Detection Theory

Catalog description

Course on statistical signal detection techniques and their applications in digital forensics. Lectures are interleaved with hands-on projects to show the students the usefulness of theoretical concepts introduced during lectures. Students will become familiar with an important forensic method for digital media that identifies digital cameras and manipulated images using the so-called sensor fingerprint. From the technical point of view, the course contains the following topics: Binary and multiple hypothesis testing, Neyman-Pearson theorem, Bayesian risk minimization, likelihood ratio test. Detection of deterministic signals and random signals, matched and generalized matched filter. Gaussian and non-Gaussian noise. Detection of signals with unknown parameters and unknown noise parameters, generalized likelihood ratio test, universally most powerful test. Basics of parameter estimation theory. The techniques will be demonstrated via hands-on projects drawn from the field of digital forensics (digital watermarking and sensor fingerprints). Prerequisite: Basic statistics. Offered every spring semester.