Framework for biometric iris recognition in video, by deep learning and quality assessment of the iris-pupil region
Eduardo Garea-Llano,Annette Morales-Gonzalez
DOI: https://doi.org/10.1007/s12652-021-03525-x
IF: 3.662
2021-10-05
Journal of Ambient Intelligence and Humanized Computing
Abstract:In the current world scenario the influence of the COVID19 pandemic has reached universal proportions affecting almost all countries. In this sense, the need has arisen to wear gloves or to reduce direct contact with objects (such as sensors for capturing fingerprints or palm prints) as a sanitary measure to protect against the virus. In this new reality, it is necessary to have a biometric identification method that allows safe and rapid recognition of people at borders, or in quarantine controls, or in access to places of high biological risk, among others. In this scenario, iris biometric recognition has reached increasing relevance. This biometric modality avoids all the aforementioned inconveniences with proven high efficiency. However, there are still problems associated with the iris capturing and segmentation in real time that could affect the effectiveness of a System of this nature and that it is necessary to take into account. This work presents a framework for real time iris detection and segmentation in video as part of a biometric recognition system. Our proposal focuses on the stages of image capture, iris detection and segmentation in RGB video frames under controlled conditions (conditions of border and access controls, where people collaborate in the recognition process). The proposed framework is based on the direct detection of the iris-pupil region using the YOLO network, the evaluation of its quality and the semantic segmentation of iris by a Fully Convolutional Network. (FCN). The proposal of an evaluation step of the quality of the iris-pupil region reduce the passage to the system of images with problems of out of focus, blurring, occlusions, light changing and pose of the subject. For the evaluation of image quality, we propose a measure that combines parameters defined in ISO/IEC 19794-6 2005 and others derived from the systematization of the knowledge of the specialized literature. The experiments carried out in four different reference databases and an own video data set demonstrates the feasibility of its application under controlled conditions of border and access controls. The achieved results exceed or equal state-of-the-art methods under these working conditions.
computer science, information systems,telecommunications, artificial intelligence