image and calculating the global feature amount correlation coefficient showing the correlation. For example, in the global feature amount correlation coefficient calculation process, the global feature amount correlation coefficient can be calculated by comparing the histograms of the global feature amount of each registered image and the global feature amount of the identification image.

The parameter set generation processing unit 324 can execute the parameter set generation processing. As shown in FIG. 17, for example, the parameter set generation process is within the setting range defined for each parameter with respect to the parameter for setting the condition of the collation process executed by the collation processing unit 262 of the image collation identification device 26. Includes a process to generate a plurality of parameter sets by combining each parameter set in the set value. In the case of the present embodiment, the parameter set generation process generates a parameter set for all combinations of set values of each parameter.

The parameter set setting processing unit 325 can execute the parameter set setting process. The parameter set setting process includes a process of setting a parameter set to be used in the search process of the search processing unit 261 based on the user's operation. The user can, for example, operate the input device 11 to set a parameter set to be used in the search process of the search process unit 261.

The visualization device 33 has a function of visualizing various information such as information received from the image matching identification device 26 and information related to the adjustment results of the photographing condition adjusting device 31 and the matching identification condition adjusting device 32 and displaying the information on the display device 12. Have. In the present embodiment, the visualization includes not only characters and numerical values but also figures and graphs to be displayed on the display device 12. As shown in FIG. 10, the visualization device 33 includes a feature point display processing unit 331, a global feature amount calculation statistical information display processing unit 332, a collation result display processing unit 333, and a condition display processing unit 334.

The feature point display processing unit 331 can execute the feature point display process. The feature point display process can be executed when various conditions and parameters are adjusted by the photographing condition adjusting device 31 and the collation identification condition adjusting device 32. In the feature point display process, for example, as shown in FIG. 21 (A), each feature point P extracted by executing the feature point extraction process in the feature point extraction processing unit 221 of the feature extraction device 22 is superimposed on an image and displayed on the display device. Includes the process to be displayed in 12. That is, the feature points P are superimposed and displayed on the subject S in the image. As a result, the user can visually recognize which part of the subject S in the image is extracted as a feature point. Further, as shown in FIG. 21A, the feature point display process is different for each local feature amount group to which the local feature amount belongs, and in this case, for each dictionary information d to which each feature point P is clustered and applied. A mode, for example, a process of displaying each feature point P on the display device 12 in a different color, size, and shape is included. In FIG. 21, the difference in color of each feature point P is represented by the difference in shade.

The global feature amount calculation statistical information display processing unit 332 can execute the global feature amount calculation statistical information display processing. The global feature amount calculation statistical information display process can be executed when various conditions and parameters are adjusted by the photographing condition adjusting device 31 and the collation identification condition adjusting device 32. The global feature amount calculation statistical information display process includes a process of displaying the global feature amount calculation statistical information on the display device 12. The global feature calculation statistical information is statistical information obtained in the process of calculating the global feature by the global feature calculation process executed by the global feature calculation processing unit 224 of the feature extraction device 22. As shown in FIG. 21 (B), the global feature calculation statistical information display process obtains the score of the histogram or the correlation value statistic obtained by voting for the dictionary information d having a high correlation with each local feature A. The process of displaying the feature amount calculation statistical information G on the display device 12 is included. That is, the global feature calculation statistical information G can be a histogram of the global feature obtained in the execution process of the global feature calculation process or a statistic score of the correlation value.

The collation result display processing unit 333 can execute the collation result display process. The collation result display process can be executed when various conditions and parameters are adjusted by the photographing condition adjusting device 31 and the collation identification condition adjusting device 32. The collation result display process includes a process of collectively visualizing the results of a plurality of collation processes executed by the collation processing unit 262 of the image collation identification device 26 and displaying them on the display device 12.

The collation result display process includes, for example, a process of graphing and displaying the collation result obtained by executing the collation process using each parameter set shown in FIG. 17 as shown in FIG. In this case, the collation result display processing unit 333 performs collation processing not only on the images of the same subject but also on the images of different subjects. Then, the collation result display processing unit 333 calculates and graphs the positive correspondence points, which are the correspondence points between the images of the same subject, and the negative correspondence points, which are the correspondence points between the images of different subjects. Then, it is displayed on the display device 12. In the graph of FIG. 22, the horizontal axis represents the condition No. and the vertical axis represents the corresponding points. Further, in FIG. 22, F1 shows a graph of positive correspondence points, and F2 shows a graph of negative correspondence points. In the example of FIG. 22, the condition No. It can be seen that N has the highest number of positive correspondence points and the smallest number of negative correspondence points. As a result, the user can grasp at a glance the condition No. which has a large number of positive correspondence points and a small number of negative correspondence points, that is, the condition No. which has good performance.

Further, the collation result display process includes a process of displaying the corresponding score evaluation information of the identification image used for collation on the display device 12 when the collation processing unit 262 of the image collation identification device 26 executes the collation process. The corresponding score evaluation information is information based on the corresponding score calculated by the collation process. For example, the corresponding score may be a numerical value as it is, or may be, for example, a level or a score of a plurality of stages.

In this case, as shown in FIG. 23, for example, the collation result display processing unit 333 displays the registered image display unit 51, the true value image display area 52, the false value image display area 53, and the true value image on the screen of the display device 12. The evaluation information display area 541, the true value image correlation information display area 542, the false value image evaluation information display area 551, and the false value image correlation information display area 552 are displayed. The registered image display unit 51 displays one of the registered image groups registered in the evaluation image registration unit 242 of the recording device 24 among the images to be collated.

In the true value image display area 52, among the images to be collated, the identification image registered in the evaluation image registration unit 242 of the recording device 24 and the image displayed in the registered image display unit 51. Images of the same subject as the subject taken under different conditions are displayed. In the example of FIG. 23, the subject S1 of the image displayed on the registered image display unit 51 and the subject S1 of the image displayed on the true value image display area 52 are the same individual. In the present embodiment, an image obtained by shooting the same subject as the subject of the image displayed on the registered image display unit 51 under different conditions is referred to as a true value image. In this case, the registered image in which the same subject is photographed and the identification image are registered in the evaluation image registration unit 242 in a state of being linked to each other, that is, in a state of forming a pair. That is, the collation result display processing unit 333 can recognize whether the same subject is photographed or a different subject is photographed by identifying the association between the registered image and the identification image.

Then, in the false value image display area 53, among the images to be collated, the identification image registered in the evaluation image registration unit 242 of the recording device 24 is displayed on the registered image display unit 51. An image of a subject different from the subject of the image taken under different conditions is displayed. In the example of FIG. 23, the subject S1 of the image displayed on the registered image display unit 51 and the subject S2 of the image displayed on the true value image display area 52 are different individuals. In the present embodiment, an image obtained by capturing a subject different from the subject of the image displayed on the registered image display unit 51 is referred to as a false value image.

Further, in the true value image evaluation information display area 541, the result of performing collation processing between the image displayed in the registered image display unit 51 and the image displayed in the true value image display area 52, that is, the true value. The corresponding score evaluation information of the image is displayed. Further, in the false value image evaluation information display area 551, the result of performing collation processing between the image displayed on the registered image display unit 51 and the image displayed on the false value image display area 53, that is, a false value. The corresponding score evaluation information of the image is displayed.

The collation result display process includes a process of displaying information based on the global feature amount correlation coefficient on the display device 12. In this case, the true value image correlation information display area 542 is composed of the global feature amount calculated from the registered image displayed on the registered image display unit 51 and the true value image displayed on the true value image display area 52. The correlation coefficient with the calculated global features is displayed. Then, in the false value image correlation information display area 552, the global feature amount calculated from the registered image displayed on the registered image display unit 51 and the false value image displayed on the false value image display area 53 are calculated. The correlation coefficient obtained by comparing with the global feature quantity is displayed.

The user operates, for example, the input device 11 to perform collation processing, that is, an image to be displayed in the registered image display unit 51, the true value image display area 52, the false value image display area 53, and the true value image evaluation information display area 541. Select an image and perform matching processing on that image. Then, when the collation processing is performed, the collation result display processing unit 333 displays the corresponding score evaluation information in the true value image evaluation information display area 541 and the false value image evaluation information display area 551 as the collation result, and also displays the corresponding score evaluation information. The global feature amount correlation coefficient is displayed in the true value image correlation information display area 542 and the false value image correlation information display area 552, respectively. As a result, the user can use the corresponding score evaluation information displayed in the true value image evaluation information display area 541 and the false value image evaluation information display area 551, and the true value image correlation information display area 542 and the false value image correlation information display area 552. By looking at the information based on the global feature amount correlation coefficient displayed in, the performance of various conditions and parameters used in the collation process can be confirmed. In this case, the information based on the global feature correlation coefficient may be the value of the global feature correlation coefficient itself, or the level or rank that the performance is expressed in several stages by processing the global feature correlation coefficient. May be.

As shown in FIG. 18, for example, the collation result display processing unit 333 may display the positive correspondence score or the negative correspondence score of the registered image and the identification image on the display device 12. Further, the collation result display processing unit 333 may separately display, for example, a registered image and an identification image, that is, the feature points that each image has uniquely and the feature points that each image has in common. good. In this case, the collation result display processing unit 333 may display the feature points uniquely possessed by each image and the number or ratio of the feature points commonly possessed. Then, the collation result display processing unit 333 sets a mode for displaying only the registered image and the identification image, a mode for displaying the registered image and the identification image by adding feature points, and a feature point and a corresponding point for the registered image and the identification image. In addition, the display mode may be arbitrarily switched.

Here, parts manufactured in the same process, such as industrial products, have extremely small differences in characteristics between individuals as compared with ecological authentication such as human faces and fingerprints. Therefore, since it is necessary to collate a large number of features appearing on the surface of an individual, the amount of calculation is large and it takes time to identify, and the amount of calculation increases explosively as the number of parameters to be identified increases.

On the other hand, the individual identification system 1 includes an image acquisition processing unit 212, a feature point extraction processing unit 221, a local feature amount calculation processing unit 222, a local feature amount group classification processing unit 223, a search target image registration processing unit 231 and a global feature. It includes a quantity registration processing unit 234, a narrowing down processing unit 2611, and a specific processing unit 2612. The image acquisition processing unit 212 can execute an image acquisition process for acquiring an image of a subject photographed by the photographing device 10. The feature point extraction processing unit 221 can execute the feature point extraction process for extracting the feature points from the image acquired by the image acquisition process. The local feature amount calculation processing unit 222 can execute a local feature amount calculation process for calculating the local feature amount of the feature points extracted by the feature point extraction process.

The local feature group classification processing unit 223 classifies a plurality of local features obtained by the local feature calculation process into a predetermined number, for example, 64 local feature groups according to the value of the local feature. Group classification processing can be executed. The global feature calculation processing unit 224 can execute the global feature calculation process for calculating the global feature based on each local feature group. The search target image registration processing unit 231 can execute the search target image registration process of registering a plurality of search target images as registered images in the search target image registration unit 241 in advance. The global feature amount registration processing unit 234 can execute the global feature amount registration process for registering the global feature amount related to the registered image in the global feature amount registration unit 244. As shown in FIG. 15, for example, the narrowing-down processing unit 2611 selects and narrows down a predetermined number of registered images having a global feature amount having a high correlation with the global feature amount of the identification image from a plurality of registered images. Is feasible. Then, as shown in FIG. 16, for example, the specific processing unit 2612 compares the local feature amount of the candidate registered image extracted by the narrowing process with the local feature amount of the identification image, and the corresponding points of the local feature amount are obtained. It is possible to execute a specific process for identifying the most registered images.

According to this configuration, the narrowing-down processing unit 2611 roughly narrows down the candidate registered images by comparing the global feature amounts of the registered image and the identification image by executing the narrowing-down processing. This narrowing down process requires less calculation than the specific process of comparing all the local features of each image and can be processed at a much higher speed, but the accuracy is difficult to obtain. Therefore, next, the specific processing unit 2612 identifies the target by comparing the registered image of the narrowed-down candidate with the local feature amount of the identification image by executing the specific processing. This specific process requires a large amount of calculation and takes a long time as compared with the narrowing process in which the global feature amount is compared, but the accuracy is good. In this way, the individual identification system 1 realizes a high-speed and highly accurate search by executing a two-step process of narrowing down by the global feature amount and specifying by the local feature amount. be able to. This is particularly effective when the population parameter of the registered image is increased.

The individual identification system 1 further includes at least one of a photographing device condition adjusting processing unit 311, a photographing target condition adjusting processing unit 312, and a feature extraction condition adjusting processing unit 321. The shooting device condition adjustment processing unit 311 can execute the shooting device condition adjustment process for adjusting the shooting device condition, which is a condition related to the shooting device 10, based on the user's operation. The shooting target condition adjustment processing unit 312 can execute the shooting target condition adjustment process for adjusting the shooting target condition, which is a condition related to the shooting target, based on the user's operation. Then, the feature extraction condition adjustment processing unit 321 can execute the feature extraction condition adjustment process for adjusting the feature extraction condition, which is a condition related to the feature point extraction process, based on the user's operation.

According to this, the user operates at least one of the shooting device condition adjustment processing unit 311, the shooting target condition adjustment processing unit 312, and the feature extraction condition adjustment processing unit 321 to set various conditions and parameters, that is, to perform tuning. It can be carried out. As a result, for example, even if the surface pattern is changed in the middle of the manufacturing process, that is, even if the surface pattern changes between the time of registering the image and the time of identification, it can be flexibly dealt with, and the accuracy is high. Can be identified well.

The individual identification system 1 includes a global feature dictionary generation processing unit 251. The global feature dictionary generation processing unit 251 performs a global feature dictionary generation process for generating a global feature dictionary having each representative value of each local feature group obtained from a plurality of learning images acquired in advance as dictionary information. It is feasible. Then, the global feature calculation process includes a process of calculating a histogram obtained by voting for dictionary information having a high correlation with each local feature as a global feature. In the global feature calculation process, the vector obtained by summing the difference vectors between each local feature and the representative value, which is dictionary information with high correlation, is calculated for the number of representative values, and the vector connecting them is calculated as the global feature. It may include the processing to be performed.

In other words, depending on the type and state of the subject, what kind of features are extracted as feature points will change. Therefore, simply classifying the local features obtained from the image and calculating the global features is sufficient. , The performance of global features may be low, that is, the discrimination ability may be low. On the other hand, according to this configuration, by using a global feature dictionary having dictionary information suitable for the subject, it is possible to obtain a global feature with good performance even if the type or state of the subject changes. As a result, the accuracy of the narrowing process can be improved.

Further, in this configuration, the global feature dictionary generation process includes a process of setting the centroid of the local feature group as a representative value in the dictionary information. The centroid of the local feature group is representative of the local feature group. Therefore, by setting the centroid of the local feature group as a representative value in the dictionary information, it is possible to obtain a global feature dictionary with good performance.

The individual identification system 1 includes an image acquisition processing unit 212, a feature point extraction processing unit 221 and a feature point display processing unit 331. As shown in FIG. 21, for example, the feature point display processing unit 331 can execute the feature point display process of superimposing the feature points extracted by the feature point extraction process on the image and displaying them on the display device 12.

The individual identification system 1 includes at least one of a photographing device condition adjusting processing unit 311, a photographing target condition adjusting processing unit 312, and a feature extraction condition adjusting processing unit 321. The shooting device condition adjustment processing unit 311 can execute the shooting device condition adjustment process for adjusting the shooting device condition, which is a condition related to the shooting device 10, based on the user's operation. The shooting target condition adjustment processing unit 312 can execute the shooting target condition adjustment process for adjusting the shooting target condition, which is a condition related to the shooting target, based on the user's operation. Then, the feature extraction condition adjustment processing unit 321 can execute the feature extraction condition adjustment process for adjusting the feature extraction condition, which is a condition related to the feature point extraction process, based on the user's operation.

For example, when the individual identification system 1 is used at a production site, some parts to be identified change with time during the manufacturing process, heat processing, or the like, and the color of the surface pattern thereof changes. For parts that are processed to change the color of the surface pattern during the manufacturing process, the position and number of feature points extracted before and after the processing, that is, at the time of image registration and at the time of identification, And the local feature amount may change significantly, and as a result, the identification accuracy is lowered.

On the other hand, according to this configuration, since the feature points extracted by the feature point extraction process are superimposed on the image and displayed on the display device 12, the user can extract which part of the image as the feature point. You can check if it is. Then, the user can focus on extracting feature points from places where the color of the surface pattern is unlikely to change even after the manufacturing process, so that the feature extraction condition adjustment processing unit 321 and the imaging target condition adjustment processing unit 312, and the shooting target condition adjustment processing unit 312, and At least one of the feature extraction condition adjustment processing unit 321 can be operated to set various conditions and parameters while checking the positions of the feature points to be extracted, that is, tuning can be performed. As a result, even if the surface pattern is changed in the middle of the manufacturing process, that is, the surface pattern changes between the time of registering the image and the time of identification, it is possible to accurately identify the image.

The individual identification system 1 includes a local feature amount calculation processing unit 222 and a local feature amount group classification processing unit 223. Then, as shown in FIG. 21A, the feature point display process displays the feature points in a mode different for each local feature amount group to which the local feature amount belongs, for example, a mode in which the color, size, and shape are different. The process of displaying in 12 is further included.

According to this, the user can grasp at a glance what kind of feature point exists in the image. This makes it easier to set various conditions and parameters while checking the positions of the extracted feature points. As a result, it is possible to more accurately identify those that are subjected to processing that changes the surface pattern in the middle of the manufacturing process.

The individual identification system 1 includes a global feature amount calculation processing unit 224, and further includes a global feature amount calculation statistical information display processing unit 332. As shown in FIG. 21 (B), the global feature calculation statistical information display processing unit 332 is obtained in the local feature group classification process in the global feature calculation process by executing the global feature dictionary generation process. It is possible to execute the global feature calculation statistical information display process for displaying the feature calculation statistical information G on the display device 12.

According to this, the user can easily grasp what kind of global feature amount is obtained. That is, by displaying information on which dictionary information of the area feature amount dictionary and how many votes have been voted, the user can grasp the bias of how the dictionary information of the area feature amount dictionary is selected. This makes it easier to set various conditions and parameters related to the global feature amount while looking at the displayed global feature amount calculation statistical information G. As a result, it is possible to more accurately identify those that are subjected to processing that changes the surface pattern in the middle of the manufacturing process.

The individual identification system 1 further includes a global feature dictionary generation processing unit 251. The global feature dictionary generation processing unit 251 performs a global feature dictionary generation process for generating a global feature dictionary having each representative value of each local feature group obtained from a plurality of learning images acquired in advance as dictionary information. It is feasible. Then, in the global feature calculation statistical information display processing, as shown in FIG. 13, in the calculation process of the global feature, the histogram or the statistic of the correlation value obtained by voting for the dictionary information d having a high correlation with each local feature. As shown in FIG. 21, the score is displayed on the display device 12 as the global feature amount calculation statistical information G.

According to this, the user can easily grasp the global feature amount visually and intuitively. This makes it easier to set various conditions and parameters related to the global feature amount while looking at the displayed global feature amount calculation statistical information G. As a result, it is possible to more accurately identify those that are subjected to processing that changes the surface pattern in the middle of the manufacturing process.

The individual identification system 1 further includes a global feature dictionary setting processing unit 322. The global feature dictionary setting processing unit 322 calculates a score for each global feature calculation statistical information, and automatically sets a score with a high score as a global feature dictionary to be used in the local feature group classification process. Dictionary setting processing can be executed.

According to this, since it is automatically set as an appropriate global feature dictionary, the user does not have to manually adjust the conditions and parameters one by one, and as a result, the user convenience is improved.

When the individual identification system 1 having this configuration is used, for example, in a factory or the like, it is assumed that the acquisition of the registered image and the acquisition of the identification image are performed at different places. In this case, dust generated in the factory, lighting in the factory, natural light such as outside light, ambient noise sources, etc. become factors of disturbance, and the shooting environment between the shooting of the registered image and the shooting of the identification image changes significantly. there is a possibility. Then, the extracted feature points, local feature amounts, and the like may change, and the discrimination performance may deteriorate.

Therefore, the individual identification system 1 having this configuration includes an image acquisition processing unit 212, a feature point extraction processing unit 221, a local feature amount calculation processing unit 222, a shooting target condition adjustment processing unit 312, and a shooting device condition adjustment processing unit. 311 and. According to this, the user operates the shooting target condition adjustment processing unit 312 and the shooting device condition adjustment processing unit 311 to determine the shooting target condition and the shooting device condition suitable for extracting the feature points and calculating the local feature amount. Can be adjusted. Therefore, by setting appropriate conditions according to changes in the environment, the influence of disturbance can be reduced, and the pattern to be characterized can be appropriately highlighted. As a result, the identification accuracy can be improved.

The individual identification device 20 further includes a brute force condition generation processing unit 313. The brute force condition generation processing unit 313 can execute a brute force condition generation process that generates a combination of a shooting target condition within a preset range and a shooting device condition within a preset range by brute force. Then, the image acquisition process includes a process of automatically acquiring an image obtained by capturing all the conditions generated by the brute force condition generation process.

According to this, it is possible to automatically shoot and acquire a large number of images in which the shooting target conditions and the shooting device conditions are changed. Therefore, when searching for the optimum combination of conditions by changing each of the shooting target condition and the shooting device condition, it is not necessary for the user to manually generate the combination of conditions and shoot. As a result, it is possible to reduce the time and effort required to adjust the shooting target conditions and the shooting device conditions. Further, since the conditions with good performance can be easily set, the identification performance can be improved.

The individual identification device 20 further includes an optimum condition presentation processing unit 314. The optimal condition presentation processing unit 314 has a large number of corresponding points between the registered image obtained by capturing the same subject and the identification image among the registered images and identification images captured under all the conditions generated by the brute force condition generation process. It is possible to execute an optimum condition presentation process that presents a combination of a shooting target condition and a shooting device condition as an optimum condition. According to this, since the user can easily acquire and set the optimum conditions having a large number of feature points, it is possible to further reduce the time and effort for adjusting the conditions of the shooting target condition and the shooting device condition. Further, since the optimum combination of the shooting target condition and the shooting device condition can be easily set, the identification performance can be improved.

In the optimum condition presentation process, a plurality of feature extraction conditions, which are conditions related to the feature point extraction process, are prepared in advance, and a registered image and an identification image in which the same subject is photographed under all the conditions generated by the brute force condition generation process are prepared in advance. It includes a process of presenting a combination of conditions that maximizes the number of corresponding points when collating using the feature points obtained by executing the feature point extraction process under a plurality of feature extraction conditions. According to this, the user can easily acquire the optimum condition among the combination with the feature extraction condition related to the feature point extraction process in addition to the conditions of the shooting target condition and the shooting device condition. As a result, it is possible to further reduce the time and effort required to adjust the shooting target condition, the shooting device condition, and the feature extraction condition. Then, since the optimum combination of the shooting target condition, the shooting device condition, and the feature extraction condition can be easily set, the identification performance can be further improved.

For example, in parts of industrial products, unlike human faces and fingerprints, the characteristics of patterns appearing on the surface of the parts vary depending on the type and material of the object to be collated and identified and the processing method. Therefore, it is necessary to adjust the parameters of the algorithm used for collation identification to appropriate ones depending on the type of the object to be identified. However, there are many parameter items, and it takes a lot of time and effort for the user to manually search for the optimum combination from all the combinations.

Therefore, the individual identification system 1 includes a feature point extraction processing unit 221 and a local feature amount calculation processing unit 222. Further, the individual identification system 1 further includes a collation processing unit 262, a parameter set generation processing unit 324, and a parameter set setting processing unit 325. The collation processing unit 262 can execute a collation process of comparing the local feature amounts of the registered image and the identification image, acquiring the corresponding points of the feature points, and collating the registered image with the identification image. The parameter set generation processing unit 324 generates a plurality of parameter sets by combining each parameter set in the set value within the setting range defined for each parameter with respect to the parameter for setting the condition of the collation process. The generation process can be executed. The parameter set setting processing unit 325 can execute the parameter set setting process for setting the parameter set based on the user's operation.

The collation process includes a process of executing a collation process using each parameter set generated in the parameter set generation process. The individual identification system 1 further includes a collation result display processing unit 333. The collation result display processing unit 333 can execute the collation result display process of collectively displaying the collation results when the collation process is executed using each parameter set on the display device 12.

According to this, a parameter set which is a combination of parameters is automatically generated, and a collation process is performed using each parameter set. Therefore, the user does not have to manually create a combination of parameters and perform collation. Then, the collation results are collectively displayed on the display device 12. Therefore, the user can confirm the performance of each parameter set by looking at the result displayed on the display device 12. Then, the user may select an appropriate parameter set by looking at the result displayed on the display device 12. Thereby, regardless of the type of the target to be collated and identified, the parameters of the algorithm used for collation and identification can be easily and accurately adjusted without hassle.

As shown in FIG. 22, the collation result display process includes a process of graphing and displaying the collation result obtained by executing the collation process using each parameter set. It is easy to visually and intuitively grasp the one with good performance among multiple parameter sets. As a result, it is possible to easily perform the parameter adjustment work with high accuracy.

The individual identification system 1 further includes an evaluation image registration processing unit 232. The evaluation image registration processing unit 232 includes a registered image group including a plurality of registered images in which different subjects are photographed, and an identification image captured in the same subject as one of the registered image groups under different conditions. It is possible to execute the evaluation image registration process of registering the images in the evaluation image registration unit 242 in a state of being linked to each other as the evaluation images of the collation process.

This eliminates the need for the user to take an image each time the parameter set is evaluated. As a result, the parameter set can be easily evaluated. Further, according to this, the registered image and the identification image obtained by photographing the same subject are registered in a state of being linked to each other. Therefore, when the collation process is executed using the registered image and the identification image, it becomes clear in advance whether the registered image and the identification image correspond to each other, that is, whether the same subject is photographed or a different subject is photographed. ing. As a result, by using an image having a clear correspondence, it is possible to clarify the result desired in the collation process, and the user can easily evaluate the parameter set used in the collation process.

That is, when the collation process is executed using the registered image and the identification image in which the same subject is photographed, the collation process is executed using the registered image and the identification image in which the corresponding points are large and different subjects are photographed. In some cases, if the number of corresponding points is small, the user can evaluate that the performance of the parameter set used for the collation process is high. On the contrary, when the collation process is executed using the registered image and the identification image in which the same subject is photographed, the number of corresponding points is small, or the collation process is executed using the registered image and the identification image in which different subjects are photographed. If the number of corresponding points is small, the user can evaluate that the performance of the parameter set used for the collation process is low.

The collation process includes a process of collating one or more registered images included in the registered image group with the identification image. Then, as shown in FIG. 23, the collation result display process includes a process of displaying the corresponding score evaluation information based on the corresponding score calculated in the collation process on the display device 12. According to this, the user can easily evaluate the performance of the parameter set by looking at the corresponding score evaluation information displayed on the display device 12, and thereby the appropriate parameters according to the type of the target to be collated and identified. You can select a set. As a result, high collation and identification performance can be exhibited regardless of the type of object.

The individual identification device 20 further includes a feature extraction condition adjustment processing unit 321. The feature extraction condition adjustment processing unit 321 can execute the feature extraction condition adjustment processing. The feature extraction condition adjustment process is different from the positive correspondence score, which is the corresponding score between the registered image and the identification image obtained by shooting the same subject as the subject of the identification image in the registered image group, and the subject of the identification image in the registered image group. It includes a process of calculating a negative correspondence score, which is a correspondence score between a registered image in which a subject is photographed and an identification image. Then, the feature extraction condition adjustment process includes a process of adjusting the feature extraction condition by the local feature amount calculation process so that the ratio of the positive correspondence score and the negative correspondence score becomes the largest.

In this case, the feature extraction condition adjustment processing unit 321 searches for a feature extraction condition having a large number of positive correspondence points and a small number of negative correspondence points so that the ratio of the positive correspondence points to the negative correspondence points is the largest. A large number of positive correspondence points means a large number of correspondence points between images of the same subject. This means that in the collation process between images of the same subject, it is determined that the subjects included in the images are likely to be the same, that is, identification between images of the same subject. It means that the performance is high. Further, the fact that the number of negative correspondence points is small means that the number of correspondence points between images of different subjects is small. This means that in the collation process between images of different subjects, it is highly likely that the subjects included in the images are different, that is, the discrimination performance between the images of different subjects is high. Means.

In this way, the feature extraction condition adjustment processing unit 321 automatically adjusts the feature extraction conditions so that the collation and identification performance between the images of the same subject and the images of different subjects is improved. It is possible to save the user's trouble of adjustment while obtaining high collation identification performance.

The individual identification system 1 includes a local feature amount group classification processing unit 223 and a global feature amount calculation processing unit 224. The individual identification system 1 further includes a global feature amount correlation coefficient calculation processing unit 323. The global feature amount correlation coefficient calculation processing unit 323 compares the global feature amount of each registered image with the global feature amount of the identification image, and calculates the global feature amount correlation coefficient showing the correlation. Calculation processing can be executed. Then, the collation result display process includes a process of displaying information based on the global feature amount correlation coefficient on the display device 12.

The global feature amount correlation coefficient is one index showing the degree of matching of the global feature amount between the registered image and the identification image, that is, the performance. By displaying the information based on the global feature amount correlation coefficient on the display device 12, the user can easily grasp the performance of the global feature amount. Therefore, the user can perform the adjustment work of various conditions and parameters on the display device 12 using the information based on the global feature amount correlation coefficient as one guideline, which makes the work easier.

The individual identification system 1 further includes a global feature dictionary registration unit 245 and a global feature dictionary setting processing unit 322. The global feature dictionary setting processing unit 322 can execute the global feature dictionary setting process. The global feature dictionary setting process selects a registered image and an identification image of the same subject based on the global feature correlation coefficient among a plurality of global feature dictionaries registered in the global feature dictionary registration unit 245. This includes a process of selecting a global feature amount dictionary that can be judged to be the same and that can determine that the registered image and the identification image obtained by capturing different subjects are not the same, and setting the global feature amount dictionary to be used for the local feature amount group classification process.

As described above, the global feature amount correlation coefficient is one index showing the degree of coincidence of the correlation of the global feature amount between the registered image and the identification image. The greater the correlation between the global features between the two images, the higher the possibility that the same subject is included in both images. Therefore, the fact that the correlation of the global features between the images of the same subject is large means that the degree of coincidence of the global features calculated for the two images of the same subject is high. That is, this means that the performance of the global feature dictionary is high between images of the same subject.

On the other hand, the smaller the correlation between the global features between the two images, the higher the possibility that the subjects included in the two images will be different. Therefore, the fact that the correlation of the global features between the images of different subjects is small means that the degree of coincidence of the global features calculated using the global feature dictionary for the images of the different subjects is low. do. That is, this means that the performance of the global feature dictionary is high between images of different subjects.

According to this configuration, the global feature dictionary setting process automatically sets a global feature dictionary with high performance between images of the same subject and high performance between images of different subjects. be able to. Therefore, the time and effort of the user's adjustment can be saved, and high collation identification performance can be obtained.

In the global feature dictionary setting process, when two global features are vectors, the absolute value is normalized to 1 and the inner product is taken, and the value of the inner product is close to 1 and the identity is high. It may include a process of determining that the value of the inner product is close to -1 and that the identity is low. Further, the global feature correlation coefficient can be a value of the distance between the vectors of the two global features. In this case, the global feature dictionary setting process includes a process of determining that the identity is high when the value of the distance between the vectors of the two global features is close to 0, and determining that the identity is low when the value is large. According to these configurations, the performance of the global feature dictionary can be quantitatively evaluated.

The individual identification system 1 includes an individual identification device 20 having an image acquisition processing unit 212, a feature point extraction processing unit 221 and a specific processing unit 2612. In addition to the individual identification device 20, the individual identification system 1 includes at least two of the visualization device 33, the photographing condition adjusting device 31, and the collation identification condition adjusting device 32.

The visualization device 33 has a feature point display processing unit 331. The feature point display processing unit 331 can execute the feature point display process of superimposing the feature points extracted by the feature point extraction process on the image and displaying them on the display device 12.

The shooting condition adjusting device 31 compares the local feature amounts in the registered image and the identification image obtained by shooting the same subject among the plurality of registered images registered in advance, acquires the corresponding points of the feature points, and acquires the corresponding points of the feature points, and the registered image and the identification image. The individual identification device 20 is made to execute the collation process for collating with the image, and the image capture target condition adjustment processing unit 312 and the image capture condition adjustment device 31 are provided. The shooting target condition adjustment processing unit 312 can execute the shooting target condition adjustment process for adjusting the shooting target condition, which is a condition related to the shooting target. The shooting device condition adjustment processing unit 311 can execute the shooting device condition adjustment process for adjusting the shooting device condition, which is a condition related to the shooting device 10.

The collation identification condition adjusting device 32 includes a parameter set generation processing unit 324 and a parameter set setting processing unit 325. The parameter set generation processing unit 324 generates a parameter set for generating a plurality of parameter sets by combining the parameters set in the set values within the range defined for each parameter with respect to the parameters for setting the conditions of the collation processing. Processing can be executed. The parameter set setting processing unit 325 can execute the parameter set setting process for setting the parameter set based on the user's operation. Then, the collation identification condition adjusting device 32 causes the collation processing unit 262 to execute a process of executing the collation process using each parameter set generated in the parameter set generation process. Then, the collation identification condition adjusting device 32 further has a collation result display processing unit 333 capable of executing the collation result display process of collectively displaying the collation results of the collation process on the display device 12 using the parameter set.

Since the individual identification system 1 includes the visualization device 33, the feature points extracted by the feature point extraction process are superimposed on the image and displayed on the display device 12, so that the user can use which part of the image as the feature point. You can check if it is extracted. Then, the user sets various conditions and parameters while checking the positions of the extracted feature points so that the feature points can be extracted mainly in the places where the surface pattern is hard to change even in the manufacturing process, that is, tuning is performed. Can be done. As a result, even if the surface pattern is changed in the middle of the manufacturing process, that is, even if the surface pattern changes between the time of registering the image and the time of identification, it is possible to accurately identify the image. Excellent effect can be obtained.

When the individual identification system 1 includes the shooting condition adjusting device 31, the user operates the shooting target condition adjusting processing unit 312 and the shooting device condition adjusting processing unit 311 to extract feature points and calculate local feature quantities. Suitable shooting target conditions and shooting device conditions can be adjusted. Therefore, by setting appropriate conditions according to changes in the environment, the influence of disturbance can be reduced, and the characteristic pattern can be appropriately highlighted. As a result, an excellent effect such as being able to improve the identification accuracy can be obtained.

By providing the collation identification condition adjusting device 32 in the individual identification system 1, the user does not have to manually create a combination of parameters and perform collation. Then, the collation results are collectively displayed on the display device 12. Therefore, the user can confirm the performance of each parameter set by looking at the result displayed on the display device 12. Then, the user may select an appropriate parameter set by looking at the result displayed on the display device 12. As a result, regardless of the type of the target to be collated and identified, it is possible to obtain an excellent effect that the parameters of the algorithm used for collation and identification can be easily and effortlessly adjusted with high accuracy.

Then, the individual identification system 1 is provided with at least two of the visualization device 33, the photographing condition adjusting device 31, and the collation identification condition adjusting device 32 at the same time, so that two or more of the above-mentioned excellent effects of each device can be simultaneously obtained. Obtainable. As a result, it is possible to respond to changes in the surface pattern of the identification target, make it less susceptible to disturbances, reduce the user's time and effort required to adjust the conditions and parameters, and improve the accuracy of the conditions and parameters. As a result, the performance of collation identification can be improved.

The individual identification system 1 includes a visualization device 33, a photographing condition adjusting device 31, and a collation identification condition adjusting device 32. According to this, by simultaneously providing the visualization device 33, the photographing condition adjusting device 31, and the collation identification condition adjusting device 32, all the excellent effects of the above-mentioned devices can be obtained at the same time. That is, according to this individual identification system 1, it is possible to respond to changes in the surface shape of the identification target, make adjustments that are not easily affected by disturbances, and further reduce the user's labor required for adjusting conditions and parameters. Can be done.

The search processing unit 261 further has a narrowing down processing unit 2611 capable of executing the narrowing down processing. In the narrowing down process, a predetermined number of registered images having a global feature amount having a high correlation with the global feature amount of the identification image are selected as candidates in the specific process from among a plurality of registered images registered in the search target image registration unit 241. Includes narrowing down processing.

According to this, the search process can realize a high-speed search by executing a two-step process of narrowing down by the global feature amount and the specific process of specifying by the local feature amount. As a result, according to the individual identification system 1 of the present configuration, as a result, it is possible to realize a search with high collation identification performance and high speed.

Although the present disclosure has been described in accordance with the examples, it is understood that the present disclosure is not limited to the examples and structures. The present disclosure also includes various modifications and modifications within a uniform range. In addition, various combinations and forms, as well as other combinations and forms that include only one element, more, or less, are also within the scope of the present disclosure.

The controls and methods thereof described in the present disclosure are realized by a dedicated computer provided by configuring a processor and memory programmed to perform one or more functions embodied by a computer program. May be done. Alternatively, the controls and methods thereof described in the present disclosure may be implemented by a dedicated computer provided by configuring the processor with one or more dedicated hardware logic circuits. Alternatively, the control unit and method thereof described in the present disclosure may be a combination of a processor and memory programmed to perform one or more functions and a processor composed of one or more hardware logic circuits. It may be realized by one or more dedicated computers configured. Further, the computer program may be stored in a computer-readable non-transitional tangible recording medium as an instruction executed by the computer.

1 ... Individual identification system, 10, 10a, 10b ... Imaging device, 12 ... Display device, 20 ... Individual identification device, 30 ... Adjustment device, 31 ... Imaging condition adjustment device, 32 ... Verification identification condition adjustment device, 33 ... Visualization device , 212 ... Image acquisition processing unit, 221 ... Feature point extraction processing unit, 222 ... Local feature amount calculation processing unit, 241 ... Search target image registration unit, 261 ... Search processing unit, 262 ... Collation processing unit, 311 ... Imaging device conditions Adjustment processing unit, 312 ... Shooting target condition adjustment processing unit, 324 ... Parameter set generation processing unit, 325 ... Parameter set setting processing unit, 331 ... Feature point display processing unit, 333 ... Collation result display processing unit, 2611 ... Narrowing down processing unit , 2612 ... Specific processing unit

Claims (3)

An image acquisition processing unit (212) capable of executing an image acquisition process for acquiring an image of a subject captured by a photographing device as an identification image, and an image acquisition processing unit (212).
A feature point extraction processing unit (221) capable of executing a feature point extraction process for extracting feature points included in the identification image acquired in the image acquisition process, and a feature point extraction processing unit (221).
A local feature amount calculation processing unit (222) capable of executing a local feature amount calculation process for calculating the local feature amount of the feature point extracted by the feature point extraction process, and a local feature amount calculation processing unit (222).
The local feature amount of the plurality of registered images registered in advance is compared with the local feature amount of the identification image, and the local feature amount of the identification image has the highest correlation among the plurality of registered images. An individual identification system (1) including an individual identification device having a specific processing unit (2612) capable of executing a specific process for specifying a registered image having a local feature amount.
In addition to the individual identification device,
It has a feature point display processing unit (331) capable of executing a feature point display process of superimposing the feature points extracted by the feature point extraction process on the image and displaying them on a display device.
Visualization device (33),
Or
Among the plurality of registered images registered in advance, the registered image obtained by photographing the same subject and the identification image are compared with each other, and the corresponding points of the feature points are obtained to obtain the registered image and the identification image. In addition to causing the individual identification device to execute a collation process for collating
A shooting target condition adjustment processing unit (312) capable of executing a shooting target condition adjustment process for adjusting a shooting target condition, which is a condition related to a shooting target, and a shooting target condition adjustment processing unit (312).
A shooting device condition adjustment processing unit (311) capable of executing a shooting device condition adjustment process for adjusting the shooting device conditions, which is a condition related to the shooting device, and a shooting device condition adjustment processing unit (311).
Shooting condition adjusting device (31),
Alternatively, for the parameters for setting the conditions of the collation process, a parameter set generation process for generating a plurality of parameter sets by combining the parameters set in the set values within the setting range defined for each parameter is executed. Possible parameter set generation processing unit (324) and
It has a parameter set setting processing unit (325) capable of executing a parameter set setting process for setting the parameter set based on a user operation.
In addition to having the collation processing unit execute a process of executing the collation process using each of the parameter sets generated in the parameter set generation process,
Further, it has a collation result display processing unit (333) capable of executing a collation result display process of collectively displaying the collation results of the collation process on the display device (12) using each of the parameter sets.
Collation identification condition adjustment device (32),
Equipped with at least two of the devices,
Individual identification system (1). The visualization device, the photographing condition adjusting device, and the collation identification condition adjusting device are provided.
The individual identification system according to claim 1. The search processing unit
Among the plurality of registered images registered in the search target image registration unit, a predetermined number of registered images having a global feature amount having a high correlation with the global feature amount of the identification image are candidates in the specific process. Further has a narrowing down processing unit (2611) capable of executing the narrowing down processing.
The individual identification system according to claim 1 or 2.

Images