All Times ET
Virtual
Detecting Fake Images via Multiscale Methods in High-Dimensional Data (304178)
Donghoh Kim, Sejong UniversityHee-Seok Oh, Seoul National University
*Minsu Park, Samsung Medical Center
Keywords: Image analysis, Intraclass correlation, Ridge regression, Wavelet packet analysis
In this paper, we propose wavelet-based procedures to identify the difference between images, including portraits and handwriting. The proposed method is based on an inventive combination of multiscale methods with a regularization technique. Beverly and Caroline (2006) introduced how to detect forged handwriting via wavelets and summary statistics. We mainly expand the scope of their method to general images and improve the results considerably. Key features that distinguish the proposed method from existing ones are two-fold: (1) a multiscale framework formed by wavelets and wavelet packets gives a rich and flexible structure that allows adaptation to signals (images) with various spatial structures; and (2) a regularized regression technique is capable of handling to high-dimensional data, which might represent images of interest properly and extend the class of images considerably. We demonstrate the promising empirical evidence of the proposed method through various experiments.