Adaptive Selection of Classifiers for Person Recognition by Iris Pattern and Periocular Image

• Published in: Neural Information Processing Vol. 13111; pp. 656 - 667
• Main Authors: Ogawa, Keita, Kameyama, Keisuke
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 2021; Springer International Publishing
• Series: Lecture Notes in Computer Science
• Subjects: Adaptive combination; Biometrics; Iris; Multimodal; Periocular image
• ISBN: 3030922723; 9783030922726
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-030-92273-3_54

Iris recognition is a type of biometric authentication that can achieve high authentication accuracy. However, its classification accuracy is significantly reduced when the image quality is low. In recent years, research on multi-modal authentication that uses not only the iris but also the periocular information that can be acquired together with the iris has been actively conducted. The purpose of this study is to improve the robustness of classification accuracy for degraded observed images by using iris and periocular modalities. In this paper, a method to select a classifier that is useful for authentication from the iris and periocular classifiers will be proposed for when either of the iris or the periocular image is of low quality. For the selection of the modal classifier, we propose and use the Multi Modal Selector that adaptively selects a classifier useful for classification by using parts of the outputs of the iris and periocular classifiers. In the experiment, it was shown that high classification accuracy can be maintained by adaptively selecting a useful classifier.