These very large datasets are extremely effective for training general purpose networks, but present some issues like the question of consent and privacy (Birhane and Prabhu, 2021). Notwithstanding the presence of these general datasets, new emerging imaging modalities or specific application scenarios require the deployment of their specific datasets, but this process is hindered by the difficulties in collecting a sufficient number of relevant data, and by the fact that the underlying imaging technologies are continuously evolving. Let us provide an example of this problem, related to the multimedia forensics area. In that field, one of the research issues is the identification of the acquisition device that generated the content.
Photo Response Non-Uniformity (PRNU), a unique pattern noise left by each sensor (Lukas et al., 2006; Iuliani et al., 2019). Its uniqueness ensures that the sensor pattern Spain phone number list noises extracted from different cameras are strongly uncorrelated, even when they belong to the same camera model. To test the performance of the methods based on this feature, several datasets have been proposed, the most adopted one being the VISION dataset (Shullani et al., 2017). The VISION dataset is currently composed by 34,427 images and 1,914 videos, both in the native format and in their social version (Facebook, YouTube, and WhatsApp are considered), from 35 portable devices of 11 major brands.
A have been released several years ago, so their imaging technology is some- what obsolete. In the meantime, with the advent of computational photography, the image acquisition pipeline is rapidly changing with novelties involving both the acquisition process, and the in-camera processing, and this process could hinder the usefulness of PRNU. Preliminary analysis carried out on over 33.000 Flickr images belonging to 45 smartphone and 25 DSLR camera models released recently, show that non-unique artifacts that may reduce PRNU noise's distinctiveness, especially when several exemplars of the same device model are involved in the analysis, appear in some of the most recent devices.