Hybrid Iris Recognition: Leveraging HOG Features and Dense Networks for Enhanced Biometric Accuracy

The article describes a new method for iris recognition, based on a hybrid approach combining edge detection, Histogram of Oriented Gradients (HOG) features, and dense neural networks(DNN), enabling improved accuracy and reduced training time. Using Circular Hough Transform (CHT)segmentation and Canny Edge Detection for localization, followed by feature extraction through HOG, the authors train the data using dense networks. The results show a 92.37% accuracy on the CASIA-Iris-Interval dataset and 96.28% on the IITD dataset, outperforming previous techniques. The authors illustrate the proposed method with results from MATLAB simulations on three datasets. The hybrid method allows for a reduction in training time by 10 minutes compared to traditional Convolutional Neural Network (CNN)methods. The effectiveness of the new method is confirmed by accuracy improvements of up to 5 %. These research results supplement existing iris recognition techniques and can be used for real-time biometric applications. The novelty and scientific contribution lie in integrating HOG and dense networks, enhancing both accuracy and computational efficiency in biometric systems.