ES2224838A1 - System for biometric identification of people, by analyzing iris, for use in access control in e.g. buildings, has determination unit determining signature of iris, and comparison unit comparing captured image with stored image - Google Patents

Abstract

The system has a capturing unit to capture the image of the eye, and a pre-processing unit pre-processing the image. An extraction and determination unit extracts characteristics of the image and determines the signature of the iris. A comparison unit compares the captured image with a stored image, and a decision-making unit provides access to the people.

Description

Biometric people identification system by iris analysis.

Technical sector to which the invention relates

1. Financial services.

2. Access control to buildings and services.

3. Computer and communications security Electronic

4. Electronic commerce.

5. Public safety.

6. Airports.

Statement of prior art

The goal of a biometric system is the identification / authentication of people through some physically measurable or biological characteristic of individual behavior, comfortably and safely. Traditionally, the solution to this problem has been based on the possession by the user of keys, PINs, or other numbers of personal identification. However these numbers can be easily forgotten, lost and even counterfeit. In this sense, Biometrics provides a higher level of security since that the biological characteristics (iris, voice, fingerprints, facial morphology, etc.) are singular and unalterable, impossible of losing, forgetting or transferring. Currently, there are research related to different biometric technologies such as those based on the characteristics of the iris, voice, footprint finger, hand shape, facial morphology, signature, shape walk, etc.

The system that is intended to be patented here is based on the potential of the iris to identify the individual since the fundamental characteristic of the iris pattern It is the uniqueness. Indeed. according to ophthalmic studies, in the visual pattern of the iris there is more information that identifies uniquely to a person who on a fingerprint. In fact the two irises of the same person have different patterns, being This is a characteristic to be taken into account in the system as it is not The image of the two interchangeable eyes. Even the brothers Twins have well differentiated iris patterns. In addition, the characteristics of the human iris give it the same following properties:

Stability against originated changes due to accidents, due to the protection granted by the cornea.

?

Easy "live subject" detection, since small variations of lighting produce alterations in the opening of the pupil, even when it comes to fixed lighting.

?

The picture of the iris can be obtained non-invasively, by being visible from the outside for the transparency of the cornea.

?

The falsification of a person's iris would involve operations Surgical that could seriously damage vision.

These properties, along with the characteristic of high uniqueness, guarantee the viability of this technique of biometric identification

The idea of using the iris pattern for people identification was initially proposed by the ophthalmologist Frank Burch in 1936, but it wouldn't be until 1987 when two American ophthalmologists, Leonard Flom and Aran Safir, They patented the concept of Burch. The development of the algorithms necessary was carried out by J. G. Daugman, professor of the Harvard University, and patented in:

\ sqbullet

JG Daugman , "Biometric personal identification system based on iris analysis", Patent US 5291560, 1994 .

Another patent related to the invention that here it is intended to patent is:

\ sqbullet

RP Wildes , JC Asmuth , SC Hsu , RJ Kolczynski , JR Matey and SE McBride , "Automated. Noninvasive iris recognition system and method", Patent US 5572596, 1996 .

However, no exploitation is known yet of the latter. More recently, and also related to the The invention presented here is the patent:

\ sqbullet

JL Cambier and C. Musgrave , "System and method of animal identification and animal transaction authorization using iris patterns", Patent WO 0135321, 2001 .

The main differences (advantages) of the system to be patented with respect to the systems designed by JG Daugman, RP Wildes et al , and JL Cambier et al ., The first of which is the basis of iris recognition systems Currently in marketing, they are:

\ sqbullet

The system which is intended to patent used in its extraction stage of features a one-dimensional signal (or sequence) (1-D) of 256 points, called iris signature, at contrary to patented systems that use the total image of the iris, that is a two-dimensional (2-D) signal, with what with the system presented here, the time spent in user identification / authentication is significantly less.

\ sqbullet

In the phase of extraction of characteristics is where we can say that there is an essential difference between the method presented here for this phase and the methods corresponding to the mentioned patents. The method on which the system to be patented is based, the representation of the zero crossings of the discrete dyadic wavelet transform of the one-dimensional signal called the iris signature, effectively contributes, with respect to the methods used in the aforementioned patents, to reduce in an important way the computation times in this phase, whether the system is implemented as a user identifier or if it is done as an authenticator. And, although the three systems (those patented by JG Daugman and RP Wildes et al . And the one presented here) use a multiresolution analysis in their feature extraction phase, the patented systems perform it on the 2- D image of the iris, with the consequent increase in computational cost; However, the one presented here uses the multiresolution analysis of a 1-D signal (the signature of the iris) and makes from it the representation of zero crossings considering, only. the optimal resolution levels, thus contributing to the reduction of the computational cost of this phase and, therefore, to the improvement of the functioning of the system itself.

\ sqbullet

In the comparison phase there are also fundamental differences, since the JG Daugman method generates an iris code of 256 bytes (2048 bits), while the representation of the iris in the system patented by RP Wildes et al . It is obtained directly from the filtered image, storing more information than is necessary in the comparison process, making it virtually unfeasible for storage in devices with small storage capacity. However, the iris pattern that is obtained with the method presented here, called the iris signature code, consists of 256 bits for each level of resolution used, much smaller than the iris pattern used by patented systems , and makes it ideal, therefore, for inclusion in a device with small storage capacity, such as, for example, a smart card.

In the last of the mentioned patents (from J.L. Cambier and C. Musgrave) describes: 1) an identification system of animals and 2) a transaction authorization system of animals by identifying the person and the animal; both based on the comparison of the characteristics of the corresponding iris patterns. In both systems, the methods used both in the feature extraction phase of the iris as in the comparison phase are the same as the proposals by J.G. Daugman in his patent (US 5291560). The patented system It focuses, therefore, on the incorporation of a device for carry out the capture phase of the iris image by video, as well as the methodology for its subsequent transmission to the base of central data where the irises comparison is made, and finally, in establishing the method to effect the animal transaction authorization. The designed device allows you to capture the image of the iris, process it until you get the iris pattern, encrypt the latter in order to secure the transmission of the pattern obtained to the central database for subsequent phase of comparison.

Regarding the scientific literature, in addition to article by J.G. Daugman, which includes details of the cited patent, also found, in various magazines specialized, research articles related to our invention, as are the following:

[1] WW Boles , "A Security System based on Human Iris Identification using Wavelet Transform", Engineering Applications of Artificial Intelligence , vol. 11, no. 1, pp. 77-85, Feb. 1998 .

[2] WW Boles ; B. Boashash , "A Human Identification Technique using Images of the Iris and Wavelet Transform", IEEE Transactions on Signal Processing , vol. 46, no. 4, pp. 1185-1188, 1998 .

[3] Y. Zhu , T. Tan and Y. Wang , "Biometric Personal Identification Based on Iris Patterns", ICPR2000: the 15th International Conference on Pattern Recognition, pp. 801-804, Barcelona, Spain.

[4] Shinyoung-Lim, Kwanyong Lee, Okhwan Byeon and Taiyun Kim, "Efficient Iris Recognition through Improvement of Feature Vectors and Classifier", ETRI-Journal , vol. 23, no.2, pp. 61-70, June 2001 .

[5] C. Sanchez-Avila , R. Sanchez-Reillo and D. de Martin-Roche , "Iris-Based Biometric Recognition using Dyadic Wavelet Transform", IEEE Aerospace and Electronic Systems Magazine , vol. 17, no. 10, pp. 3-6, October 2002 .

[6] C. Sanchez-Avila and R. Sanchez-Reillo , "Multiscale Analysis for Iris Biometric" Proceedings IEEE 2002 International Carnahan Conference on Security Technology, pp.35-38.

       \ newpage
    

The main ones are described below. differences of our invention regarding ideas and methods proposed in the aforementioned articles.

Main differences with respect to the method published in [1] and [2] . Regarding the system presented here, one of the main differences found with the method presented in articles [1] and [2] is that they only show the phases of feature extraction and comparison, without being treated in any case the image capture and preprocessing module, which are the previous steps, but fundamental (since they condition the whole process), in every biometric identification system. Referring to the two stages that are considered, in the method described in [1] and [2] it also uses the representation of zero crossings of the wavelet transform of what the authors call the iris signature, a term they use to designate the values of the gray levels of a concentric virtual circle of 45 pixels in diameter of the isolated iris. In the system presented here, the average of the values of the gray levels of a circular crown formed by concentric virtual circles from a given internal radius, r_ {i} to what we call the external radius is considered as the iris signature , r_ {e}, thereby achieving better results.

Another of the fundamental differences of the method on which the system presented here is based on that designed by WW Boles et al . is that the pattern that is stored for the subsequent phase of comparison is the representation of zero crossings of three levels of resolution (from 4 to 6) of the wavelet transform (each of these levels is a sequence of 256 points), what which demands considerable storage requirements. In the system presented here, in addition to using the optimal resolution levels (from 3 to 5), the pattern that is stored (called the iris signature code) consists of only 256 bits for each level of resolution, which can be Store in a small device. In addition, in the comparison phase, WW Boles et al . they only use dissimilarity functions related to the representation of zero crossings of the wavelet transform, with the consequent increase in computational cost. In the system presented here, the Hamming distance between the different signature codes of the irises that are intended to be compared is used in the comparison phase, which involves only o-exclusive logical operations (\ oplus), with the consequent saving in time and computational cost. Finally, it is necessary to comment that the main similarity with our system is the use of a 1-D iris signature as a vector of iris characteristics (which is an advantage over the patented systems of JG Daugman and RP Wildes et al ., which use 2-D images); as well as the use of the representation of the zero crossings of the discrete dyadic wavelet transform. In fact, article [2] was the starting point of our research since 1999, but today it has given rise to the system that is intended to be patented whose method differs essentially from that proposed by WW Boles et al .

Main differences with respect to the system published in [3] . The main and essential difference of our system compared to that presented in [3] lies in the phase of extracting characteristics. In [3] the characteristics of each iris are extracted using a global multiscale texture analysis, applying a multi-channel Gabor filter, followed by the application of the 2-D wavelet transform with 5 resolution levels. The essential advantage of the system that is intended to be patented with respect to it is that, in addition to obtaining significantly better results than those obtained with the system proposed in [3] (which uses the weighted Euclidean distance), it is avoided, in the extraction phase of features, the application of the Gabor filter for the extraction of textures and, in addition, the wavelet transform that is applied is always 1-D with 3 levels of resolution, comparatively achieving a significant saving in the computational cost of the system at this stage .

Main differences with respect to the system published in [4] . In [4], the authors propose an iris-based biometric recognition system that, in addition to the image capture stage, includes:

one)

a scheme of representation of the iris pattern by the wavelet transform two-dimensional (2-D) with 4 levels of resolution, using as a basic wavelet the wavelet of Haar, in the stage of feature extraction;

2)

two new methods of competitive learning for LVQ, in the final stage of comparison, in the identifier / verifier versions.

The essential differential (disadvantage) of the system proposed in [4] compared to the one presented here, fundamentally, in the feature extraction stage, already that instead of working with a 1-D signal, in [4] it works with the image (2-D), 450x60, of the iris full. In addition, in the system presented in [4], they are obtained, as a result of applying the wavelet transform four times 2-D, 4 sub-images for the 4 resolution levels, from which the 87 bit pattern. However, although the size of the iris pattern is smaller than the one obtained with the method presented here, The computational cost of this stage is extraordinary, considering also that 8 sub-images are obtained additional ones that are not used in any other process step. By On the other hand, the final stage of comparison is based on the use of a LVQ type neural network, whose mechanism of learning is faster than the algorithm of retro-propagation, but its smooth operation depends on a correct initialization of the weight vectors of the net. This method is much slower and requires a number. greater computational resources than distance application Hamming In addition, the results obtained with the proposed system in [4] they are worse, both in classification (with success in 98.4% classification, compared to 100% obtained with the system that patent is intended) as in verification (reaching an EER rate 2.56% versus 0.12% achieved with the system here Presents itself).

Main differences with respect to the system published in [5] and [6] . As can be seen, the authors of [5] and [6] are the researchers authors of the system presented here and intended to be patented. The main similarity stands out in the first place: both systems use as a mathematical tool the representation of the zero crossings of the discrete dyadic wavelet transform of a one-dimensional (1-D) signal, considered as a vector of iris characteristics. In both methods, it is called the iris signature, but the construction of this iris signature is different in the two systems. In those published in [5] and [6] the idea of WW Boles et al . Is followed at all times, so that its establishment differs essentially from that presented here. On the other hand, 4 levels (3 ≤ j ≤ 6) of resolution were considered, without analyzing the number and type of optimal resolution levels. Nor was any type of iris signature code established, which is entered here. In fact, the results obtained with the system to be patented are significantly better, both in recognition and in user authentication, than those obtained with the systems published in [5] and [6].

Explanation of the invention.

The biometric identification system Developed consists of four stages:

1. Capture the eye image

The image capture device of the eye of A person can be a camera or video camera. As soon as the lighting environment must be taken into account ability to reflect the cornea, which, as a smooth and well lubricated surface, reflects every ray of light that arrives For this reason, depending on the device used for the capture, the environment of illumination. In this sense, video cameras have greater sensitivity, which would avoid the use of high intensity foci, at the cost of a lower resolution.

On the other hand, the possibility of use infrared light, instead of visible light, both in the lighting as in the capture. This would avoid perception by User part of the lighting bulb.

2. Preprocessed iris image and establishment of the iris signature

The objective of this stage is to adapt the signal to the requirements of the feature extraction block. Taking into account the characteristics of the captured images, the first step is a conversion to black and white, followed by a Histogram stretch. Once this operation has been carried out, will proceed to:

2.1. The detection of the internal and external edges of the iris , in order to proceed to the isolation of the iris within the image, by means of an algorithm of detection of edges in images.

2.2. The determination of the iris signature . A main objective of the invention is the determination or extraction of the iris signature, which is performed as follows. Once the iris is isolated, the centroid of the pupil detected as a reference point is chosen to obtain the data set that we call the iris signature. Thus, the values of the gray level, g_ {r} of the outline of each virtual circle, with center in the centroid and with radius r , are extracted such that r_ {i}, \ leq r \ leq r_ {e}, being r_ {i} = 5 and r_ {e} = 45 the internal and external radii, respectively (Figure 1). Finally, the data set called iris signature (IS) is calculated (Figure 2):

IS = \ frac {1} {r_ {e} -r_ {i} +1} \ sum \ limits ^ {r_ {e}} _ {r = r_ {i}} g_ {r}

This way you get the entrance to the feature extraction stage be a signal unidimensional, not two-dimensional as is the case with the techniques current, with which two great advantages are achieved: 1) greater processing speed, and 2) lower requirement in terms of storage requirements since the iris signature consists of 256 points.

3. Feature extraction and code determination of the iris signature

Another main objective of the invention is the representation of iris characteristics by representing zero crossings of its discrete dyadic wavelet transform. Thus, in this third block of the system, the discrete dyadic wavelet transform of the iris signature is calculated using as a basic wavelet a quadratic spline, only for the scales (or resolution levels) corresponding to parameter j = 3 to j = 5 ( Figure 3). Next, the representation of the zero crossings of the previously calculated transform is carried out and only for the scales 3 j j 5 5, which is shown in Figure 4. It is convenient that the algorithm be used to carry out This representation is invariant against rotation and includes a mechanism to eliminate false crossings by zero.

Another main objective of this invention is the determination of the iris signature code that is carried out in this last step of the stage, in order to adapt the representation of zero crossings obtained to the block of comparison. The value of each bit in the iris signature code (Figure 5) is obtained by taking the representation of zero crossings for the 3 scales considered and associating the value "1" for positive or null values, and "0" for values negative

4. Comparison and decision making

The last stage of the system is that of comparison of the characteristics obtained from the captured iris image with a previously stored pattern. As in any system biometric, it can be configured well as a

\ sqbullet

classifier (biometric recognition), in which the characteristics extracted are compared with the patterns of all users who are already recruited by the system, in order to determine the identity of the person; or as a

\ sqbullet

Verifier (biometric authentication), in which the features extracted are only compared with the pattern of the user who claims to be, in order to authenticate the identity of the user in response to the question: is this user the person who claims to be?

For this task you can use a distance of dissimilarity that allows to identify a person, if the system has been implemented as a classifier; or authenticate your identity if it has been implemented as a verifier. In concrete, another of the main objectives of the system that we intend to patent is the realization of this comparison by the binary Hamming distance, d b H, which is described as the shape:

d ^ {b} {} H} = \ frac {1} {M-K + 1} \ sum \ limits ^ {L} _ {j = K} \ left (\ sum \ limits ^ {L} _ {i = 1} y_ {i} \ oplus p_ {i} \ right)

which does not measure the difference between the components of the feature vectors but the number of components whose values differ. In the above formula, L is the dimension of the feature vector; y_ {i} is the i-th component of the feature vector that is to be compared and p_ {i} the ith component of the pattern feature vector; K and M are the finest and grossest resolution levels, respectively, considered in the analysis. This comparison process results in a level of accuracy that quantifies the similarity between both feature vectors. If the level is higher than a certain threshold it can be concluded that both patterns come from the same iris.

System decision making is therefore conditioned by the results in both schemes. In the case of implement the system as a classifier the result of the comparison can be positive (the system accepts the user), if the distance between the pattern to be classified and that of the user with a highest probability always remains below a certain threshold; or it can be negative (the system rejects the user), otherwise.

In the case of implementing the system as a Verifier, the results are expressed based on two rates of error:

\ sqbullet

Cup of False Rejection (FRR): percentage of cases in which the system rejects a system user.

\ sqbullet

Cup of False Acceptance (FAR): percentage of cases in which an intruder as a user of the system.

These two rates get different values according to the threshold that is taken and they keep a relationship of inverse variation. The choice of the threshold depends exclusively on the requirements of System security and operation for a specific environment. Thus, for high security environments it will be necessary to apply a threshold that allows to obtain minimum values of the FAR even at the cost of increase the FRR.

However, for forensic applications it will be a threshold is required with which very low FRR values are achieved still at the cost of increasing the FAR. Usually, from these two rates are obtained the so-called ROC curve (FRR vs. FAR), which allows to determine the threshold to be applied according to the scope of system application.

Brief description of the drawings

Description of Figure 1 . Figure 1 represents the eye of the person to be identified, where (1) represents the iris isolated from the eye; (2) the pupil; (3) and (4) the outer and inner circles (centered on the centroid of the iris) and radii 5 and 45, respectively, that delimit the circular ring of the iris (5) from which the iris signature is calculated.

Description of Figure 2 . Figure 2 shows the iris signature of the isolated iris. It is a one-dimensional signal (or sequence) that is formed by taking for each point of it the arithmetic mean of the gray values at that point of the circles that make up the circular ring shown in Figure 1.

Description of Figure 3 . The three resolution levels, corresponding to the scales j = 3, 4 and 5, of the discrete dyadic wavelet transform of the iris signature shown in Figure 2 are shown in Figure 3.

Description of Figure 4 . Figure 4 shows the representation of zero crossings of the three resolution levels of the iris signature previously calculated.

       \ newpage
    

Description of Figure 5 . Figure 5 shows the iris signature code (represented in Figure 2) that is obtained by associating the value 1 for the positive or null values of the representation of zero crossings and 0 for the negative values, for the 3 scales considered .

Description of Figure 6 . Figure 6 shows the results in verification obtained in the practical implementation of the system detailed in the corresponding section, for different threshold values.

Detailed statement of at least one embodiment of the invention

An embodiment is detailed below. of the system to be patented, with which they have been obtained optimal results and improving those obtained with the systems currently patented.

1. Capture the eye image

The photographic image capture of the eye is made using a Kodak Professional digital camera DCS 315 that allows the storage of photographs in PCMCIA flash cards making it possible to download them in Any computer with a PCMCIA card reader. To get the  necessary approach you use a goal Micro-Nikkor 105 mm f / 2.8D. When choosing the lighting form, we must take into account the high capacity of cornea reflection. To avoid this problem a localized and polarized lighting bulb that is placed in the bottom of the camera, so that it illuminates from below to above, causing a reflection of light located in the lower cone of the iris To eliminate this reflection, a polarizer is placed in quadrature with that of the light emitter located on the target of the camera. However, the use of the polarizing filter in the lens causes a loss of camera sensitivity, so that the flash built into the camera must be used in order of increasing incident light. This fact causes a reflection perfectly located near the center of the pupil and can be easily removed.

2. Preprocessed iris image and establishment of the iris signature

The objective of this stage is to adapt the signal to the requirements of the feature extraction block. Taking into account the characteristics of the captured images, the first step is a conversion to black and white, followed by a Histogram stretch. Once this operation has been carried out, will proceed to:

2.1. internal and external edge detection of the iris, in order to isolate the iris within the image, and to

2.2. Iris signature extraction.

2.1. Iris isolation

The steps to follow for the detection of Iris edges are as follows:

one)

You get a 4 times smaller copy of the image, in order to lighten the computational cost

2)

It is eliminated, through a threshold, the parts of the image overexposed, due to the use of the flash and, subsequently, the histogram is stretched again.

3)

It establishes a grid at whose ends are located the centers that are will use in the edge detection algorithm. The point that determine the best edge will be taken as the center of the iris, performing iris location and detection in this way of the outer edge, simultaneously.

4)

It starts from an arbitrarily chosen point (x_ {0}, y_ {0}), it is taken as the center and from 61 the image is sampled, taking the corresponding points from an increase of the radius Δ_ {r} and an increase in the angle Δ_ {theta}, looking for the multiple n of Δ_ {r} that maximizes D ,

D = \ sum \ limits_ {m} \ sum \ limits ^ {5} _ {k = 1} (I_ {n, m} -I_ {n-k, m})

being

I_ {i, j} = I (x_ {0} + i \ Delta_ {r} \ cos (j \ Delta_ {\ theta}), and_ {0} + i \ Delta_ {r} \ sin (j \ Delta_ {theta}))

where m are those multiples of Δ {that correspond to points in the lateral cones of the iris and I (x, y) are the intensity values of the image.

5)

Once the point of the grid that provides the maximum for D has been found , a new grid with higher resolution is created, covering only the square formed by the points between the previous and the later of the chosen one, both on the horizontal axis and on the vertical.

6)

The procedure indicated in point 4); so, being the higher resolution grid, you get a center more and more refined.

7)

Once the center is located, the value of \ Delta_ {r} is reduced in order to obtain greater precision in the position of the edge, determining the distance from the center to the edge which is: n \ Delta_ {r}, where n the value that maximizes D.

Once the outer edge of the iris is detected, it cut the original image into a square that encompasses the iris detected and all those points that are left out of the circumference that frames the iris. Then the same process for the detection of the internal edge, that is to say eliminate the points of overexposure and the histogram is stretched. This will allow to take advantage of the entire dynamic range of intensity in the iris so that, even if it is dark, it can be distinguished from the pupil when detecting the inner edge.

For the detection of the internal edge, the same center could be used and a new maximum of D could be sought, if we consider that the iris and pupil are concentric, a fact that is not true, in general, since the iris is slightly displaced down and towards the nose, this displacement being in some cases 15%. Thus, starting from this center, a new grid is created that covers ± 20% of the size of the image obtained in the previous step, and the same process is reapplied, from points 4) to 7), for the edge external.

Finally we get a central point -centroid- as well as the distance from that point to the border between the iris and The pupil. The points within the circumference defined, they are canceled in the image resulting from the detection of outer edge and a stretch of the histogram, obtaining the iris isolated from the rest of the image.

2.2 Iris signature establishment

Once the isolated iris is obtained, proceed to extract the iris signature, as detailed in the explanation of the invention.

3. Extraction of the signature characteristics of the iris. Iris signature code

The extraction of the signature characteristics of the iris and the final establishment of the iris signature code It is carried out by the following steps:

\ sqbullet

Calculation of the discrete dyadic wavelet transform ( Discrete Dyadic Wavelet Transjorm : DDWT) of the one-dimensional (1-D) signal (or sequence) called iris signature, using the quadratic spline of compact support as the basic wavelet. For this, a fast calculation algorithm of the discrete dyadic wavelet transform for 1-D sequences is used. Thus, and restricting ourselves to dyadic scales, we have that the iris signature DDWT ( IS ), is the set of the following sequences:

\ {{A_ {2 8}} (IS), (D_ {2 j}}, (IS)) {1 \ leq j \ leq 8} \}

where A_ {2 <8}} is the grossest approximation sequence (corresponding, therefore, to the highest scale) and (D_ {2} {(IS)) _ {1 \ leq j \ leq 8} the set of detail sequences, from the scale j = 1 to the scale j = 8. In order to obtain a robust representation in a noisy environment and, at the same time, reduce the computational load, both the level plus gross as the finest levels. As is known, the information contained in the finest resolution levels is heavily contaminated by the noise that affects the system as well as quantification errors. To reduce such effects in the representation of zero crossings, only intermediate resolution levels are considered. In particular, three levels of resolution are considered, corresponding to the 3 \ leq j \ leq5 scales (Figure 3).

\ sqbullet

Multi-scale representation of zero crossings of the iris signature DDWT . The zero crossings of (D2 {j}} (IS)) {3 le j ≤ 5} are calculated from the sign changes of your samples. The position of each zero crossing is estimated by a linear interpolation between two samples of different signs. Thus, if the iris signature ( IS ) has N samples other than zero, and since there are at most Nlog (N) samples in its discrete dyadic wavelet transform, the number of operations required to obtain the zero crossings position is O (Nlog (N)). Thus, for any pair of consecutive zero crossings of D_ {2 ^ {j}} with 3 \ leq j \ leq5, whose abscissa are z_ {n-1} and z_ {n} and, respectively, it is possible to calculate the value of the integral:

in} = \ int ^ {z_ {n}} {{z_ {n-1}} D_ {2 ^ {j}} (IS) dx

and for any D_ {2 ^ {}} the position of the zero crossings (z_ {n}) _ {n \ in Z} can be represented by the following constant function a pieces:

\hskip0.5cm

x\in[z_{n-1}, z_{n}]Z_ {2 ^ {j}} (IS) = \ frac {e_ {n}} {z_ {n} -z {_ {n-1}}}

 \ hskip0.5cm 

x \ in [z_ {n-1}, z_ {n}]

Therefore, the multi-scale representation of zero crossings of the signature of the iris (IS) is considered, on scales 3 j j le5, to the sequence set (Figure 4):

\ {(Z_ {2j}} (IS)) 3 \ leq j \ leq5} \}

Done in this way, the representation obtained is invariant against rotation. As a previous step to the obtaining the iris signature code, it is convenient to implementation of a mechanism to eliminate false crossings by zero since, although the distortion produced in this representation of zero crossings due to white noise is greatly reduced in low resolution levels, in some critical positions with high levels of resolution, crosses may appear and disappear by zero. The inflection points of the wavelet transform are Some of those points. As a result, two identical irises can have different representations of zero crossings if one of them It is in a noisy environment.

However, comparing the number of crosses by zero of the unknown iris representation (in environments noisy), with the number of zero crossings of the representation of our model (in a noise-free environment) at the same level of resolution, it is possible to eliminate false zero crossings if Use a suitable algorithm. Also, as the representation is periodically, the number of zero crossings must be even. So, When eliminating false crosses, it must be done two by two. This can be done using the following algorithm:

With the default resolution range (3 \ j j \ leq5), starting with the grossest level of resolution, j = 5, and ending with the finest level, j = 3, follow these steps:

one.

If the number of zero crossings is the same, go to step 4, if not, go to step 2.

2.

Modify the representation that has a greater number of zero crossings of the following form:

\ sqbullet

Calculate the algebraic area of the rectangular pulses between each pair of consecutive zero crossings.

\ sqbullet

 Remove the pulse that has the lowest absolute value of the area.

\ sqbullet

Join the two zero crossings adjacent to the one eliminated to form a new pulse. The area of the new pulse is calculated following the following equation:

A_ {new} = A_ {n + 1} + A_ {n} + A_ {n-1}

where A_ {n} is the area Pulse algebraic removed. The two positions of the crosses by zero of the new pulse are defined by the positions left and right of the zero crossings of the left neighbors and right of the pulse removed, respectively.

3.

Repeat step 2 until the number of zero crossings of the model and the candidate match.

Four.

Go to the next finest resolution level, and return to step 1.

The last step in this stage is the extraction of called the iris signature code (Figure 5) from the multi-scale representation of zero crossings of the iris signature, and after the application of the false crossing elimination algorithm, as indicated in the explanation of the technique. This code of the iris signature will constitute the iris pattern of the person with which will compare the signature code of the person's iris that you want to identify or authenticate. Your storage can be done on a small device, a smart card by example, in an encrypted form (using an encryption algorithm of secret or public key) or not.

4. Comparison and decision making

As detailed in the explanation section of the invention, the system can be configured as a classifier , or as a user verifier . In both cases, as in any pattern recognition system, the system database, which will be constituted by all the iris signature codes of all persons (users) authorized to enter the system, must be previously prepared.

If the system is used as a classifier (biometric recognition), the iris signature code is compare with all iris signature codes stored in the system database. In the realization carried out, this comparison has been carried out by calculating the distance of Hamming between the iris code of the person to be identified and all database codes. The decision is made from the as follows: the system will identify the person you intend access as that user whose iris signature code stored in the database get the shortest distance from Hamming in the comparison process. In the realization carried out by the authors the operation of the system has been tested with all users, that is to say the ability to system identification for all users that make up the database, obtaining 100% ranking success, is say the system has correctly identified each user in All the tests performed.

If the system is used as a verifier (biometric authentication), the iris signature code of the person who intends to enter the system is compared to the code of the signature of the iris of the user who claims to be, stored in the system database. In this case, decision making is like  continue: the system will authenticate the user as one who claims to be yes Hamming's distance between the iris signature code of the same and the one that is stored in the database is less than a certain predetermined threshold.

In the embodiment carried out for different thresholds, the results shown in the Figure 6. It is important to highlight as the main result the Obtaining null False Acceptance (FAR) rates for False Rejection (FRR) very small (0.19%, for a threshold value equal to 89); which makes the system ideal for environments High security In addition, the Equal Error Rate (EER) obtained; that is, the point where the FAR and FRR rates are equal, resulted be 0.12%, lower than that obtained by other systems biometric based on the human iris pattern currently in use.

Moreover, the computational cost in this stage is minimal, as the calculation of Hamming's distance requires only logical operations of o-exclusive, and therefore the time required in each comparison is very small (less than 1 msg).

Industrial application

The way to apply the system industrially described follows from the description itself. Do not However, we will indicate as more important its applicability in the industry related to computer security services, financial, governmental, police, airport, sanitary, domotic and inmotic and, in general in the industry related to all those areas or services that require the identification or authentication of users for access to said area or service So, within its multiple applications industrial include, among others, manufacturing and commissioning operation of:

\ sqbullet

ATMs automatic with iris recognition;

\ sqbullet

chambers of iris-based authentication for access control (airport, home automation, real estate, sanitary, etc);

devices capable of joining a PC for user authentication via iris in commerce electronic, electronic libraries, etc.

Claims (7)

1. Biometric identification system of people by analyzing the iris, which includes the following consecutive stages:

\ sqbullet

capture of the image of the eye of the person to be identified;

\ sqbullet

preprocessed image to isolate the iris of the eye within the image and establishment of a one-dimensional signal (or sequence) (iris signature) that characterize the iris isolated;

\ sqbullet

analysis of the iris signature to extract its characteristics through the wavelet transform and determination of the signature code of the analyzed iris; Y

\ sqbullet

comparison of characteristics obtained from the image of the iris captured with a Previously stored pattern and decision making.

2. Biometric identification system of people by analyzing the iris according to claim 1, characterized in that the preprocessing step of the image to isolate the iris and the establishment of the iris signature comprises:

\ sqbullet

the black and white conversion of the eye image and stretch of the histogram;

\ sqbullet

the detection of the inner and outer edges of the iris through a image edge detection algorithm, to proceed to iris isolation within the image;

\ sqbullet

the choice of the centroid of the pupil detected as a reference point to obtain the data set called iris signature by extracting the values of the gray level, g_ {r}, of the contour of each virtual circle, centered on the centroid and radius r , such that r_ {i} \ leq r \ leq r_ {e}, where r_ {i} = 5 and r_ {e} = 45 the internal and external radii respectively; Y

\ sqbullet

the calculation of the data set called iris signature (IS) of the shape:

IS = \ frac {1} {r_ {e} -r_ {i} +1} \ sum \ limits ^ {r_ {e}} _ {r = r_ {i}} g_ {r} 3. Biometric identification system of persons by means of the iris analysis according to claims 1 and 2, characterized in that the stage of extraction of characteristics and determination of the iris signature code comprises:

\ sqbullet

the application of the discrete dyadic wavelet transform to the whole of data called iris signature;

\ sqbullet

the representation of zero crossings of said transform;

\ sqbullet

the use of a false crossing elimination algorithm by zero; Y

\ sqbullet

the construction of the iris signature code: for each of the resolution levels of the representation of zero crossings of the Transformed discrete dyadic wavelet of the iris signature, the value of each bit in the iris signature code is obtained by associating the value "1" for positive or null values, and a "0" for negative values.

4. Biometric identification system of people by means of iris analysis according to claims 1 to 3, characterized in that the stage of comparing the characteristics of the image of the captured iris with a previously stored pattern and the decision-making includes the following steps:

\ sqbullet

the o-exclusive logical operation (\ oplus) between corresponding bits, for each resolution level considered, of the signature codes of the irises that are compared;

\ sqbullet

the sum of the components of each resulting binary vector, for each level of resolution;

\ sqbullet

the calculation of the average of the resulting sums, which is defined as distance of Hamming between two codes of the signature of the irises that compare, and

\ sqbullet

the Hamming distance conversion calculated to a measure for generate the decision of whether (or not) both iris signature codes they come (or not) from the same iris, and therefore from the same person: assigning the identity to a person, whose code is compared, with the of the one with whose code the minimum distance of Hamming of all with which it compares.

         \ newpage
      

5. Biometric identification system of people by iris analysis according to claims 1 to 4, characterized in that the preferable levels of resolution are the intermediate levels, that is to say j = 3, 4 and 5. 6. Biometric identification system of people by means of the analysis of the human iris according to claims 1 to 5, characterized in that infrared light is used in the image capture stage of the eye. 7. Biometric identification system of people by iris analysis according to claims 1 to 6, characterized in that the iris signature code is stored in an encrypted form, using a secret key or public key encryption algorithm.