In this paper, we propose approaches to prevent automatic inference of scene class by classifiers and also enhance (or maintain) the visual appeal of images. The task is part of the Pixel Privacy challenge of the MediaEval 2019 workshop. The fusion based approaches we propose apply adversarial perturbations on the images enhanced by image enhancement algorithms instead of the original images. They combine the benefits of image style transfer/contrast enhancement and the white-box adversarial

more »

... ack methods and have not been previously used in the literature for fooling the classifier and enhancing the images at the same time. We also propose to use simple Euclidean transformations which include image translation and rotation and show their efficacy in fooling the classifier. We test the proposed approaches on a subset of the Places365-standard dataset and get promising results.