Image Segmentation Based Visual Security Evaluation
Christof Kauba, Stefan Mayer, Andreas Uhl · 2016
In this paper we present a metric for visual security evaluation of encrypted images, also known as visual security metric. Such a metric should be able to assess whether an image encryption method is secure or not. In order to consider intelligibility of objects in encrypted images our metric is based on image segmentation and applying a measure designed to evaluate the segmentation result. The visual security metrics' performance is evaluated using a selective encryption approach and compared to some general image quality metrics like PSNR, metrics suggested for encrypted images like Irregular Deviation and two metrics specifically designed for visual security evaluation. Our visual security metric performs better than all of the other tested metrics on the dataset and encryption algorithm we used during our experiments in terms of different correlation measures.