Iris recognition under non-ideal conditions
Akhila P (41718002)
Iris recognition under non-ideal conditions - MPhil CS 2018-2019
A biometric system provides automatic identication of an individual based on unique physiological
or behavioral features or characteristics possessed by an individual. Iris recognition
is such a technology for identifying humans by capturing and analyzing the unique patterns
of the iris in the human eye. Unlike other biometrics such as ngerprints or face, the distinctive
aspect of iris comes from randomly distributed features in the iris. Iris recognition is
regarded as the most reliable and accurate biometric identication system available at present.
Iris recognition/classification under non ideal imaging conditions is a challenging task. Nonideal
conditions include iris images that are off-angled, occluded, blurred, noisy, iris captured
at a distance and so on.
Segmentation of the iris from the eye image is an important task in iris recognition. Iris
segmentation is the process of isolating the iris region from the surrounding structures of the
captured image of the eye. In ideal situations, we can easily segment the iris region from the
eye images. However, in the case of non ideal situations, iris segmentation is a challenging task.
One can never get valid recognition result without the right type of segmentation. Previous
studies have shown that the performance of an iris recognition systems highly dependents on
the performance of its segmentation algorithm. Several iris segmentation algorithms such as
those based on Hough transform, active contour and integro-differential operator have been
developed. However, these segmentation algorithms do not perform well under non-ideal and
less constrained conditions. So, one of the objective of this research is to develop an automatic
segmentation method for extracting the iris region. Towards this in this thesis, we propose
a novel mechanism for segmentation of iris images using a encoder-decoder network. The
proposed method gave an accuracy of 98% on non ideal iris images taken from UBIRIS v2
database.
Classification of iris images acquired under non ideal conditions is another challenge
due to the poor quality of the images. Currently, there are some works on iris recognition/classification
based on CNN. They use one of the pre-trained models to extract the features
and then SVM is used for the classification. SVM is a non parametric model where the complexity
grows as the number of training samples increases. To reduce the complexity and improve
the recognition accuracy, we propose a novel mechanism for iris classication using LSTM
sequence prediction model. LSTM can extract the long-term dependencies of the data features
in the sequence. It is very similar to the recognition by the human visual system. We tested the
proposed method on a UBIRIS v2 database which includes iris images under various non ideal
conditions and compared its performance with existing SVM based approach. The proposed
method gave an accuracy of 81% and outperformed the existing SVM based method (78%
accuracy).
IRIS RECOGNITION
RECURRENT NEURAL NETWORK
CONVOLUTIONAL NEURAL NETWORK
Iris recognition under non-ideal conditions - MPhil CS 2018-2019
A biometric system provides automatic identication of an individual based on unique physiological
or behavioral features or characteristics possessed by an individual. Iris recognition
is such a technology for identifying humans by capturing and analyzing the unique patterns
of the iris in the human eye. Unlike other biometrics such as ngerprints or face, the distinctive
aspect of iris comes from randomly distributed features in the iris. Iris recognition is
regarded as the most reliable and accurate biometric identication system available at present.
Iris recognition/classification under non ideal imaging conditions is a challenging task. Nonideal
conditions include iris images that are off-angled, occluded, blurred, noisy, iris captured
at a distance and so on.
Segmentation of the iris from the eye image is an important task in iris recognition. Iris
segmentation is the process of isolating the iris region from the surrounding structures of the
captured image of the eye. In ideal situations, we can easily segment the iris region from the
eye images. However, in the case of non ideal situations, iris segmentation is a challenging task.
One can never get valid recognition result without the right type of segmentation. Previous
studies have shown that the performance of an iris recognition systems highly dependents on
the performance of its segmentation algorithm. Several iris segmentation algorithms such as
those based on Hough transform, active contour and integro-differential operator have been
developed. However, these segmentation algorithms do not perform well under non-ideal and
less constrained conditions. So, one of the objective of this research is to develop an automatic
segmentation method for extracting the iris region. Towards this in this thesis, we propose
a novel mechanism for segmentation of iris images using a encoder-decoder network. The
proposed method gave an accuracy of 98% on non ideal iris images taken from UBIRIS v2
database.
Classification of iris images acquired under non ideal conditions is another challenge
due to the poor quality of the images. Currently, there are some works on iris recognition/classification
based on CNN. They use one of the pre-trained models to extract the features
and then SVM is used for the classification. SVM is a non parametric model where the complexity
grows as the number of training samples increases. To reduce the complexity and improve
the recognition accuracy, we propose a novel mechanism for iris classication using LSTM
sequence prediction model. LSTM can extract the long-term dependencies of the data features
in the sequence. It is very similar to the recognition by the human visual system. We tested the
proposed method on a UBIRIS v2 database which includes iris images under various non ideal
conditions and compared its performance with existing SVM based approach. The proposed
method gave an accuracy of 81% and outperformed the existing SVM based method (78%
accuracy).
IRIS RECOGNITION
RECURRENT NEURAL NETWORK
CONVOLUTIONAL NEURAL NETWORK