Overview: iris vs. fingerprint
Fingerprint recognition is a popular security feature in the newer generation of smartphones and is a well-known biometric technology. Since Microsoft introducing iris recognition feature in its smartphones, there were comparisons between these two biometric traits. We will be discussing both these biometric technologies, their capabilities and security features.
Iris and fingerprint recognition both have higher accuracy, reliability and simplicity as compared to other biometric traits. These attributes make iris and fingerprint recognition perform better and a particularly promising security solution in today’s society. The process starts by capturing the images of iris and fingerprint which are then pre-processed to remove any noise effects. The distinguishing features are then extracted and matched to find similarity between both the feature sets. The matching scores that are generated from the individual recognizers are given to the decision module which decides if a person is genuine or an impostor.
Add this infographic to your site by copying this code
How does iris recognition work?
Iris pattern is unique to each individual and remains constant throughout the lifetime of a person. A circular black disk known as pupil lies in the center of the eyeball that dilates on exposure to light and contracts in dark. So the size of the pupil varies with respect to the amount of light it is exposed to. The annular ring that is located between the sclera and pupil boundary is called the iris and contains a large number of minute details. The iris also has an extremely data rich physical structure and contains the flowery pattern that is unique to each individual. This pattern remains unchanged with age.
This unique flowery pattern is extracted from the rest of the captured eye image and transformed into strip to which a pattern matching algorithm is applied. It is important that the acquired iris image is rich in texture because all steps in iris recognition depend upon the quality of the image. An automated iris recognition system had been proposed in which multi-scale quadrature wavelets is used to extract the structure information of the iris. A 2048 bit iris code is generated and the difference between a pair of iris representations is compared by comparing their Hamming distance using XOR operator. At various resolution levels of a virtual circle on an iris image, the zero-crossing representation of 1-D wavelet transform that characterizes the texture of the iris has been calculated. The iris texture is obtained with a Laplacian pyramid that has been constructed from four distinct resolution levels. Normalized correlation is then used to confirm if the input and the model image belong to the same class.
What are the steps involved in iris recognition?
Prior to proceeding with any of the steps mentioned above, it is important to acquire an image of the iris that is rich in texture because all subsequent steps will depend upon the quality of the iris image. A 3CCD camera is used to capture the image of the iris under a controlled lab environment. This captured image is then passed to a localization module that detects the iris portion from the rest of the image.
The pupil needs to be detected and removed from the acquired eye image since only the iris pattern is used for matching purpose. The pupil is the darkest portion of the eye and the first step is to look for the contours of the acquired image. The pupil region contains the lowest intensity values and so its edges can be found easily. The next step after edge detection is to find the center of the pupil. This is achieved by dilating the edge detected image and then the dilated image with filled pupil circle is used to compute the Euclidean distance between non-zero points. This distance helps to form the spectrum that shows the largest filled circle. The overall intensity of the spectrum is maximum at the center as the pupil is the largest filled circle in the image. From this spectrum image, the pupil center can be computed as the pixel position that has the maximum value. The distance between the pupil center and the nearest non-zero pixel is the radius of the pupil.
The outer iris boundary is detected by using intensity variation approach. Concentric circles having different radii are drawn from the detected center. In this approach, the iris circle is detected by locating the circle that shows the highest change in intensity compared to the previously drawn circles. Iris images that display sharp variation between iris boundary and sclera work fine with this approach. The radius of iris and pupil boundary is used in transforming the annular portion of the iris to a rectangular block known as strip.
The localized iris image is converted into strip. The Cartesian coordinates are first transformed into its polar equivalent after which the mapping is done. The transformed iris image is composed of points which are taken from the pupil boundary to the outer iris boundary. This essentially means that the same group of points is considered for every image. The iris image needs to be normalized to ensure that the size of strip remains constant for different images. The size of same iris image however might vary due to the expansion and dilation of pupil. So the size of iris strip is constant for every iris image.
The unique characteristics of the iris are derived by extracting the attributes or values of the image. These attributes or values are known as features and are extracted from the iris image using Haar Wavelet decomposition process. The Haar Wavelet process decomposes the image into four coefficients – horizontal, diagonal, vertical and approximation. The approximation coefficient is again decomposed into four coefficients and the sequences of steps are repeated for five levels. The last level coefficients are combined to form a vector and binarized. This allows the iris codes to be easily compared for database and query image. The binarized vectors are then passed to the matching module for comparisons.
The hamming distance approach is used to compare the iris codes (IC) that are generated for the database and query images. The difference between the bits of two codes is counted in this approach and the number is divided by the total number of comparisons. This matching score is provided as input to the fusion module that generates the final matching score.
Strengths of iris recognition
- Iris recognition has a proven highest accuracy rate. A biometric products testing final report found
- iris recognition to have no false matches in over two million cross-comparisons.
- Iris recognition is able to handle very large populations at high speed. It has the capacity to perform very large 1: all searches within extremely large databases.
- Iris biometric is very convenient and the individual simply needs to do look into a camera for a few seconds. The process captures a video image that is non-invasive and inherently safe.
- The iris is very stable and remains unchanged throughout a person’s life. There are also no changes to the physical characteristics of the iris even when the person ages.
- Iris recognition is an affordable biometric modality and has very low maintenance costs. Moreover, it allows seamless interoperability between different hardware vendors and can also work well with other applications.
What are the steps involved in fingerprint recognition?
Fingerprint recognition is a proven technique to verify the identity of individuals and hence it is one of the most widely used biometric technologies. A fingerprint is basically composed of ridges and valleys that are on the surface of the finger. In fingerprint recognition, there are three major steps that are applied to acquired images using the minutiae matching approach.
There are various kinds of noises such as creases, smudges and holes that can corrupt a fingerprint image. The quality of the fingerprint image cannot be improved for the unrecoverable regions of the fingerprint. Moreover, it is not possible to recover the true ridge/valley structures from these unrecoverable regions. Hence it is required to use an enhancement algorithm that can improve the clarity of ridges and valley structures of fingerprint images in the recoverable regions and mask out the unrecoverable regions.
The first step in the image enhancement phase is to normalize the input fingerprint image. This is done so that the fingerprint image has pre-specified mean and variance. The normalized input fingerprint image is used to estimate the orientation image. The normalized input fingerprint image and the orientation image are further used to compute the frequency image. After this computation, each block in the normalized input fingerprint image is classified into a recoverable or unrecoverable block to obtain the region mask. The last step is to apply a bank of Gabor filters to the ridge and valley pixels in the normalized input fingerprint image to obtain the final enhanced fingerprint image. The Gabor filters that were applied to the normalized input image were tuned to local ridge orientation and ridge frequency.
Binarization is applied to the enhanced fingerprint image and a thinning algorithm is used to reduce the ridge thickness to one pixel wide. The points of ridge endings and bifurcations are known as the minutiae points and are extracted using a skeleton image. A feature set is then formed by extracting and storing the location of minutiae points and their orientation. For extraction of minutiae points, the crossing number (CN) method is used. The crossing number method uses a 3×3 window to examine the local neighborhood of each ridge pixel. The ridge endings and bifurcations will then be extracted from the skeleton image.
Minutiae is extracted from the database and query fingerprints and then stored as points in the two dimensional plane. In minutia based matching, the idea is to find the alignment between the template and the input minutiae sets with the maximum number of minutiae pairings.
Strengths of fingerprint recognition
- Fingerprint recognition is a widely accepted biometric modality and is excellent for background checks. It has found numerous applications in the areas of law enforcement and government forensics such as the AFIS database.
- For populations that have a low incidence of “outliers”, fingerprint biometric has a relatively low false rejection rate and false acceptance rate. This however may not be the case for large groups or groups that have race and gender variations.
- Fingerprint solutions are provided by a wide range of vendors.
- Fingerprint technology has the ability to enroll multiple fingers.
Fingerprint technology has been widely used within the law enforcement community and in the AFIS database. Thus it is a very popular biometric technology and also widely accepted. However, fingerprint readers may not be sufficient to handle the large variation in populations that need to be enrolled. Performing a search in large-scale deployments may take many minutes and might also require ancillary data such as age, sex etc. for partitioning the database in order to increase search speed. Moreover, the search may also return multiple matches. Thus a back-up identification method such as iris recognition is required that can provide resolution for these multiple matches.
Fingerprint technology works best for background check applications. Iris recognition has a very high accuracy rate and is also a non-invasive biometric technology. In this article, we have reviewed both fingerprint and iris recognition. We have also highlighted the benefits and weaknesses of these two biometric modalities.