An Effective Framework of Iris Recognition Using UNet-Based Segmentation and Attention-Based Classification
Gorla Babu, Pinjari Abdul Khayum · 2024
Due to its numerous applications in security fields such as homeland security, border control, and airports, Iris Recognition (IR) has been a hotly debated field over the past few years. Despite the fact that many of the previous efforts at IR attain a very high rate of accuracy, a significant amount of preprocessing steps that are needed, and the use of handcrafted features make it inapplicable and ineffective for various datasets of the iris. Even though numerous types of research in this field of application are available, they still face numerous issues like complexity in computation and larger consumption of time. In this paper, a novel IR system is proposed to address the aforementioned issues. At first, the benchmark datasets are used to collect the raw iris images. Pre-processing is then performed on the collected images at the next phase using median filtering. The segmented images are obtained utilizing UNet from the resulting pre-processed image. Finally, the Attention-based ResNeXt (A-ResNeXt) method is used for the classification procedure. The performance is then calculated using a variety of parameters and is compared with other conventional approaches. As a result, the model's output undoubtedly demonstrates the improved recognition performance of the developed IR model.