Multimodal Eye Biometric System for Unconstrained Eye

Biometrics frameworks have essentially enhanced individual authentication, playing a

significant part in personal, national, and global security. Existing ocular biometric

system achieves good accuracy results for images acquired using NIR cameras in ideal

condition only. When visible wavelength images are acquired in unconstrained

environment, noise is introduced such as illumination, reflection, motion blur etc.

which degrade the recognition performance. This research presents a multimodal eye

biometric framework utilizing Support value-based fusion (SVBF) matching process to

enhance biometric authentication by combining the of iris, sclera, and pupil

characteristics from unconstrained coloured eye images. A multimodal biometric

architecture using the fusion-associating support-value method is introduced in this

report to improve biometric authentication. The proposed strategy is portrayed in

subsequent steps; initially CNN (Convolutions Neural Network) segmentation based

on quality feature selection using entropy is applied to cluster iris, pupils and sclera

region. Subsequently effective features are extracted from the segmented iris, pupil and

sclera region, for example colour histogram, Log Gabor and sclera Y- shape features.

On the basis of the extricated features, the support value-based fusion is determined,

and the matching score is calculated by means of the minimum, maximum value and

support value derived from the features. Finally, authentic person is predictable by

computing a Euclidean distance of training and testing matching scores. The proposed

findings are tested on constrained database MMU, unconstrained image database

UBIRIS.V2 and mobile image database MICHE to show with the current techniques

the efficiency of the proposed authentication technique. Experimental results shows

that proposed multimodal biometric system provides better results as compared to

existing state-of-art. Segmentation performed using E-CNN improves results for

segmentation accuracy up to 97.99% for iris, 98.08% for sclera and 99.43% for pupil

segmentation under uncontrolled environment by reducing segmentation time up to

0.9sec. Proposed SVBF framework also highlights the role of feature level fusion to

enhance the recognition accuracy up to 97% for unconstrained visible wavelength

images.