JP3995181B2 - Individual identification device - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to an individual identification device that performs individual identification by extracting an arbitrary feature from an image.
[0002]
[Prior art]
Conventionally, as described in, for example, the following literature, a technique using an iris has been proposed as an individual identification technique.
[0003]
[References] US Pat. No. 5,291,560, United States Patent BIOMETRIC PERSONAL IDENTIFICATION SYSTEM BASED ON IRIS ANALISYS
[0004]
In the prior art as described above, a circle circumscribing the pupil is detected in image data acquired by a television camera, and a coordinate system is set based on each circle. By setting the coordinate system in this way, changes in the size of the pupil, changes in the distance between the subject and the camera, and the like are absorbed. Based on the coordinate system set in this way, the iris is divided into several regions, and an iris code is generated for each region. Individual identification is performed by comparing the iris codes.
[0005]
[Problems to be solved by the invention]
However, in the above conventional technique, normal or abnormal pupil shape or iris shape is caused by the influence of image distortion due to the positional relationship between the camera and subject at the time of data acquisition, the data acquisition environment, etc. Even when collation processing is performed, it is difficult to obtain the expected collation accuracy.
[0006]
[Means for Solving the Problems]
  The present invention employs the following configuration in order to solve the above-described problems.
<Configuration 1>
  In an individual identification device that generates an iris code from an eye image and performs individual identification based on the iris code, an input image storage unit that stores eye images of a plurality of different scenes of one individual, and an input image storage unit For each of the plurality of eye images, a feature data generation unit that generates an iris code having an extraction error by changing the value of each parameter of the iris coordinate system, and each iris code generated by the feature data generation unit In contrast, a predetermined number of iris codes obtained from an arbitrary image in which the average difference between an iris code obtained from an arbitrary image in the same individual and an iris code obtained from other images is smaller than a predetermined value A dictionary data selection unit that selects only the iris code for registrationIndividualBody identification deviceWhen the dictionary data selection unit has a Hamming distance between each iris code generated by the feature data generation unit and an iris code generated from another image of the same individual is smaller than a threshold used at the time of matching An individual identification device characterized by selecting an arbitrary number of pieces of data having a higher matching performance as registration data, assuming that the larger the number, the higher the matching performance, and the higher the matching performance.
[0016]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, embodiments of the present invention will be described in detail using specific examples.
[0017]
<< Specific Example 1 >>
<Constitution>
FIG. 1 is a block diagram showing a specific example 1 of the individual identification apparatus of the present invention.
The apparatus shown in FIG. 1 includes a camera 1, an input image storage unit 2, an iris coordinate extraction unit 3, an iris coordinate variation unit 4, an iris code generation unit 5, a data storage unit (iris code storage unit) 6, and a dictionary data selection unit 7. The feature data generation unit 8 includes the iris coordinate extraction unit 3, the iris coordinate variation unit 4, and the iris code generation unit 5.
[0018]
The camera 1 shoots the eyes of a person or animal that registers an iris code, and the video acquired by the camera 1 is A / D converted by an A / D conversion unit (not shown) and is input as an input image storage unit. 2 is stored in the memory. The input image storage unit 2 includes a memory, a hard disk, and the like.
[0019]
In addition, shooting by the camera 1 is performed for a certain period of time while artificially changing the brightness and shooting angle, and a plurality of frames having good image quality are recorded among the videos shot during that time.
[0020]
The feature data generation unit 8 generates iris codes for all input images recorded in the input image storage unit 2. That is, the iris coordinate extraction unit 3 performs a coordinate system setting for generating an iris code. The extraction of the coordinate system is configured to be automatically performed on the input image using an image processing technique. Here, the iris coordinate system is the value of parameters such as the center coordinates and radius when a figure such as a circle or ellipse is applied to the outer contour of the pupil or the outer contour of the iris. By doing so, different iris codes are generated even for the same image. Moreover, even if the images are selected from the same individual, different iris codes are generated between the two images even if the iris coordinate system is accurately set due to differences in imaging conditions and the like.
[0021]
The iris coordinate changing unit 4 has a function of changing each parameter of the iris coordinate system extracted by the iris coordinate extracting unit 3 within a predetermined range. The iris code generating unit 5 generates an iris code for the parameter changed by the iris coordinate changing unit 4 and adds both the input image and the changed parameter information to the data storage unit 6. . Further, the variation of the iris coordinates and the generation of the iris code are repeated until a predetermined type of iris code is obtained.
[0022]
The dictionary data selection unit 7 is a functional unit that selects an iris code to be recorded for registration from the iris code stored in the data storage unit 6, and details will be described in the operation section.
[0023]
In the above configuration, each function of the iris coordinate extraction unit 3, the iris coordinate variation unit 4, the iris code generation unit 5, and the dictionary data selection unit 7 includes software corresponding to each process, and a processor and memory for executing the software. The input image storage unit 2 and the data storage unit 6 are composed of a storage device such as a semiconductor memory or a hard disk device.
[0024]
<Operation>
Here, a case will be described in which an object to be registered and verified is a person, and an iris code is generated by applying a circle to the contour of the pupil and the contour outside the iris.
[0025]
FIG. 2 is a flowchart showing operations from video input to iris code generation to register an iris code.
[0026]
For the eyes of the person to be registered, operations such as a change in illumination that can occur at the time of collation, a change in the relative position between the camera and the person, and movement of the eyes are performed, and images in all these states are recorded (step S101). Among the acquired videos, good images from various scenes are stored as digital images by A / D conversion on the memory of the input image storage unit 2 or on the hard disk (step S102).
[0027]
Therefore, a plurality of image files are generated for one eye. The video may be temporarily recorded on a video tape or the like, or may be converted into a still image on the spot and recorded on a hard disk or a memory.
[0028]
Next, the iris coordinate extraction unit 3 performs processing such as edge extraction on each of the images of the plurality of frames recorded in the input image storage unit 2, applies a circle to the outline of the pupil, and sets the center coordinates to Cp (Xp , Yp), the radius is Rp (step S104), and this is called a pupil circle.
[0029]
Similarly, a circle is also applied to the outer contour of the iris, the center coordinate is Ci (Xi, Yi), the radius is Ri (step S105), and this is called the iris circle. For fitting these circles, for example, it is possible to use a conventional technique such as that described in the conventional literature, and therefore detailed description thereof is omitted here.
[0030]
Each parameter of the circle extracted in step S104 and step S105 is a parameter that serves as a reference for generating an iris code by the following processing, and is referred to as a reference parameter.
[0031]
FIG. 3 is an explanatory diagram of an eye image and a coded image.
FIG. 3A is an explanatory diagram in the case where a circle is applied to the contours of the pupil and the iris. In order to simplify the description, the centers of the pupil circle and the iris circle are the same (Xp = Xi, Yp). = Yi), but in order to correct image distortion due to the positional relationship between the eyes and the camera 1 at the time of shooting, a method of handling these separately may be used.
[0032]
Steps S106 to S109 are parts for generating an iris code for the input image. In consideration of the fitting error of the pupil circle and the iris circle, the standard parameters of these circles are changed by a constant step width to obtain new parameters (step S106). This parameter is called a variation parameter.
[0033]
In this step S106, it is assumed that only the radii Rp and Ri are varied, and the values of the variation parameters are represented as rp (= Rp + Δrp) and ri (= Ri + Δri), respectively. In addition, each value of the center coordinates Cp and Ci can be changed.
[0034]
In step S107, an intermediate image (referred to as a coded image) for generating an iris code is generated based on the variation parameter. FIG. 3B illustrates an example of a method for generating a coded image. Since the upper part of the iris is hidden by eyelids or eyelashes, the range from π−θ to 2π + θ is coded as shown.
[0035]
Coded images are generated so that a, b, c, and d in FIG. 3A correspond to a, b, c, and d in FIG. Specifically, if an arbitrary point p (xp, yp) on the coded image corresponds to a point q (xq, yq) on the input image (FIG. 3A), the relationship between the point p and the point q Is represented by the following equation.
[0036]
The point q (xq, yq) is a polar coordinate system having the center Cp of the pupil circle as the origin, and expressed as q (rq, θq) (the distance from Cp is rq, and the angle with the horizontal axis is θq). Then
xp = w × (θq− (π−θ)) / (π + 2θ)
yp = h * (rq-rp) / (ri-rp)
[0037]
In step S108, an iris code is generated by applying a bandpass filter to the coded image generated in step S107. Since a known technique can be applied to the iris code, a detailed description thereof is omitted here. However, since the iris code depends on the change in the density of the coded image, a plurality of different iris codes can be obtained from the same input image by changing these parameters.
[0038]
Steps S106 to S109 are repeated until the standard parameter variation is completed within the determined range. If the number of repetitions is N, N types of iris codes are generated from one input image, and are recorded on the memory or disk as dictionary candidate data together with image file information and fluctuating parameter information. (Step S110, Step S103).
[0039]
FIG. 4 is an explanatory diagram showing the contents of dictionary candidate data.
Assume that an M-frame image is stored as a file for one eye in steps S101 and S102 in FIG. In steps S106 to S109, it is assumed that Δrp and Δri are each changed by −0.1 to 0.1 with a step width of 0.1. Thus, 3 × 3 = 9 codes are generated from one frame image. These nine codes are collectively referred to as a code set CD (m), where m is an image number.
[0040]
Further, a code in which Δrp and Δri are both 0 in the code set (an iris code generated without changing the parameter extracted in step S106) is referred to as a standard code and is represented as CDs (m). Therefore, M code sets (M × 9 codes) are generated for one eye. The n (n = 1 to N, N = 9) th code for the m (m = 1 to M) th still image is represented as C (m, n). N (= 9) varies depending on the variation width and the step width of Δrp and Δri.
[0041]
FIG. 5 is a flowchart of processing for obtaining a matching frequency for selecting registered data from dictionary candidate data.
That is, first, for all codes in the dictionary candidate data, the code collation frequency Q (m, n) = {D, AVE} (m = 1 to M, n = 1 to N, D is the number of collations, AVE calculates the average Hamming distance).
[0042]
For codes C (m, n) (n = 1 to N) in a certain code set CD (m) (m = 1 to M) (steps S201 to S204, step S215), other than CD (m) The Hamming distance HD (L) with the standard code CDs (L) (L = 1 to M, L ≠ m) (steps S205 to 207, step S212) of the code set is calculated (step S208), and HD (L) is calculated. If it is below the threshold value Ta for distinguishing the person from others at the time of collation (step S210), the collation number D is increased by 1 (step S211).
[0043]
When the calculation of the Hamming distance and the comparison with the threshold Ta are completed for all L, the average AVE of the Hamming distance is calculated (Step S209, Step S213), and D and AVE are compared with the code C (m, n). Register as Q (m, n) = {D, AVE} (step S214).
[0044]
Therefore, D is a number that can be collated when standard codes CDs (L) of other code sets are input, with C (m, n) as a dictionary.
[0045]
The verification frequency is calculated for all codes in the code set (steps S215 and S204) and for all code sets (steps S216 and S201). It should be noted that the higher the number of verifications and the smaller the average Hamming distance if the number of verifications is the same, the higher the verification frequency.
[0046]
For the calculation of the matching frequency, not only codes in a certain code set CD (m) and standard codes of other code sets CD (L) but also all codes included in CD (L) may be used. Good.
[0047]
Next, processing for selecting registration data based on the collation frequency Q (m, n) = {D, AVE} obtained by the above-mentioned collation frequency calculation processing and registering it in the dictionary set will be described.
[0048]
FIG. 6 is a flowchart of the registration data selection process.
A dictionary set is for registering multiple codes for one eye, taking into account differences in conditions such as brightness and camera position at the time of data acquisition, and errors in fitting iris coordinates. If the number of codes to be registered is RM, the dictionary set is expressed as CM (r) (r = 1 to RM).
[0049]
First, the codes in the code sets are sorted in descending order of collation frequency. The sorted code is expressed as CS (m, n) (m = 1 to M, n = 1 to N) (step S301). Here, for the same m, it is assumed that the smaller the number of n, the higher the matching frequency. Further, the code set including the code C (mm, nn) having the highest collation frequency among all dictionary candidate data is defined as CD (mm), and the first dictionary with the C (mm, nn) as the first registered data. Register in the set CM (1) (step S302).
[0050]
K = mm + 1 and r = 2 (step S303). K is a code set number. If K> M, K = 1 is set (step S304). Among the codes not registered in the dictionary set in the code set CD (K), the code CS (K, n) having the highest matching frequency and all codes CM (r in the already registered dictionary set ) To obtain the minimum value. If this minimum value is larger than the threshold value Tr for registration (step S307), this code is registered in the dictionary set CM (r), and the value of K and the value of r are each increased by 1 (step S308).
[0051]
As a result, when each code in the dictionary set is used as a dictionary, it can be compared with other input data at a high rate, and the other codes in the dictionary set are different from the code space of the individual (its A set of all codes that can be generated from an individual).
[0052]
FIG. 7 is an explanatory diagram showing the relationship between the code CM (r) and the code space.
The state shown in the figure means that the codes in the dictionary set are appropriately scattered in the code space, and if this is used, the fluctuation of the iris code due to the change of the environment at the time of collation can be covered.
[0053]
In step S307 in FIG. 6, if the minimum value of the Hamming distance is smaller than Tr, the same processing is performed for the next highest matching frequency in the code set CD (K) (steps S309, S306, and S307).
[0054]
As a result of performing the processes of steps S309, S306, and S307 for all codes in the code set CD (K), if there is no code satisfying the condition in CD (K), K is incremented by 1 (step S311). Return to step S304. However, if there is no code satisfying the condition of step S307 for all codes (all codes in the dictionary candidate data) of all code sets of K = 1 to M (step S310), Tr is a predetermined value. Decrease by dr (relax the condition of step S307) (step S312), and return to step S304. The above steps S304 to S312 are repeated until the number of codes in the dictionary set reaches RM (until r> RM in step S305).
[0055]
If the result of the above processing for all the code sets CD (K) is less than the dictionary set CM (RM), the dictionary data selection unit 7 determines that registration has failed and displays a message indicating this. The information is output such as displaying. That is, in such a case, there are few good images, such as an image obtained by the camera 1 being unclear or an inappropriate lighting condition. Therefore, the operator can recognize the registration failure by viewing such a message.
[0056]
<effect>
As described above, according to the first specific example, a plurality of iris codes corresponding to the eye images of a plurality of different scenes of an individual, and an iris code having a pseudo extraction error generated from each of these iris codes The number of iris codes whose Hamming distance between a certain iris code and other iris codes is smaller than a predetermined value is selected from among them, and this is selected as an iris code for registration in the same individual. When collating eyes, even if image distortion occurs due to environmental conditions during collation or positional relationship with the camera, there is a high possibility that collation is possible because a plurality of codes are registered.
[0057]
In addition, since the iris code hamming distance is registered more than a certain distance, the iris code is registered, so the code space generated from that eye can be covered with the determined number, so the number of registration dictionaries for one eye Can be limited, and the amount of data can be reduced.
[0058]
<< Specific Example 2 >>
Specific example 2 is an example in which collation is performed using the registration dictionary of specific example 1.
[0059]
<Constitution>
FIG. 8 is a configuration diagram of the second specific example.
The apparatus shown in the figure includes a camera 11, an iris coordinate extraction unit 12, an iris code generation unit 13, a verification unit (iris code comparison unit) 14, a verification result output unit 15, and a registration database 16, and includes an iris coordinate extraction unit 12 and an iris code. The generation unit 13 constitutes a feature data generation unit 17.
[0060]
The camera 11 shoots the eyes of a person or animal to be verified, and the video acquired by the camera 11 is A / D converted by an A / D converter (not shown), and is not shown in a memory or hard disk. And stored in the iris coordinate extraction unit 12.
[0061]
The iris coordinate extraction unit 12 is a functional unit that performs a coordinate system setting for generating an iris code, and this coordinate system extraction is configured to be automatically performed on an input image using an image processing technique. Yes. The iris code generation unit 13 has a function of calculating an iris code using the coordinate parameters extracted by the iris coordinate extraction unit 12. That is, the feature data generation unit 17 has a function of generating an iris code from the input image.
[0062]
The collation unit 14 has a function of calculating the degree of difference between the iris code calculated by the iris code generation unit 13 and the dictionary data of the registration database 16 and sending it to the collation result output unit 15. The collation result output unit 15 is a functional unit that outputs the collation result calculated by the collation unit 14.
[0063]
The registration database 16 is a database in which iris codes, which are dictionary data for collation, are registered, and this is generated by the apparatus of the first specific example.
[0064]
In the above configuration, each function of the iris coordinate extraction unit 12, the iris code generation unit 13, and the collation unit 14 includes software corresponding to each process and a processor or memory for executing the software. And hard disk device. The collation result output unit 15 includes a display, a printer, a network, or the like.
[0065]
<Operation>
FIG. 9 is a flowchart illustrating the operation of the second specific example.
Here, a specific example in the personal identification device, that is, a case where the identification target is a person will be described.
[0066]
Prior to the description of FIG. 9, the contents of the registration database 16 in FIG. 8 will be described.
If the number of persons registered in the database is NR, the database consists of NR dictionary sets. It is assumed that the dictionary set is the same as that created in the above-described specific example 1, but is sorted in the order of the matching frequency in the dictionary set. That is, if a code included in a certain dictionary set CM (r) is expressed as DC (r, n) (r = 1 to NR, n = 1 to ND (r)),
[0067]
Q (r, n) ≧ Q (r, n + 1)
Q (r, n) is the matching frequency of DC (r, n)
It becomes. However, ND (r) is the number of codes included in the dictionary set CM (r), and the number may be different depending on the dictionary set.
[0068]
In FIG. 9, when an image of the human eye is input to the iris coordinate extraction unit 12 (step S401), the iris coordinate extraction unit 12 creates an iris code IC after extracting the iris coordinates (step S402). Next, the minimum value holder MD (r) (r = 1 to NR) is initialized with a sufficiently large value (step S403). The minimum value holder MD (r) stores the minimum value of the Hamming distance between the code included in each dictionary set CM (r) and the input iris code.
[0069]
In step S404, the number of repetitions cr, the function R (cr), the number r of dictionary sets, and the variable n are initialized. Here, the number of repetitions cr (cr = 1 to NC) counts the number of repetitions of step S405 (step S412), and NC is an integer smaller than the number of codes included in each dictionary set. Each time this is repeated, the dictionary set CM (r) is sorted in ascending order of minimum value holder MD (r) (step S413). As a result of the sorting, the contents are changed so that the number of the dictionary set CM (r) can be matched with the included code and the minimum value holder MD (r).
[0070]
The function R (cr) is a function that determines how many dictionary sets are used as codes to be compared with the Hamming distance for the number of repetitions cr. This is a monotonically decreasing function. When cr = 1, R (cr) = NR and the minimum value is 1. R (cr) = NR means that all dictionary sets are to be searched. For example, if R (cr) = n (n <NR), the minimum value holder in the dictionary set. This means that the top n dictionary sets having a small MD (r) value are to be searched. Therefore, when R (cr) = 1, the dictionary set to be searched is narrowed down to one. R (cr) is, for example, the following formula:
[0071]
R (1) = NR
R (cr) = R (cr-1) / 2 cr> 2
if R (cr) <1 then R (cr) = 1
Function. This is a function that reduces the number of dictionary sets to be compared to half each time it is repeated.
[0072]
N (cr) in step S406 is a function that determines how many codes in the registration set are used for the Hamming distance calculation with respect to the number of repetitions cr. The variable used for checking N (cr) is n (step S406), and this n is 1 every time the steps S407, S408, S409, S410, and step S411 are repeated after being initialized in step S404. It increases gradually. That is, in step S407, the processing from step S408 to step S411 is repeated until r> R (cr). When r> R (cr), n is incremented by 1 (step S415), and step S406 is performed. Return to. Since the order of codes in the registration set is not rearranged, the code once calculated for the Hamming distance is not used for the calculation of the Hamming distance thereafter.
[0073]
In step S408, the Hamming distance HD between the input iris code IC and the nth code DC (r, n) in the rth registered set CM (r) is calculated. If the hamming distance HD is smaller than the threshold value Ta for distinguishing the person from others (step S409), the person is the same person as the person in the dictionary set CM (r), and the result is output (step S416).
[0074]
If the condition of step S409 is not satisfied, the content of the minimum value holder MD (r) corresponding to the dictionary set CM (r) is updated (step S410), the value of r is incremented by one (step S411), and step S407 is performed. Return to.
[0075]
When all the repetitions are completed in step S405, the result is output assuming that no person matches the input iris code (step S414).
[0076]
Next, the collation operation will be further described using an example.
FIG. 10 is an explanatory diagram illustrating an example of a repeated portion in FIG.
Here, the number of registered persons is NR = 8, the number of codes in each dictionary set is ND (r) = 7 (r = 1 to NR, all dictionary sets have the same number of codes), The number of repetitions NC = 4. The number of codes to be compared for each repetition is
[0077]
N (1) = 1
N (2) = 3
N (3) = 5
N (4) = ND (max) = 7
R (cr) is a function of reducing the number of dictionary sets to half each time the above expression is repeated.
[0078]
In the first iteration, the first code in all dictionary sets is used for comparison of the Hamming distance with the input iris code IC. As a result, when collation cannot be performed, CM (1) to CM (8) are sorted in the order of the smallest difference. Based on the expression of R (cr), the dictionary set that proceeds to the second iteration is the top four of the sorted dictionary sets.
[0079]
The contents of CM (1) to CM (4) at the second iteration are different from CM (1) to CM (4) at the first iteration. For example, in the first iteration, the minimum difference is CM (3) <CM (6) <CM (2) <CM (5) <CM (1) <CM (4) <CM (8) <CM (7). , It is CM (3), CM (6), CM (2), and CM (5) that proceeds to the second iteration, and the numbers 1 to 4 are reassigned to these. Yes.
[0080]
In the second iteration, since N (2) = 3, up to the third code in each dictionary set is used to calculate the Hamming distance. However, since the first code has already been calculated, only the second and third codes are to be compared here. When the second iteration is completed, CM (1) to CM (4) are sorted in ascending order of the smallest difference. Based on the equation of R (cr), the dictionary set that proceeds to the third iteration is the top two of the sorted dictionary sets.
[0081]
The contents of CM (1) and CM (2) at the time of the third repetition are the same as the description at the time of the second repetition, and the contents of CM (1) and CM (2) at the time of the second repetition are described above. ) Is different. For example, if the minimum difference in the second iteration is CM (2) <CM (4) <CM (1) <CM (3), it is CM (2), CM that proceeds to the third iteration. (4), and the numbers 1 and 2 are reassigned to these.
[0082]
In the third iteration, since N (3) = 5, codes up to the fifth in each dictionary set are used for the Hamming distance calculation. However, since the first to third codes have already been calculated, only the fourth and fifth codes are to be compared here. When the third iteration is completed, CM (1) and CM (2) are sorted in ascending order of minimum dissimilarity. Based on the equation of R (cr), the dictionary set that proceeds to the fourth iteration is the one with the smallest minimum degree of difference among the sorted dictionary sets.
[0083]
In the fourth iteration, since N (4) = ND (max) (maximum value of the number of codes registered in the dictionary set) = 7, all the codes in the last remaining dictionary set are to be compared.
[0084]
If there is a code determined to match the person being collated during the first to fourth iterations, the ID of the person recorded in the dictionary set to which the code belongs is output as a result. If there is no person who can be verified in the first to fourth repetitions, a result indicating that the person is not registered is output.
[0085]
<effect>
As described above, according to the second specific example, a plurality of iris codes are registered for one person, arranged in advance in the order of high possibility of collation, and collation is performed in that order. Is shortened. In addition, even if a large number of people are registered, the number of codes registered for each person is determined for each number of repetitions, and the search target is determined each time the repetition is performed. Since comparison is performed while narrowing down, it is possible to search each dictionary set without bias.
[0086]
In the specific examples 1 and 2 described above, an example of a person as an identification target has been described. In addition, the device is applicable not only to the iris code but also to other identification devices as long as the device recognizes based on the difference between the registered data and the input data. Further, although the hamming distance is used for the dissimilarity of the feature data, other distance scales such as Euclidean distance and Mahalanobis distance may be used as long as the dissimilarity is shown.
[Brief description of the drawings]
FIG. 1 is a block diagram showing a specific example 1 of an individual identification apparatus of the present invention.
FIG. 2 is a flowchart showing an operation from video input for registering an iris code to generation of an iris code.
FIG. 3 is an explanatory diagram of an eye image and a coded image.
FIG. 4 is an explanatory diagram showing the contents of dictionary candidate data.
FIG. 5 is a flowchart of processing for obtaining a matching frequency for selecting registered data from dictionary candidate data.
FIG. 6 is a flowchart of registered data selection processing.
FIG. 7 is an explanatory diagram showing a relationship between a code CM (r) and a code space.
FIG. 8 is a configuration diagram of a specific example 2;
FIG. 9 is a flowchart showing the operation of the specific example 2;
10 is an explanatory diagram showing an example of a repeated portion in FIG. 9. FIG.
[Explanation of symbols]
2 Input image storage
6 Data storage
7 Dictionary data selector
8, 17 Feature data generator
14 Verification part
16 Registration database

Claims (1)

In an individual identification device that generates an iris code from an eye image and performs individual identification based on the iris code,
An input image storage unit for storing eye images of a plurality of different scenes of one individual;
For each of the plurality of eye images stored in the input image storage unit, a feature data generation unit that artificially generates an iris code having an extraction error by changing the value of each parameter of the iris coordinate system;
For each iris code generated by the feature data generation unit, the average difference between the iris code obtained from an arbitrary image in the same individual and the iris code obtained from other images is smaller than a predetermined value. a number body recognition apparatus and a dictionary data selection unit to iris code for registration by selecting the iris code obtained from any of the image by a predetermined number,
The dictionary data selection unit, for each iris code generated by the feature data generation unit, the number when the Hamming distance between the iris code generated from another image of the same individual is smaller than the threshold used at the time of matching An individual identification apparatus characterized in that an arbitrary number of data having higher collation performance are selected as registration data on the assumption that the larger the number, the higher the collation performance is.

Images