LONG-SHORT TERM CONVOLUTIONAL NEURAL NETWORKS FOR IRIS RECOGNITION UNDER NON-IDEAL CONDITIONS

Author Name: Akhila P

Volume: 01/02

Country: India

DOI NO.: DOI Link: https://doi-ds.org/doilink/07.2024-87334124/IJISRE

Affiliation:

1. Kerala, India

ABSTRACT

Iris recognition system provides automatic identification of an individual based on unique patterns of the iris in the human eye. These features include the ridges, arches, folds, crypts, freckles, corona, furrows, etc. Existing iris recognition systems are heavily dependent on high user cooperation. A major drawback of this stop-and-stare condition is that they lead to low throughput. Also, the user is often required to look at the camera which may not facilitate recognition under covert conditions. In environments where user cooperation is not guaranteed, prevailing segmentation and matching schemes of the iris region are confronted with many problems, such as obstruction by eyelids, eyelashes, specular, non regular lighting reflections in the eye area, poorly focused images, partial, out-of iris images, invalid off-angle rotations and motion blurred irises. Currently, there are some works on iris recognition/classification based on convolutional neural network (CNN). They use one of the pre-trained models to extract the features and then support vector machine (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 classification using long short term memory (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 95.25% and outperformed the existing SVM based method.

Key words: CNN · Super Resolution · Semantic Segmentation · Feature Extraction LSTM.