Reduced complexity correlation-based blind clustering of camera fingerprints

We consider the problem of common image source detection on a quite large set of pictures (1000 images) based on similarities of their noise patterns. We experiment with different clustering techniques in a blind scenario, when the informations about the cameras are not available. We propose a method for clustering combining Fair clustering on RCIC and FICFO algorithms, based on the similarity scores. We add a refinement step and an attraction stage in order to obtain less groups and better performances. We evaluate our method on the Dresden images database showing that the methodology is highly effective.