DEEP LEARNING BASED IRIS RECOGNITION SYSTEM

Author Name(s): Dr Puja S Prasad, Dr Vinit Kumar Gunjan
Author Email: puja.s.prasad@gmail.com

Abstract

The main focus on this paper is to use multiple instance of a single   biometric traits that integrates the left and right iris of a person for authentication. Deep structured learning or we say trained deep learning approach is proposed whose architectural design is based on the combination of and Softmax classifier and Convolutional Neural Network (CNN) for the extraction of distinct floral features from the input iris images. In this work, for learning rate adaptation and weights updating a selective training scheme of CNN that is based on a combination of mini-batch Ada Grad optimization method and back-propagation algorithm and is being proposed. Two databases are used for evaluating performance of this proposed system.  The results outputted from the work of this proposed system outperform achieved an identification rate Rank-1 of 99.8% on all the corresponding different databases that is used here as well as recognition time is also reduces.

Keywords

Iris, Unimodal biometric, Ada Grad optimization, Convolutional Neural Network, Back-propagation algorithm.

Introduction

For authentication of an individual person biometric systems are growing rapidly and also it is a challenging technologies that can be used for identification and verification of a person without the use of traditional method for authentication like Cards, passwords, IDs. By this reason, iris recognition for authenticating a person has been used in many useful applications, authenticating a person for accessing in sensitive area, airport access, database access national identification cards, as well as financial services and is said to be unique as well as best reliable and accurate biometric systems. Multimodal biometric systems have been used widely in many areas for authentication purposes   real-world applications. The limitations present in unimodal system like noise, non-universality, intra-class variability, inter-class variability, population coverage, and vulnerability to spoofing. In our work, multimodal   system is given which is actually  based on the concept of deep learning  for representations of left and right irises of a person, and after that the features obtain from both the left and right irises are fused   by a method name called ranking-level fusion Numerous studies have demonstrated that the iris features have a very unique characteristics compare to   other biometric traits (e.g., palm print, earlobe, DNA, fingerprint), which make it highly demanding for application where high reliability as well as more accurate decision required[6,9,10].

Conclusion

In the above proposed paper multimodal biometric method for authentication is proposed by using the both irises as a biometric. A deep learning method is proposed which uses CNN and Softmax as feature extractor. The deep learning structure which is in fact based on a combination of the both Softmax classifier and CNN. As iris consists of rich feature this technique is used to extract the features of both left and right irises without any need of domain knowledge. After that it is classified into classes. After that Identification and ranking scores are calculated for each individual. For learning rate adaptation and weights updating a selective training scheme of CNN that is based on a combination of mini-batch Ada Grad optimization method and back-propagation algorithm and is being proposed. Two databases are used for evaluating performance of this proposed system.  The outcome obtained from the work of proposed system outperform achieved a Rank-1 identification rate which is equal to 99.8% on all the corresponding   databases that is used here as well as recognition time is also reduces.

187 total views, 3 views today

Download PDF File

About the author: admin