Video Iris Recognition Based On Iris Image Quality Evaluation And Semantic Classification

The use of video in biometric applications has reached a great height in the last five years. The iris as one of the most accurate biometric modalities has not been exempt due to the evolution of the capture sensors. In this sense, the use of on line video cameras and the sensors coupled to mobile devices has increased and has led to a boom in applications that use these biometrics as a secure way of authenticating people, some examples are secure banking transactions, access controls and forensic applications, among others. In this work, an approach for video iris recognition is presented. Our proposal is based on a scheme that combines the direct detection of the iris in the video frame with the image quality evaluation and segmentation simultaneously with the video capture process. A measure of image quality is proposed taking into account the parameters defined in ISO /IEC 19794-6 2005. This measure is combined with methods of automatic object detection and semantic image classification by a Fully Convolutional Network. The experiments developed in two benchmark datasets and in an own dataset demonstrate the effectiveness of this proposal.