Security Assessment of Selectively Encrypted Visual Data: Iris Recognition on Protected Samples

Abstract

Security assessment of partially encrypted visual data is known to be difficult. We show that iris recognition performance on encrypted sample data turns out to be a good predictor for protection strength of encryption schemes for visual data, in particular in settings where image structure and intelligibility is preserved to some extent. This is of particular interest, as common image quality metrics used for this task exhibit specifically poor prediction quality in this setting.