HUMAN IRIS RECOGNITION USING CIRCLE BASED SEGMENTATION AND OPTIMIZED CLASSIFIER
Loading...
Date
2020
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
FACULTY OF ENGINEERING AND TECHNOLOGY SAM HIGGINBOTTOM UNIVERSITY OF AGRICULTURE, TECHNOLOGY AND SCIENCES (FORMERLY ALLAHABAD AGRICULTURAL INSTITUTE) NAINI, PRAYAGRAJ-211007
Abstract
These days, Iris Recognition (IR) is a technique of biometric verification of the individual
authentication process centered upon the human iris unique pattern that is implemented to
control system intended for high security. The recognition of a person centered on iris pattern is
attaining more fame owing to the uniqueness of the pattern amongst the people. Iris recognition
system’s performance relies upon the method of segmentation and classification of iris from the
eye image. Segmentation and classification of the iris region are the most controversial issues in
iris recognition system because the poor result of these stages can shatter or spoil the
effectiveness of iris recognition systems.An Enhanced AMF algorithm was proposed to
effectively eradicate noise while at the same time preserves the image detail. During noise
detection a new variable was introduced to reflect the correlation betwixt pixel values. The
filtering algorithms were applied under noise ratio (PNR) of 30%. Experimental outcomes on
CASIA database reveals that with SNP of 30%, PSNR values of 14.1%, 19.58 and 23.42% were
attained for CMF, AMF and proposed enhanced AMF filters respectively.
The proposed ‘two-level segmentation’ methodology starts by converting the iris images
picked from database from RGB to Gray scale images and then normalized using Daugman’s
rubber sheet model. The interior and exterior boundaries are then detected from the
denoisedimage.In interior boundary segmentation (IBS) section, the image was segmented
utilizing some methods like Gaussian pyramid, anisotropic diffusion,thresholding, centroid
computing, polar transform, and radius computing.Exterior boundary segmentation (EBS)
section performs zigzag collarette process.Finally, the IBS was subtracted from EBS; to give the
segmented iris. Three databases namely CASIA, UBRIS and MMU1.0 were utilized for testing
the proposed approach. The proposed method reveals an accuracy of 97.33%, 97.97% and 97%
in CASIA, UBIRIS and MMU databases respectively
.ExperimentalresultswhencomparedwithACWOEandK-meansdemonstrated the superiority of the
proposed technique.
The proposed iris recognition framework takes the circle based iris segmented images as
the input. Then, the Log Gabor, GLAC, Contour-let transform, LGXP, and Canny Edge
Detection features are extracted. Thirdly, the extorted features are specified to the ANFIS for
achieving the training process. Amid the training process, the ANFIS parameters are
simultaneously optimized by the ACFO. The optimized parameters in ANFIS by ACFO
efficiently perform the IR process. At last, the recognized iris images are found. The experiment
outcomes for the proposed IR technique demonstrates its superiority over latest methodologies
pertaining to precision (96% for CASIA, 99% for UBRIS and 96.19% for MMU database),
Recall (100% for all databases), F-Measure (97.9% for CASIA, 98% for UBRIS and 99% for
MMU database) together with Accuracy (98.6% for CASIA, 97.9% for UBRIS and 98% for
MMU database).
Description
Ph. D. Thesis
Keywords
null