Persistent Homology to Detect Fake Face Images

Fri 12 Apr 2019
14:15 - 15:15


Franciscan Building Room F06
Room F06, Verney Park


School of Computing
+44 (0)1280 828322

Aras AsaadA talk by Aras Asaad as part of our work in progress seminars.

Persistent Homology (PH) is the main machinery in Topological Data Analysis that depends on filtration of spaces to measure the topological information. I will present a recent PH-based morph detector which rely on the discriminating power of the pattern of change in the topological features over the entire sequence of the nested shapes constructed form image landmarks. In addition, two different types of image landmarks proposed to construct simplicial complexes that are sensitive to morphing. Two large digital datasets used to test the discriminating power of the proposed PH-based morph detector together with a print-scanned dataset.