User profiles’ image clustering for digital investigations

Sharing images on Social Network (SN) platforms is one of the most widespread behaviors which may cause privacy-intrusive and illegal content to be widely distributed. Clustering shared through SN according acquisition cameras embedded in smartphones regarded as a significant task forensic investigations cybercrimes. The Sensor Pattern Noise (SPN) caused by camera sensor imperfections due manufacturing process has been proved an effective robust fingerprint that can used for several tasks, such digital evidence analysis, smartphone fingerprinting user profile linking well. uploaded users their profiles way sources it considered challenging since upload different types images, i.e., taken users’ (taken images) single from sources, cropped or generic Web (shared images). make perturbation clustering task, they do not usually present sufficient characteristics SPN related sources. Moreover, are directly referable user’s device so have detected removed process. In this paper, we propose profiles’ image method without prior knowledge about type number hierarchical graph-based clusters both images. strengths our include overcoming large-scale datasets, presence perturb loss details compression platforms. evaluated VISION dataset, public benchmark including 35 smartphones. dataset perturbed 3000 simulating except Experimental results confirm robustness proposed against datasets its effectiveness clustering.