In recent decade Iris recognition has emerged as a highly precise and dependable form of biometric identification among humans. In this research article, we introduce a meticulous framework for an iris recognition system that employs the Long Short-Term Memory algorithm to extract and classify features. Dense Fully Convolutional Neural network algorithm is used for segmentation. This method provides highly accurate segmentation. Normalization is done by Daugman’s method. The normalized image is subsequently enhanced to serve as input for the LSTM model. The dataset is then divided into training and testing subsets. The LSTM network proceeds to extract relevant features and subsequently classify the corresponding labels. The experimental procedure concludes with the utilization of both the CASIA 1000 and IITD Datasets. The performance measure is evaluated in terms of Precision, Recall, Specificity, F-Score, and Accuracy rate on both datasets. The findings have demonstrated that the suggested approach attains the utmost precision of 99.85% and 99.88% in the CASIA 1000 and IITD datasets, respectively. This method is compared with the proposed DFCN and Gabor method. The DFCN and LSTM combination achieves little higher accuracy than DFCN and Gabor method.

LSTM, Daugman, DFCN