Quantization lattice estimation for multimedia forensics

The most widely used information lossy source coding schemes for multimedia signals rely on the quantization of the content samples in a linearly transformed domain. A number of forensic applications (e.g., processing history estimation, tampering detection, software identification) can be posed as the estimation of the equivalent lattice quantizer from the observed samples. We present a new lattice estimation algorithm based on the observation of noisy points of the lattice. Although inspired by the work of Neelamani et al., our scheme uses the so-called “dual lattice” to achieve significant performance improvements with respect to its predecessors as measured by the number of vectors of the lattice basis that can be correctly estimated. Such performance improvement is even more dramatic when small pieces of the contents are considered, which indeed is especially relevant for forensic applications.