KR102286455B1 - Method for generating fake iris using artificial intelligence, recording medium and device for performing the method - Google Patents

Abstract

A fake iris generating method using artificial intelligence comprises: a step of learning a process for generating an iris image from an edge image through CycleGAN by using an iris image and an edge image acquired from an iris recognizer as a data set; a step of quantizing an inputted iris image to convert an iris code having a real part and an imaginary part into an arbitrary positive value if the iris code is 1 and convert the iris code into an arbitrary negative value if the iris code is 0; a step of inverting the converted iris code through a formula of a one-dimensional Gabor filter expressed in a system of equations to reconstruct the iris code into an iris image; a step of converting the reconstructed rectangular iris image into an orthogonal coordinate system to generate a conic iris image; and a step of inserting the conic iris image into the iris image generated by learning through CycleGAN to generate a fake iris image. Accordingly, negation samples can be secured in iris research.

Description

Method for generating fake iris using artificial intelligence, recording medium and apparatus for performing the same

The present invention relates to a method for generating a forged iris using artificial intelligence, a recording medium and an apparatus for performing the same, and more particularly, to a counterfeit iris sample in a study of detecting a fake iris by supplementing the restoration vulnerability of Dogman's iris code. It's about acquiring the technology.

Biometrics is a technology that can verify the identity of an individual. It has many advantages over existing authentication systems based on passwords and ID cards by using personal information such as fingerprints, iris, and fingerprints to verify identity. Since biometric information cannot be copied or stolen, there is no risk of change or loss, so it is widely used in the security field.

However, there is a high risk of leakage because personal information used in biometrics cannot be changed. For this reason, biometric technologies currently used do not store and use personal biometric information as it is for security, but use the information as data in the form of a template. Currently, a method for reconstructing an image from such a template is being studied.

Fake iris detection research has been previously conducted using Purkinje Phenomenon or search in frequency space. However, in the course of these studies, the problem of the absence of counterfeit iris samples was caused.

Also, in the case of iris, various methods of recognition and counterfeit iris detection were studied in the iris recognition stage. However, studies and experiments have been conducted to identify counterfeit iris using printed output or artificial iris due to the absence of data on counterfeit iris.

Accordingly, there is a problem in that the existing fake iris image has a large difference from the real iris image or has a random iris code.

KR 10-2019-0135062 A KR 10-0787114 B1 JP 3598109 B2

J. Daugman, "High Confidence Visual Recognition of Persons by a Test of Statistical Independence", IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 15, no. 11, pp. 1148-1161, Nov. 1993 J. Daugman, "How Iris Recognition Works", IEEE Trans. on Circuits and Systems for Video Technology, Vol. 14, No. 1, pp. 21-30, Jan. 2004 S Minaee, A Abdolrashidi, "Iris-GAN: Learning to Generate Realistic Iris Images Using Convolutional GAN." arXiv preprint arXiv:1812.04822, 2018

Accordingly, it is an object of the present invention to provide a method of generating a fake iris using artificial intelligence in order to secure a fraudulent sample.

Another object of the present invention is to provide a recording medium in which a computer program for performing the method for generating a forged iris using artificial intelligence is recorded.

Another object of the present invention is to provide an apparatus for performing the method for generating a fake iris using the artificial intelligence.

In a method for generating a fake iris using artificial intelligence according to an embodiment for realizing the object of the present invention, an edge image through CycleGAN using an iris image and an edge image obtained from an iris recognizer as a data set learning a process of generating an iris image from quantizing the input iris image to convert an iris code having a real part and an imaginary part into an arbitrary positive value when the iris code is 1, and into an arbitrary negative value when the iris code is 0; reconstructing the iris code into an iris image by inversely calculating the converted iris code through a formula of a 1D Gabor filter expressed as a simultaneous equation; generating a circular iris image by converting the reconstructed rectangular iris image into a Cartesian coordinate system; and generating a fake iris image by inserting the circular-shaped iris image into the iris image generated by learning through the cyclegen.

In an embodiment of the present invention, the learning of the process of generating an iris image from an edge image through the CycleGAN includes: acquiring an iris image and an edge image from an iris recognizer as a data set; normalizing the acquired data set; Learning the normalized data set with a U-Net structure of Cyclogen, learning by bidirectional encoding converting an iris image to an iris edge image and an iris edge image to an iris image; and generating an iris image from the edge image through the learned encoding.

In an embodiment of the present invention, the step of reconstructing the iris code into an iris image comprises: calculating the iris code by dividing the iris code line by line in the vertical direction using a reconstruction estimation matrix in which the filter equations of the real part and the imaginary part are vertically combined; and synthesizing them in the separated order and outputting them as one image.

In an embodiment of the present invention, the step of quantizing the input iris image and converting the iris code having a real part and an imaginary part into an arbitrary positive value when 1 is 1 and into an arbitrary negative value when 0 includes the steps of: When it is 1, it can be converted to 1, and when the iris code is 0, it can be converted to -1.

In an embodiment of the present invention, in the generating of the circular iris image, the empty space in the middle may be replaced with a pixel value of 0.

In an embodiment of the present invention, the method for generating a fake iris using artificial intelligence may further include quantizing the input iris image.

In an embodiment of the present invention, the step of quantizing the input iris image may include separating the iris region from the iris image recognized by the iris recognizer; normalizing the iris region through polar coordinate transformation; extracting a feature having a real part value and an imaginary part value using a 1D Gabor filter; and generating a quantized iris code in which the image of the real value and the image of the imaginary value are intersected in a horizontal direction.

In a computer-readable storage medium according to an embodiment for realizing another object of the present invention, a computer program for performing the method for generating a fake iris using artificial intelligence is recorded.

An apparatus for generating a fake iris using artificial intelligence according to an embodiment of the present invention for realizing another object of the present invention uses the iris image and edge image obtained from the iris recognizer as a data set through CycleGAN. a learning unit for learning a process of generating an iris image from an edge image; a code conversion unit that quantizes the input iris image and converts the iris code having a real part and an imaginary part into an arbitrary positive value when the iris code is 1, and into an arbitrary negative value when the iris code is 0; an image restoration unit for reconstructing the iris code into an iris image by inversely calculating the converted iris code through a formula of a 1D Gabor filter expressed as a simultaneous equation; a coordinate conversion unit that converts the reconstructed rectangular iris image into a Cartesian coordinate system to generate a circular iris image; and a fake iris generator generating a fake iris image by inserting the circular iris image into the iris image generated by learning through the cyclegen.

In an embodiment of the present invention, the image restoration unit separates and calculates the iris codes one by one in the vertical direction by using the restoration estimation matrix in which the filter equations of the real part and the imaginary part are combined in the vertical direction, and synthesizes them in the separated order to form one It can be output as a video.

According to this method of generating a fake iris using artificial intelligence, an iris code is arbitrarily inserted in the step of generating a fake iris image to overcome the problem of the absence of a fake iris sample in a study for detecting a fake iris. Accordingly, through the present invention, it is possible to secure a negative sample necessary for iris research, and it is possible to increase the level and accuracy of a counterfeit iris detection study at the current point in time when iris recognition is important.

1 is a block diagram of an apparatus for generating a fake iris using artificial intelligence according to an embodiment of the present invention.
FIG. 2 is a conceptual diagram for explaining a learning process of the CycleGAN of FIG. 1 .
FIG. 3 is a conceptual diagram for explaining a learning result of the learning unit of FIG. 1 .
4 is a diagram for explaining a process of converting a circular iris image into a rectangle.
5 is an example of a normalized image, a real part feature, an imaginary part feature, and an iris code generated by the present invention.
6 is an example of an original image, real and imaginary part features, a reconstructed image, and an orthogonal coordinate system-converted iris image generated by the present invention.
7 is a statistical graph of the Hamming distance calculated by comparing the iris of others in order to verify the performance of the present invention.
8 is a statistical graph of the Hamming distance calculated by comparing the iris of the same person in order to verify the performance of the present invention.
9 is a statistical graph of a Hamming distance calculated by comparing an original image and a reconstructed image in order to verify the performance of the present invention.
10 is a flowchart of a method for generating a fake iris using artificial intelligence according to an embodiment of the present invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS [0010] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS [0010] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS [0023] Reference is made to the accompanying drawings, which show by way of illustration specific embodiments in which the present invention may be practiced. These embodiments are described in sufficient detail to enable those skilled in the art to practice the present invention. It should be understood that the various embodiments of the present invention are different but need not be mutually exclusive. For example, certain shapes, structures, and characteristics described herein with respect to one embodiment may be embodied in other embodiments without departing from the spirit and scope of the invention. In addition, it should be understood that the location or arrangement of individual components within each disclosed embodiment may be changed without departing from the spirit and scope of the present invention. Accordingly, the detailed description set forth below is not intended to be taken in a limiting sense, and the scope of the invention, if properly described, is limited only by the appended claims, along with all scope equivalents to those claimed. Like reference numerals in the drawings refer to the same or similar functions throughout the various aspects.

Hereinafter, preferred embodiments of the present invention will be described in more detail with reference to the drawings.

1 is a block diagram of an apparatus for generating a fake iris using artificial intelligence according to an embodiment of the present invention.

The device for generating a fake iris (hereinafter referred to as device 10) using artificial intelligence according to the present invention relates to a technique for generating a fake iris through a GAN model or generating an iris image by inversely calculating an iris code to obtain a sample of a fake iris .

In the present invention, the restoration vulnerability of the iris code is analyzed by analyzing the process of creating the iris code from the iris image and calculating it in reverse.

To this end, we propose our own iris recognizer based on Dog's own iris recognition method. The iris recognizer proposed in the present invention performs a 1D Gabor filter by segmenting and normalizing only the iris region in the iris image. Finally, the iris code is formed by quantizing the filtered image.

In the present invention, it is assumed that the parameters of the 1D Gabor filter are known, the filter is calculated inversely to restore the iris image from the iris code, and the restored image and the original image are matched to verify the restoration vulnerability of the iris code based on the authentication result. do.

Referring to FIG. 1 , an apparatus 10 according to the present invention includes a learning unit 100 , a code converting unit 310 , an image restoration unit 330 , a coordinate converting unit 350 , and a fake iris generating unit 370 . include The learning unit 100 may include a CycleGAN 110 and a bidirectional encoder 130 .

In the device 10 of the present invention, software (application) for performing fake iris generation using artificial intelligence may be installed and executed, and the learning unit 100, the code conversion unit 310, and the image restoration The configuration of the unit 330 , the coordinate transformation unit 350 , and the fake iris generating unit 370 is to be controlled by software for generating a fake iris using the artificial intelligence executed in the device 10 . can

The device 10 may be a separate terminal or a module of the terminal. In addition, the configuration of the learning unit 100, the code converting unit 310, the image restoration unit 330, the coordinate converting unit 350, and the forged iris generating unit 370 is formed as an integrated module, It may consist of one or more modules. However, on the contrary, each configuration may be formed of a separate module.

The device 10 may be movable or stationary. The apparatus 10 may be in the form of a server or an engine, and may be a device, an application, a terminal, a user equipment (UE), a mobile station (MS), or a wireless device. (wireless device), may be called other terms such as a handheld device (handheld device).

The device 10 may execute or manufacture various software based on an operating system (OS), that is, the system. The operating system is a system program for software to use the hardware of the device, and is a mobile computer operating system such as Android OS, iOS, Windows Mobile OS, Bada OS, Symbian OS, Blackberry OS, and Windows series, Linux series, Unix series, It can include all computer operating systems such as MAC, AIX, and HP-UX.

The learning unit 100 learns a process of generating an iris image from the edge image through CycleGAN using the iris image and the edge image obtained from the iris recognizer as a data set.

FIG. 2 is a conceptual diagram for explaining a learning process of the CycleGAN of FIG. 1 . FIG. 3 is a conceptual diagram for explaining a learning result of the learning unit of FIG. 1 .

The CycleGAN 110 of the learning unit 100 sets the iris image and the edge image obtained from the iris recognizer as a data set, and learns through the U-Net structure after the normalization of the set data set.

The two-way encoder 130 of the learning unit 100 is a two-way auto encoder learner that converts an iris image into an iris edge image and an iris edge image into an iris image. The learning unit 100 generates an iris image from the edge image through the learned Auto Encoder.

GAN (Generative Adversarial Network) is one of neural network-based generative models, and iteratively learns and derives an optimal solution through a generator and a discriminator. Producers generate fake images by inputting random noise generated through probability distribution. The discriminator determines whether the input data is a generated image or an original image. The objective function of the GAN is shown in Equation 1 below.

[Equation 1]

A fake iris image can be generated through such a neural network-based production model, and even when there is no fake iris sample as in the prior art, the generated fake iris image can be used to solve the problem.

CycleGAN is one of the models of GAN, and Fig. 2 shows the model structure of CycleGAN and uses an unpaired training set. Using an unpaired training set removes the restrictions on using an existing Parade training set. In addition, CycleGAN has the advantage of maintaining more detailed images generated using U-Net. Equation 2 below is a loss function of CycleGAN.

[Equation 2]

CycleGAN uses the bidirectional loss function to generate images, so the image quality is improved. This prevents the noise of the probability distribution from being biased due to iterative learning of producers and discriminators.

Referring to FIG. 3 , it can be confirmed that the probability distributions of the producer and the discriminator are consistent with the noise bias.

The present invention proposes an iris recognizer based on Dogman's iris recognition method in order to analyze the restoration vulnerability of the iris code. The iris recognizer consists of the steps of segmenting, normalizing, feature extraction, and authenticating the iris region. In order to extract the features of the iris from the iris image, the iris region must first be segmented.

In order to segment only the iris region, it is necessary to find the inner boundary (the pupil boundary) and the outer boundary of the iris. In this case, in the case of the pupil boundary, which is the inner boundary of the iris, the contrast value between the pupil and the iris is certain because the pupil has a characteristic that most of the pupil has a low pixel value, and thus it can be easily obtained through a method using the Hough transform.

On the other hand, in the case of the outer boundary, the Hough transformation method is difficult because the contrast value between the iris and the sclera is not clear. Therefore, in the case of the outer boundary, the Integro-differential operator as in Equation 3 below is used.

[Equation 3]

This operator is an operator that finds the inner and outer boundaries of the iris by using the contrast difference in the iris image using the characteristic of the iris, which is always brighter than the pupil and darker than the sclera.

Here, ( x ₀ , y ₀ ) represents the center of the iris, and r represents the radius. The place with the greatest difference by changing each value to make the boundaries of the circle shape, finding the sum of the pixel values ( I(x,y) ) located around the circle, and then finding the difference between the values that change according to the radius r is the boundary of the iris.

The size of the divided iris is different for each individual, and even for the same person, the size of the iris image may be different depending on the distance between the camera and the user in the process of obtaining the iris image, and the size of the pupil may vary according to the surrounding illumination. For correct authentication, the divided iris should be normalized to a certain standard, and normalization is performed through polar coordinate system transformation. The coordinate system transformation is shown in Equation 4 below.

[Equation 4]

r is the distance between the inner and outer boundaries of the iris, and θ is the angle of the iris. Through the coordinate system transformation, the circular iris is spread out into a rectangular shape as shown in FIG. 4 .

After the image is normalized, the features of the iris are extracted using a 1D Gabor filter. The 1D Gabor filter is a cosine and sine function transformed by a Gaussian function, and is a filter based on the Fourier transform method. Equations 5 and 6 below are real and imaginary values, respectively. Calculate.

[Equation 5]

[Equation 6]

Here, σ is a value that determines the width of the filter kernel, and p is a value that determines the period. Equation 5 is a form of a cosine function transformed by a Gaussian function, and an image having a real value of an iris feature is acquired through a corresponding filter. Equation 6 is a sine function transformed by a Gaussian function and acquires an image having an imaginary part value.

After intersecting the obtained real-valued image and the imaginary-valued image in the horizontal direction, quantization is performed to finally generate an iris code. 5 is an example of a normalized image and an iris code. FIG. 5(a) is a normalized normalized image, FIG. 5(b) is a real part feature, FIG. 5(c) is an imaginary part feature, and FIG. 5(d) is an example of an iris code.

This type of code is generated for each input iris image, and whether or not the user is authenticated is determined through code comparison. A Hamming Distance is used for comparison of iris codes. The Hamming distance equation is the following equation (7).

[Equation 7]

X _j and Y _j are code values that are compared with each other. The XOR operation yields a value of 0 if the values are the same, 1 if they are different. The value of the Hamming distance is a numerical value that quantifies the degree of similarity between the two codes, and authentication is determined based on a certain threshold.

The code converter 310 quantizes the input iris image and converts the iris code having a real part and an imaginary part into an arbitrary positive value when the iris code is 1, and into an arbitrary negative value when it is 0.

The iris code was generated by extracting the features of the iris using a 1D Gabor filter. If this feature extraction process is expressed mathematically, it can be expressed as a single matrix. A method of estimating the original iris image by calculating the inverse of this matrix can be restored. In this case, it is assumed that 1D knows the filter value of the burr filter.

The iris code is quantized and consists of values of 0 or 1. In order to mathematically calculate and restore this value, it is expressed in the form of positive and negative numbers, which are the values before quantization. A value of 1 is an arbitrary positive value, and in the case of 0, it is calculated by changing it to an arbitrary negative value.

In one embodiment, the code value 1 is left as 1, and 0 is changed to -1. Since the filter value of the 1D Gabor filter and the value of the iris code are known, calculations to estimate the original image of the iris can be performed. Equations 8 and 9 below express this process as an equation.

[Equation 8]

[Equation 9]

In the above equation, A is the filter value, X is the original image, and B is the iris code. Here, since A and B are known values, the original image can be estimated by calculation using the inverse matrix as in Equation 9. In order to calculate Equations (8) and (9), the process of creating an iris code must first be expressed as a matrix.

The image restoration unit 330 inversely calculates the converted iris code through a 1D Gabor filter equation expressed by a simultaneous equation to restore the iris code to an iris image.

The filter expression is expressed by solving a system of equations, and this is expressed as a matrix. If the filter process of the one-dimensional Gabor filter is mathematically expressed, it can be expressed as a system of equations. Because it is a one-dimensional filter, image restoration proceeds in order from the top row of the image. Equation 10 below is an example of expressing the filter process as a system of equations based on one line in the image.

[Equation 10]

In the equation, a ₁ to a _x are the Gabor filter values, x ₁ to x _y are the restoration estimates, and b ₁ to b _z are the iris code values. If Equation 10 is expressed as a determinant, Equation 11 is as follows.

[Equation 11]

In Equation 11, since the filter value and the iris code value are known, a restoration estimate can be obtained by multiplying both sides by the inverse matrix of the filter value. The process is shown in Equation 12 below.

[Equation 12]

In this way, a restoration estimate can be obtained. However, in the case of the 1D Gabor filter, the filter formula consists of a real part filter and an imaginary part filter. Since the iris code is a result of using these two filters, in the case of restoration, if the two filters are calculated respectively, the estimated original image is calculated as two results, so that restoration into one image is impossible.

Therefore, in order to obtain one reconstructed image that simultaneously satisfies the filtering process of the real part and the imaginary part, the form of the formula must be transformed to estimate a value that simultaneously satisfies the two filter formulas in the reconstruction process. This form is shown in Equation 13 below.

[Equation 13]

Equation 13 is expressed as an inverse formula to obtain a restored value, as shown in Equation 14 below.

[Equation 14]

The above restoration process is a method of calculating the calculation process of the 1D filter by mathematically expressing it. The iris code is calculated by dividing the iris code by one line in the vertical direction, and finally, one image is created by pasting them in the order in which they were separated. The entire restoration process is expressed in pseudo code as follows.

The filter value matrix A and the code value matrix B used in the pseudo code represent the filter value matrix and the code result matrix of the calculation process of Equation (13).

The coordinate converter 350 converts the reconstructed rectangular iris image into a Cartesian coordinate system to generate a circular iris image.

The restored image result is a rectangular image in the form of a normalized iris image as shown in FIG. 6(c). However, since the iris recognizer requires a circular iris image, it must be converted to the form before normalization by performing Cartesian coordinate system transformation. At this time, when converted to the Cartesian coordinate system, since it comes out in a ring shape, the empty space in the middle is filled with a pixel value of 0 so as to be similar to the pixel value of the pupil.

Fig. 6(a) is an original image, Fig. 6(b) is a real code and an imaginary part code, Fig. 6(c) is a reconstructed image, and Fig. 6(d) is an example of an iris image converted to a Cartesian coordinate system.

The fake iris generator 370 generates a fake iris image by inserting the circular iris image into the iris image generated by learning through the cyclegen.

Accordingly, it is possible to overcome the problem of the absence of a fake iris sample in a fake iris detection study by arbitrarily inserting an iris code in the fake iris image generation step. Through the present invention, it is possible to secure a negative sample necessary for iris research, and it is possible to increase the level and accuracy of a counterfeit iris detection study at the present time when iris recognition is important.

Hereinafter, the performance evaluation of the recognizer manufactured for the reliability of the restoration method proposed in the present invention and the restoration experiment result will be described. The performance evaluation of the restoration experiment is performed with the recognition rate of the restored image and the original image based on the FAR (False acceptance rate) of 1%, 0.1%, and 0.01% using the manufactured recognizer.

The experiment was implemented using Visual Studio 2015 on a PC equipped with Intel Core i7-6700 3.40GHz CPU and 8.0GB memory. The library used Opencv 3.20. The database used iris images of 120 people from the CASIA Database. Experiment with a total of 480 iris images using 4 iris per person.

The recognition performance evaluation method calculates the Hamming distance threshold based on the FAR 1%, 0.1%, and 0.01% standards by experimenting with same-person comparison and comparison with others using the iris image of the database and evaluates the recognition performance. 7 is a statistical graph of a Hamming distance calculated by comparing the iris of another person.

For the number of comparisons, a total of 114,240 (119*4*120*4/2) comparisons were made using 480 iris images. Table 1 below is the Hamming distance threshold based on FAR 1%, 0.1%, and 0.01%.

[Table 1]

8 is a statistical graph of a Hamming distance calculated by comparing the iris of the same person.

It is the result of comparing 720(4*3/2*120) times in total. The maximum value of the Hamming distance from the comparison of the iris of the same person is 0.3638. This value showed a result lower than the value of 0.3751, which is the threshold based on FAR 0.01%, resulting in a recognition success rate of 100% in case of recognition of the same person.

In addition, restoration vulnerability was evaluated using the iris recognizer. 480 images were generated as iris codes, and the generated iris codes were restored by the method proposed in the present invention. Then, the original image and the reconstructed image were compared 1:1 and recognized using the manufactured recognizer. 9 is a hamming distance graph calculated by comparing the original image and the reconstructed image.

The performance of restoration was based on the FAR standard, and Table 2 below shows the results.

[Table 2]

Dogman's iris code is a method that can be authenticated while protecting iris information, and is a method used by most of the currently commercially used recognizers. Currently, a technique for reconstructing an image using only the iris code information is being studied. In the present invention, a method of reconstructing an image from such an iris code by an inverse computation method using a matrix is proposed.

Dogman's iris code generation process and recognition method were analyzed, and the iris image was restored using only the iris code by calculating it in the reverse direction. The reconstructed image is different from the general iris image with the naked eye, but when this image was matched with the original image in the recognizer, it showed a high recognition rate.

10 is a flowchart of a method for generating a fake iris using artificial intelligence according to an embodiment of the present invention.

The method for generating a fake iris using artificial intelligence according to the present embodiment may be performed in substantially the same configuration as the device 10 of FIG. 1 . Accordingly, the same components as those of the device 10 of FIG. 1 are given the same reference numerals, and repeated descriptions are omitted.

In addition, the method for generating a forged iris using artificial intelligence according to the present embodiment may be executed by software (application) for generating a fake iris using artificial intelligence.

Referring to FIG. 10 , in the method for generating a fake iris using artificial intelligence according to the present embodiment, an iris image and an edge image obtained from an iris recognizer are used as a data set to generate an iris image from an edge image through CycleGAN. The process of generating is learned (step S10).

Learning the process of generating an iris image from an edge image through the CycleGAN includes: acquiring an iris image and an edge image from an iris recognizer as a data set; Normalizing the acquired data set; Learning the data set with U-Net structure cyclegen, learning by bidirectional encoding that converts iris image to iris edge image, and iris edge image to iris image, and learns iris image from edge image through the learned encoding comprising the steps of creating

The input iris image is quantized and converted into an arbitrary positive value when the iris code having a real part and an imaginary part is 1, and into an arbitrary negative value when it is 0 (step S20). For example, when the iris code is 1, it is converted into 1, and when the iris code is 0, it is converted into -1.

In the quantizing of the input iris image, the iris region is separated from the iris image recognized by the iris recognizer and normalized through polar coordinate transformation of the iris region. Then, a feature having a real value and an imaginary part is extracted using a 1D Gabor filter, and a quantized iris code is generated and quantized by intersecting the image of the real part and the image of the imaginary part in the horizontal direction. .

The iris code is restored to the iris image by inversely calculating the converted iris code through the formula of the 1D Gabor filter expressed by the simultaneous equation (step S30).

In the step of reconstructing the iris code into the iris image, the iris code is calculated by dividing the iris code line by line in the vertical direction using a reconstruction estimation matrix in which the filter equations of the real part and the imaginary part are combined in the vertical direction. Thereafter, they are synthesized in the separated order and output as one image.

The reconstructed rectangular iris image is converted into a rectangular coordinate system to generate a circular iris image (step S40). At this time, when converted to the Cartesian coordinate system, since it comes out in a ring shape, the empty space in the middle is filled with a pixel value of 0 so as to be similar to the pixel value of the pupil.

A fake iris image is generated by inserting the circular-shaped iris image into the iris image generated by learning through the cyclegen (step S50).

Accordingly, it is possible to overcome the problem of the absence of a fake iris sample in a fake iris detection study by arbitrarily inserting an iris code in the fake iris image generation step. Through the present invention, it is possible to secure a negative sample necessary for iris research, and it is possible to increase the level and accuracy of a counterfeit iris detection study at the present time when iris recognition is important.

Such a method for generating a fake iris using artificial intelligence may be implemented as an application or implemented in the form of program instructions that may be executed through various computer components and recorded in a computer-readable recording medium. The computer-readable recording medium may include program instructions, data files, data structures, etc. alone or in combination.

The program instructions recorded on the computer-readable recording medium are specially designed and configured for the present invention, and may be known and available to those skilled in the computer software field.

Examples of the computer-readable recording medium include a hard disk, a magnetic medium such as a floppy disk and a magnetic tape, an optical recording medium such as a CD-ROM, a DVD, and a magneto-optical medium such as a floppy disk. media), and hardware devices specially configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like.

Examples of program instructions include not only machine language codes such as those generated by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like. The hardware device may be configured to operate as one or more software modules to perform processing according to the present invention, and vice versa.

Although the above has been described with reference to the embodiments, those skilled in the art can variously modify and change the present invention within the scope without departing from the spirit and scope of the present invention described in the claims below You will understand.

Biometric authentication, such as fingerprints, veins, and iris, is rapidly expanding in types and fields. Such biometric authentication has the advantage of being more convenient than existing security technologies (eg, passwords and security cards), but security is important to that extent. The present invention can generate a fake iris image and can be used for research on the identification of fake iris, and this technology can be commercialized in the biometric technology market, such as the equipment business that acquires biometric images such as fingerprints and iris, and the field of providing biometric authentication solutions do.

The biometric technology market is projected to grow from $3.24 billion in 2016 to $12.22 billion (annual average of 20.8%) in 2023 (Research and Markets, August 2017). In addition, the fields of application of biometric technology are predominantly mobile, wearable, healthcare, payment, and access control. is rising

10: Device for generating fake iris using artificial intelligence
100: study department
110: CycleGAN
130: bidirectional encoder
310: code conversion unit
330: image restoration unit
350: coordinate conversion unit
370: fake iris generator

Claims (10)

learning a process of generating an iris image from the edge image through CycleGAN using the iris image and the edge image obtained from the iris recognizer as a data set;
quantizing the input iris image to convert an iris code having a real part and an imaginary part into an arbitrary positive value when the iris code is 1, and into an arbitrary negative value when the iris code is 0;
reconstructing the iris code into an iris image by inversely calculating the converted iris code through a formula of a 1D Gabor filter expressed as a simultaneous equation;
generating a circular iris image by converting the reconstructed rectangular iris image into a Cartesian coordinate system; and
generating a fake iris image by inserting the circular-shaped iris image into the iris image generated by learning through the cyclegen;
Further comprising the step of quantizing the input iris image,
Quantizing the input iris image includes:
Separating the iris region from the iris image recognized by the iris recognizer;
Separating the iris region comprises:
After calculating the sum of the pixel values located on the circumference of the circle created by changing the radius from the center of the iris, the difference between the values that change according to the radius is obtained. A method of generating a fake iris using artificial intelligence, characterized in that the regions are separated.
The method of claim 1 , wherein the step of learning the process of generating an iris image from an edge image through the CycleGAN comprises:
acquiring an iris image and an edge image from an iris recognizer as a data set;
normalizing the acquired data set;
Learning the normalized data set with the U-Net structure of Cyclogen, the iris image to the iris edge image, and learning by bidirectional encoding converting the iris edge image to the iris image; and
Generating an iris image from an edge image through the learned encoding; A method for generating a fake iris using artificial intelligence, comprising a.
The method of claim 1, wherein the step of restoring the iris code to an iris image comprises:
calculating an iris code by dividing the iris code line by line in the vertical direction using a reconstruction estimation matrix in which the filter equations of the real part and the imaginary part are vertically combined; and
A method of generating a fake iris using artificial intelligence, including; synthesizing in the separated order and outputting it as a single image.
The method of claim 1, wherein quantizing the input iris image and converting the iris code having a real part and an imaginary part into an arbitrary positive value when 1 is 1, and into an arbitrary negative value when 0, comprises:
When the iris code is 1, it is converted into 1, and when the iris code is 0, it is converted into -1.
The method of claim 1, wherein the generating of the circular iris image comprises:
A method of generating a fake iris using artificial intelligence in which the empty space in the middle is replaced with a pixel value of 0.
delete The method of claim 1, wherein quantizing the input iris image comprises:
normalizing the iris region through polar coordinate transformation;
extracting a feature having a real part value and an imaginary part value using a 1D Gabor filter; and
generating a quantized iris code obtained by intersecting the image of the real part and the image of the imaginary part in a horizontal direction;
A computer-readable storage medium in which a computer program for performing the method of generating a forged iris using artificial intelligence of claim 1 is recorded.
a learning unit configured to learn a process of generating an iris image from an edge image through CycleGAN using the iris image and the edge image obtained from the iris recognizer as a data set;
a code conversion unit that quantizes the input iris image and converts an iris code having a real part and an imaginary part into an arbitrary positive value when the iris code is 1, and into an arbitrary negative value when the iris code is 0;
an image restoration unit for reconstructing the iris code into an iris image by inversely calculating the converted iris code through a formula of a 1D Gabor filter expressed as a simultaneous equation;
a coordinate conversion unit that converts the reconstructed rectangular iris image into a Cartesian coordinate system to generate a circular iris image; and
and a fake iris generator to generate a fake iris image by inserting the circular iris image into the iris image generated by learning through the cyclegen.
The code conversion unit,
Separating the iris region from the iris image recognized by the iris recognizer,
After calculating the sum of the pixel values located on the circumference of the circle created by changing the radius from the center of the iris, the difference between the values that change according to the radius is obtained. A device for generating fake iris using artificial intelligence, characterized in that the regions are separated.
The method of claim 9, wherein the image restoration unit,
A fake iris using artificial intelligence that calculates the iris codes by dividing them vertically one by one using the reconstructed estimation matrix in which the filter formulas of the real and imaginary parts are merged vertically, and synthesizes them in the separated order and outputs them as a single image. generating device.

Images