Please use this identifier to cite or link to this item:
|Title:||Statistical image source model identification and forgery detection||Authors:||Cao, Hong||Keywords:||DRNTU::Engineering::Electrical and electronic engineering::Computer hardware, software and systems||Issue Date:||2010||Source:||Cao, H. (2010).Statistical image source model identification and forgery detection. Doctoral thesis, Nanyang Technological University, Singapore.||Abstract:||Advances of digital technology have given birth to numerous unprecedented tools, which make image forgery easier than ever. To restore the traditional trustworthiness on digital photos, image forensics analyses that can reliably tell the origin, integrity and authenticity of a given image are urgently needed. In this thesis, we propose several new image forensics tools for: 1) Accurate detection of image demosaicing regularity as a general type of image forensics features; 2) Identification of various common image sources including digital still camera models, RAW conversion tools and the low-end mobile camera models; 3) Universal detection of a wide range of common image tampering and 4) Prevention of the image recapturing threat. These forensics tools help expose common image forgeries, especially those easy-to-make forgeries, which can hardly be seen directly by human eyes. The common theme behind our proposed forensics tools is through statistical detection of some intrinsic image regularity or tampering anomalies. Our tools are not constrained by the strict end-to-end protocol requirement such as prior image hash computation or prior information hiding; hence have bright application prospect. Advanced pattern classification techniques including feature reduction techniques and nonlinear classification methods are employed to achieve extremely good and better forensics performances than state-of-the-arts forensics methods based on large-scale experimental tests. In the universal image tampering detection framework, we have also proposed a novel FusionBoost learning to combine a set of lightweight probabilistic tampering detectors into a strong ensemble tampering detector. Experimental results demonstrate its competency over the conventional boosting algorithms or fusion methods.||URI:||https://hdl.handle.net/10356/42662||DOI:||10.32657/10356/42662||Fulltext Permission:||open||Fulltext Availability:||With Fulltext|
|Appears in Collections:||EEE Theses|
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.