KR101862639B1 - Device and method for iris recognition using convolutional neural network - Google Patents

Abstract

The present invention relates to an iris recognition device. More specifically, the present invention relates to a device and a method for recognizing an iris in which noise using a convolutional neural network (CNN) is included. The present invention can improve iris recognition accuracy without removing the noise such as lighting reflection light, eyebrows and eyelids included in an iris region from an eye image captured through a visible ray camera. The iris recognition device comprises: an eye image input part; an eye image preprocessing part; and an iris recognizing part.

Description

TECHNICAL FIELD [0001] The present invention relates to an iris recognition apparatus and method using a spiral neural network technique,

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an iris recognition apparatus, and more particularly, to an apparatus and a method for recognizing a noise including a noise using a convolutional neural network (CNN).

Recently, iris recognition systems have been widely used in fields such as financial transactions, access control, and smartphone security. The iris recognition system consists of near-infrared illumination and infrared camera to obtain a clear iris pattern. Such an iris recognition system is advantageous in that it can not be counterfeited or modulated. However, as shown in FIG. 1, there is a disadvantage in that iris recognition accuracy is lowered due to iris pattern blurring due to illumination reflected light, eyebrows and eyelids.

The iris recognition system that has been studied in the prior art is a system in which iris images as shown in Fig. 2 (a) obtained at a distance of 4 m to 8 m using a visible ray camera are binarized as shown in Fig. 2 (b) . However, the above-described method has a problem that it is impossible to remove various noise elements, and the iris pattern is also removed at the time of noise removal.

In order to solve the above-described problems, the present invention proposes a method of improving the recognition rate by using an iris image in which a noise component is not removed as shown in (d) of FIG.

Background Art of the present invention is disclosed in Korean Patent Laid-Open Publication No. 10-2014-0141917.

The present invention provides an iris recognition apparatus and method which are superior in recognition performance without removing noise such as illumination reflected light, eyebrows and eyelids included in an iris region from an eye image photographed through a visible ray camera.

According to an aspect of the present invention, there is provided an iris recognition apparatus including an eye image input unit into which an eye image imaged by an iris recognition apparatus is input through an image photographing apparatus, a pupil and an iris position in an input eye image, An eye image preprocessing unit for generating at least one image and an image around the eye and normalizing the size of the generated iris image and the eye surrounding image, a deep learning based spiral neural network model for the normalized iris image and the surrounding image, And an iris recognition unit for performing iris recognition by applying the iris recognition unit.

According to another aspect of the present invention, there is provided an iris detection method and a computer program for executing the iris detection method.

The iris detection method and the computer program for executing the iris detection method in accordance with an embodiment of the present invention include the steps of inputting a photographed eye image through a photographing apparatus, detecting pupil and iris positions in the input eye image, A pre-processing step of generating an iris image and an eye-surrounding image based on the position, normalizing the size of the iris image and an eye-surrounding image, performing spiral-based neural network-based learning on a normalized iris image and an eye- And performing iris recognition based on the neural network model.

The present invention uses a spiral neural network model in which eye images including various noises (illumination reflected light, eyebrows, eyelids) are learned based on deep learning, It is possible to provide an iris recognition apparatus and method with excellent performance.

1 is a diagram illustrating noise of an iris image in which iris recognition performance is degraded;
2 is a diagram illustrating an iris image used in a conventional iris recognition apparatus and an iris image used in an iris recognition apparatus of the present invention.
3 is a view for explaining an iris recognition system according to an embodiment of the present invention.
4 to 8 are views for explaining an iris recognition apparatus according to an embodiment of the present invention.
9 to 14 are views for explaining an iris recognition method according to an embodiment of the present invention.

While the present invention has been described in connection with certain exemplary embodiments, it is to be understood that the invention is not limited to the disclosed embodiments, but, on the contrary, is intended to cover various modifications and similarities. It is to be understood, however, that the invention is not to be limited to the specific embodiments, but includes all modifications, equivalents, and alternatives falling within the spirit and scope of the invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, the present invention will be described in detail with reference to the accompanying drawings. In addition, the singular phrases used in the present specification and claims should be interpreted generally to mean "one or more " unless otherwise stated.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings, wherein like reference numerals refer to like or corresponding components throughout. .

FIG. 3 is a view for explaining an iris recognition system according to an embodiment of the present invention, and FIGS. 4 to 8 are views for explaining an iris recognition apparatus.

Referring to FIG. 3, the iris recognition system includes a photographing device 10 and an iris recognition device 20. FIG.

The image capturing apparatus 10 captures an eye area of a user using a visible light camera and inputs the captured eye image to the iris recognition apparatus 20. [

The iris recognition device 20 recognizes the iris by applying a deep learning-based convolutional neural network to the input eye image.

The iris recognition apparatus 20 includes an eye image input unit 100, an eye image preprocessing unit 200, and an iris recognition unit 300.

The eye image input unit 100 inputs the eye image photographed through the image photographing apparatus.

The eye image preprocessing unit 200 detects pupil and iris positions in the input eye image, generates one or more iris images and an eye surrounding image based on the pupil and iris positions, Normalize the size.

4, the eye image preprocessing unit 200 includes a pupil and iris position detecting unit 210, an iris / eye surrounding image generating unit 220, and an iris / eye surrounding image normalizing unit 230.

The pupil and iris position detection unit 210 detects pupil and iris positions in the input eye image. At this time, the pupil and iris position detecting unit 210 can detect the pupil and iris positions by calculating the center coordinates and the radius of the pupil and iris using a circular edge detector (CED).

The iris and eye surrounding image generating unit 220 generates an iris image eye surrounding image using the detected pupil and iris positions. Here, the iris / eye surrounding image generating unit 220 may generate at least one eye surrounding image by enlarging the iris radius at the pupil position.

The iris and eye surrounding normalization unit 230 normalizes the iris image and the eye surrounding image into a non-square image of a predetermined size by polar coordinate transformation. For example, the normalization unit 230 may convert the iris image and the eye periphery image of the orthogonal coordinate system into a polar coordinate system and normalize the image to a size of 256 (horizontal) x 8 (vertical). Also, the normalization unit 230 may normalize the pixel brightness values of the iris image and the eye surrounding image.

The iris recognition unit 300 performs iris recognition by applying a deep learning-based spiral neural network model to the normalized iris image and the eye surrounding image.

Referring to FIG. 5, the iris recognition unit 300 includes a spiral neural network modeling unit 310, a score level combining unit 320, and a similarity determination unit 330.

The spiral neural network modeling unit 310 extracts a plurality of feature vectors by applying a spiral neural network model to each of the iris image and the perimeter image of a predetermined scale. The spiral neural network modeling unit 310 will be described in more detail in FIGS. 6 and 7. FIG.

The score level combining unit 320 combines a plurality of feature vectors to extract one final feature vector.

The similarity determination unit 330 calculates the Euclidean distance between the last feature vector and the previously registered feature vector to perform iris recognition.

Referring to FIGS. 6 and 7, the spiral neural network modeling unit 310 includes an image input layer 311, a feature extraction layer 312, and a classification layer 313.

The image input layer 311 inputs an iris image and an eye surrounding image of a predetermined scale.

The feature extraction layer 312 extracts features by applying a spiral neural network model to the input iris image and the eye surrounding image. In this case, the scale of the input iris image and the eye surrounding image is 256 (horizontal) × 8 (vertical), and the vertical size is very small to be applied to a general neural network model. This size makes convolution impossible in a general neural network model using a forward filter. Therefore, in the present invention, a spiral neural network model for learning a non-square filter suitable for iris recognition is designed. For example, the feature extraction layer can use a 1-by-13 or 1-by-11 non-square filter.

The feature extraction layer 312 may include (1) eight convolutional layers and (2) batch normalization layers, and (3) each convolutional layer may include a rectified linear unit layer And (4) may include four Max pooling layers. Here, the first convolution layer receives an image of 8 x 256 x 3 pixels, and can be convolved at intervals of one pixel using 64 filters of 1 x 13 x 1 size. The size of the feature map in the first convolution layer is 8 (= ((8 - 1)) by the output height (or width) = input height (or width) - filter height (or width) + 2 × padding) / stride + 1 + 2 × 0) / 1 + 1)) × 244 (= ((256-13 + 2 × 0) / 1 + 1)) × 64. From this, the feature map output from the first convolution layer may be input to the second convolution layer via the Batch Normalization layer and the Reforming Linear Unit (ReLU) layer. The iris recognition apparatus 20 can improve the learning speed by using a batch normalization layer. The second convolution layer outputs a feature map of 8 × 232 × 64 size using 64 × 1 × 13 × 1 filters. And then processed by a first max-pooling layer comprising 1x2 size filters applied at intervals of one pixel via a batch normalization layer and a calibration linear unit layer and downsampled to a size of 8 x 116 x 64. An image of 8 × 116 × 164 pixels is input to the third convolution layer, and the 1 × 13 size filter is convoluted at intervals of one pixel using 128 filters. The outputs of the third convolution layer are input to the fourth convolution layer via the bulk normalization layer and the calibration linear unit layer, and the 1x13 filter can be convoluted at intervals of one pixel using 128 filters have. The feature map output from the third convolution layer is input to the fourth convolution layer via the batch normalization layer and the calibration linear unit layer. The fourth convolution layer has 8 × 104 × 128 feature maps. The 1 × 13 size filter is convoluted at intervals of one pixel by using 128 filters, and passes through the batch normalization layer and the calibration linear unit layer , Processed by a second Max-Pulling layer including 1x2 filters applied at intervals of one pixel, and downsampled to a size of 8x46x128. The fifth convolution layer outputs a feature map of 8x36x256 size through a feature map of 8x46x128 size, 256 filters of 1x11x1 size, a batch normalization layer, and an alignment linear unit layer. The sixth convolution layer has a feature map of 8x36x256 size, which is divided into 1x2 size 1x11x1 pixels that are applied at intervals of 1 pixel after passing through 256 filters, a bulk normalization layer, and a calibration linear unit layer of 1x11x1 size Lt; RTI ID = 0.0 > 8x13x256 < / RTI > The seventh convolution layer outputs a feature map of size 6x11x512 through a filter of 8x13x256 size, 512 filters of 3x3x1 size, a batch normalization layer, and a calibration linear unit layer. The eighth convolution layer is a 1 × 2 size convolutional filter that applies a 6 × 11 × 512 feature map through 512 filters of 3 × 3 × 1 size, a batch normalization layer, and a calibration linear unit layer, Lt; RTI ID = 0.0 > 4x5x512 < / RTI > Finally, a feature map of size 4x5x512 may be used as the input of the first fully connected layer through eight convolution layers.

In addition, the feature extraction layer 312 may further include three fully connected layer sets next to the convolution layers. The first complete connection layer sets may further include a full connection layer, a bulk normalization layer, and a calibration linear unit layer. The first and second fully connected layer sets may include a bulk normalization layer and a calibration linear unit layer, and the third fully connected layer set may include a softmax layer. Here, the first complete connection layer may have 512 and 4096 nodes as input and output, respectively. The second complete connection layer may have 4096 and 4096 nodes as inputs and outputs, respectively. The third complete connection layer may have 4096 and N (N is a natural number) nodes as input and output, which can be finally obtained through a soft max function.

Referring to FIG. 8, the iris recognition apparatus 20 according to an embodiment of the present invention may change the feature extraction layer 312 according to the case of learning the eye image and the case of authenticating the eye image. For example, when the iris recognition apparatus 20 according to an embodiment of the present invention learns various eye images using the helical neural network model described above, eight feature extraction layers 312 (FIG. 8 And a feature vector output through the first through third complete connection layers. On the other hand, when the input eye image is recognized using the learned spiral neural network model, the iris recognition device 20 extracts eight feature extraction layers 312 and a first complete connection layer as shown in FIG. 8 (b) The output feature vector can be used for recognition. Accordingly, the iris recognition apparatus 20 according to an embodiment of the present invention can solve the overfitting phenomenon in which the spiral neural network model is excessively biased to the input training data, so that the learning result loses generality.

The classification layer 313 classifies the class into a predetermined number of classes based on features extracted by applying a spiral neural network model. The classifying layer 313 according to an embodiment of the present invention may include a third complete connecting layer and a soft max function for learning a variety of eye images using a helix neural network model, Classes can be classified.

9 to 14 are views for explaining an iris recognition method according to an embodiment of the present invention.

Referring to FIG. 9, in step S910, the iris recognition device 20 receives the eye image photographed through the image photographing device 10. FIG. Here, the eye image is an eye image photographed through a visible ray camera.

In step S920, the iris recognition device 20 detects pupil and iris positions in the input eye image. The iris recognition apparatus 20 can detect pupil and iris positions by calculating the center coordinates and the radius of the pupil and iris using, for example, a circular edge detector (CED) as shown in equation (1) .

(1)

(One)

와

범위에서 홍채의 중심 좌표와 반경을 조금씩 변화시키면서 영상의 화소 밝기 변화량이 가장 크게 되는 중심과 반지름을 찾음으로써 홍채 위치를 검출할 수 있다. 또한, 홍채 인식 장치(20)는 0~

범위에서 동공의 중심 좌표와 반경을 조금씩 변화시키면서 영상의 화소 밝기 변화량이 가장 크게 되는 중심과 반지름을 찾음으로써 동공 위치를 검출할 수 있다.In (1), (x, y) is the center of the iris, (x ', y') is the pupil center, r is the iris radius, and r 'is the pupil radius. The iris recognition device 20

Wow

The iris position can be detected by finding the center and the radius at which the pixel brightness variation of the image becomes the largest while slightly changing the center coordinates and radius of the iris in the range. In addition, the iris recognition device 20 may be configured to perform,

The pupil position can be detected by finding the center and the radius at which the pixel brightness variation of the image becomes the largest while slightly changing the center coordinates and the radius of the pupil in the range.

 In step S930, the iris recognition apparatus 20 generates an iris image and an eye surrounding image based on the detected pupil and iris positions. Referring to FIG. 10, the iris recognition apparatus 20 may detect the pupil position 1010 and the iris position 1020 as shown in FIG. 10A to generate an iris image. The iris recognition device 20 can change the iris radius based on the center coordinates of the pupil to generate at least one eye surrounding image including the sclera, eyelid, and eyelashes as shown in Figs. 9 (b) and 9 have.

In step S940, the iris recognition device 20 normalizes the sizes of the iris image and the eye surrounding image. Referring to FIG. 11, the iris recognition apparatus 20 normalizes the iris image and the eye periphery image into a non-square image of a predetermined size by polar coordinate transformation. For example, the iris recognition apparatus 20 converts the row (?) And theta (?) In the iris image of the orthogonal coordinate system into a polar coordinate system as shown in Fig. 11 (b) ) X 8 (length).

In step S950, the iris recognition device 20 extracts the iris feature vector and the eye feature vector by applying a learned spiral neural network model to each of the normalized iris image and the eye periphery image. The learned spiral neural network model is the same as the method described above with reference to Figs. 7 and 8.

In step S960, the iris recognition apparatus 20 combines the iris feature vector and the eye feature vector through a weighted product rule to generate a final feature vector. Referring to FIG. 12, the iris recognition apparatus 20 outputs three feature vectors by applying a spiral neural network model to one iris image and two eye neighbor images, respectively. ) Can be applied to generate a final feature vector.

In Equation (2), V is the final matching score calculated by the score level combination, B is the matching score calculated in the iris region, and A and C are the matching scores calculated in the area around the eye.

In step S970, the iris recognition apparatus 20 calculates the Euclidean distance between the final feature vector and the previously registered feature vector, and performs iris recognition through the similarity determination. For example, the iris recognition apparatus 20 is intra-class as the sum of the Euclidean distances between the elements of the feature vectors A and B is smaller, and as the sum of the Euclidean distances is larger, .

The iris recognition device 20 used a NICE.II (Noisy Iris Challenge Evaluation-Part II) data set for learning and verification of the spiral neural network model. Referring to Table 1, the NICE.II data set is composed of 171 classes including 1000 eye images each having a size of 400 x 300 pixels. The eye image is an image obtained by photographing a person walking at a distance of 4 m to 8 m away from the visible ray camera. The eye image is obtained by rotating, low illumination, blurring, and off-angle Noise is included. For the purpose of learning, the data set is divided into halves as shown in Table 1, and the A group including 86 classes and the B group including 85 classes are constructed.

Dataset  group Number of class Number of image
before data augmentation Number of image
after data augmentation NICE.II DataSet 86 515 41715 NICE.II DataSet B 85 485 39285 NICE.II DataSet A + B (original) 171 1000 81000

배 확장하여 총 81,000장의 데이터셋을 구성할 수 있다.The iris recognition apparatus 20 calculates an iris radius from a single eye image as described above with reference to FIGS. 10 and 11, and generates an image around the two eyes by changing one iris image and iris radius. In this way, a single image

It can be expanded to form a total of 81,000 data sets.

The iris recognition device 20 normalizes the iris image and the perimeter image of the orthogonal coordinate system into a non-square image of 256 (horizontal) × 8 (vertical) pixel size by the polar coordinate transformation and uses it for learning and verification of the spiral neural network model .

The iris recognition apparatus 20 learned a spiral neural network model by a caffe framework method. Here, the parameter values used in the cafe framework method are Momentum = 0.9, Momentum2 = 0.999, Epsilon = 1e-08, Batch size = 128, Epoch = 60-65 and Bias = Here, the learning of the spiral neural network model is the same as the method described with reference to Figs. 6 to 8 above.

The iris recognition apparatus 20 cross validates the A group and the B group of the above-described data set for verification of the learned spiral neural network model to determine an average error rate (EER) And a decidability value was measured. Here, the EER means the average error rate at the moment when the false acceptance rate (FAR) and the false rejection rate (FRR) are equal to each other, and the decidability value decides the truth or falsehood of any predicate in logic (3) can be calculated as a probability value that can be obtained.

(3)

(3)

는 결정가능성 값,

는 본인수락률의 평균,

는 타인 수락률의 평균,

는 본인수락률의 표준편차,

는 타인수락률의 표준편차를 의미한다.In equation (3)

Lt; / RTI >

Is the average of the acceptance rate,

The average of the acceptance rate of others,

Is the standard deviation of the acceptance rate,

Is the standard deviation of the acceptance rate of the other person.

As a result of the experiment, when A group was studied and B group was verified, EER was 8.462% and d 'was 2.8697. When A group was studied, A and E were 12.2576% and 2.3752, respectively.

Referring to FIG. 14, the iris recognition apparatus according to the present invention has an average irreducible value of 2.6225, which is higher than other iris recognition methods using NICE.II data.

The iris recognition method according to an embodiment of the present invention may be implemented in the form of a program command that can be executed through various computer means and recorded in a computer-readable medium. The computer readable medium may include program instructions, data files, data structures, and the like, alone or in combination. Program instructions to be recorded on a computer-readable medium may be those specially designed and constructed for the present invention or may be available to those skilled in the computer software arts. Examples of computer-readable media include magnetic media such as hard disks, floppy disks and magnetic tape; optical media such as CD-ROMs and DVDs; magnetic media such as floppy disks; Includes hardware devices specifically configured to store and execute program instructions such as magneto-optical media and ROM, RAM, flash memory, and the like. The above-mentioned medium may also be a transmission medium such as optical or metal lines, waveguides, etc., including a carrier wave for transmitting a signal designating a program command, a data structure and the like. Examples of program instructions include machine language code such as those produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter or the like. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the present invention, and vice versa.

The embodiments of the present invention have been described above. It will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims. Therefore, the disclosed embodiments should be considered in an illustrative rather than a restrictive sense. The scope of the present invention is defined by the appended claims rather than by the foregoing description, and all differences within the scope of equivalents thereof should be construed as being included in the present invention.

10: Video shooting device
20: iris recognition device
100: eye image input unit
200: eye image preprocessing unit
300: iris recognition unit

Claims (12)

delete delete In the iris recognition apparatus,
Wherein the iris recognition device comprises:
An eye image input unit to which an eye image photographed through the image photographing apparatus is input;
An eye image preprocessing unit for detecting pupil and iris positions in the input eye image, generating at least one iris image and an eye surrounding image based on the pupil and iris positions, and normalizing the size of the generated iris image and the eye surrounding image, ; And
And an iris recognition unit for performing iris recognition by applying a deep learning based convolutional neural network to the normalized iris image and the eye surrounding image,
The eye image pre-
Wherein at least one iris image and an eye surrounding image are normalized into a non-square image of a predetermined size by a polar coordinate system transformation,
The iris recognition unit
A spiral neural network modeling unit for extracting an iris feature vector and an eye feature vector by applying a spiral neural network model to an iris image and an eye peripheral image of a predetermined scale;
A score level combining unit for combining the iris feature vector and the eye feature vector to extract a final feature vector; And
And a similarity determination unit for performing iris recognition by calculating an Euclidean distance between the final feature vector and the previously registered feature vector.
The method of claim 3,
The spiral neural network-based spiral neural network modeling department
An image input layer for inputting an iris image and an eye surrounding image of a predetermined scale;
A feature extraction layer for extracting features by applying a spiral neural network model to an input iris image and an eye surrounding image; And
According to the extracted feature, a classification layer for classifying the iris into one of N (N is a natural number)
/ RTI >
5. The method of claim 4,
The feature extraction layer
An iris recognition device comprising eight convolution layer sets and three fully connected layer sets.
6. The method of claim 5,
The eight convolution layer sets
A first convolution layer into which 8 × 256 × 3 pixels of image are input and convolutes using 64 filters of 1 × 13 × 1 size;
A second convolution layer for convolutioning 8 × 244 × 64 pixels output from the first convolution layer using 64 × 1 × 13 × 1 filters;
A first Max-Pulling layer for applying a 1x2 filter to output 8x116 pixels;
A third convolution layer for convolutioning the 8 × 116 × 64 pixels using 128 filters of 1 × 13 × 1 size;
A fourth convolution layer for convoluting 8 × 104 × 128 pixels output from the third convolution layer using 128 filters of 1 × 13 × 1 size;
A second max-pulling layer applying a 1x2 filter to output 8x46x128 pixels;
A fifth convolution layer for convolutioning the 8 × 46 × 128 pixels using 256 × 1 × 11 × 1 filters;
A sixth convolution layer for convoluting 8 × 36 × 256 pixels output from the fifth convolution layer using 256 filters of 1 × 11 × 1 size;
A third Max-Pulling layer applying a 1x2 filter to output 8x13x256 pixels;
A seventh convolution layer for convoluting the 8x13x256 pixels using 512 filters of size 3x3x1;
An eighth convolution layer for convoluting 6 × 11 × 512 pixels output from the seventh convolution layer using 512 filters of 3 × 3 × 1 size; And
A fourth Max-Pulling layer for outputting 4x5x512 pixels by applying a 1x2 filter;
And an iris recognition device.
6. The method of claim 5,
The feature extraction layer
Further comprising a Batch Normalization layer or a Rectified Linear Unit in each of the convolution layer and the fully connected layer.
6. The method of claim 5,
The three fully connected layer sets
A first full connection layer having 512 and 4096 nodes as inputs and outputs, respectively;
A second full connection layer having 4096 and 4096 nodes as inputs and outputs, respectively; And
4096, and N (N is a natural number) nodes as inputs and outputs, respectively.
delete In the iris recognition method,
In the iris recognition method,
A step of inputting a snow image photographed through a video image pickup device;
A preprocessing step of detecting pupil and iris positions in the input eye image, generating an iris image and an eye surrounding image based on the detected pupil and iris positions, and normalizing the size of the iris image and the eye surrounding image; And
Performing spiral-based neural network-based learning on the normalized iris image and an eye surrounding image, and performing iris recognition based on the learned spiral neural network model,
Performing the spiral neural network-based learning on the normalized iris image and the eye surrounding image, and performing the iris recognition based on the learned spiral neural network,
Classifying the iris image and the eye surrounding image included in the data set into N (N is a natural number) classes;
Transforming the size of the iris image and the eye surrounding image included in the data set into a predetermined size by polar coordinate transformation;
Learning a spiral neural network model using modified eye images; And
And performing iris recognition using the learned spiral neural network model.
In the iris recognition method,
In the iris recognition method,
A step of inputting a snow image photographed through a video image pickup device;
A preprocessing step of detecting pupil and iris positions in the input eye image, generating an iris image and an eye surrounding image based on the detected pupil and iris positions, and normalizing the size of the iris image and the eye surrounding image; And
Performing spiral-based neural network-based learning on the normalized iris image and an eye surrounding image, and performing iris recognition based on the learned spiral neural network model,
Wherein the performing the iris recognition comprises:
Extracting an iris feature vector and an eye feature vector based on the learned spiral neural network;
Generating a final feature vector through score level combination of the iris feature vector and the eye feature vector; And
Further comprising calculating an Euclidean distance between the final feature vector and the pre-registered feature vector to determine a degree of similarity.
11. A computer program for executing the iris recognition method of claim 10 or 11 and recorded on a computer-readable recording medium.

Images