Image Clustering by Source Camera via Sparse Representation

The discovery of clusters of images sharing the same origin based on camera fingerprints, such as Sensor Pattern Noise (SPN), plays an important role in the realm of multimedia forensics. In this work, we present a new approach for grouping images having the same origin based on Sparse Subspace Clustering (SSC). Due to noisy nature of data, we propose to find sparse representation of each camera fingerprint by solving ℓ1-regularized least squares, and to estimate appropriate parameter in a data-driven fashion. These sparse representations characterize underlying data segmentation. Experimental results confirm the effectiveness of our approach in comparison with existing works.