Insights into the results of MICHE I - Mobile Iris CHallenge Evaluation. (February 2018)

• Record Type: Journal Article
• Authors: De Marsico, Maria; Nappi, Michele; Narducci, Fabio; Proença, Hugo
• Is Part Of: Pattern recognition. Volume 74(2018:Feb.)
• Page Start: 286
• Page End: 304
• Publication Date: 2018-02
• Subjects: Mobile Iris Recognition -- Evaluation -- Biometric algorithm fusion; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2017.08.028
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: The paper presents the analysis of results of the MICHE-I competition on iris recognition. Compared methods work on a dataset captured on mobile devices in visible light. Segmentation and recognition methods were recombined in all possible ways. The paper evaluates score-level fusion of recognition algorithms. Abstract: Mobile biometrics technologies are nowadays the new frontier for secure use of data and services, and are considered particularly important due to the massive use of handheld devices in the entire world. Among the biometric traits with potential to be used in mobile settings, the iris/ocular region is a natural candidate, even considering that further advances in the technology are required to meet the operational requirements of such ambitious environments. Aiming at promoting these advances, we organized the Mobile Iris Challenge Evaluation (MICHE)-I contest. This paper presents a comparison of the performance of the participant methods by various Figures of Merit (FoMs). A particular attention is devoted to the identification of the image covariates that are likely to cause a decrease in the performance levels of the compared algorithms. Among these factors, interoperability among different devices plays an important role. The methods (or parts of them) implemented by the analyzed approaches are classified into segmentation (S), which was the main target of MICHE-I, and recognition (R). The paper reports both the results observed for either S or R, and also for different recombinations (S+R) of such methods. Last but not least, we also present the results obtained by multi-classifier strategies.