Citation

Jiao, Li Xian (2026) Digital image blind forensics: Copy-Move Forgery Detection. PhD thesis, Multimedia University.

Full text not available from this repository.

Abstract

Copy-Move Forgery Detection (CMFD) is a fundamental task in digital image forensics, aiming to identify duplicated regions within the same image. Even though several CMFD algorithms have been proposed, current ones still struggle to strike the right balance between accuracy, robustness, and interpretability when images are captured under challenging conditions, such as changing lighting, compression artifacts, and non-rigid geometric transformations. To address these issues, this dissertation proposes a progressive CMFD framework integrating handcrafted and deep features, composed of three detection models. First, a hybrid handcrafted CMFD algorithm based on adaptive uniform keypoint detection and hybrid moment feature description is developed. It uses an adaptive uniform FAST detector and the hybrid Phase-Locked, Multi-Scale Fast Quaternion Generic Polar Complex Exponential Transform (PLMS-FQGPCET) + Root-Normalized Histogram of Oriented Gradients (Root-HOG) descriptor to find things quickly and accurately. Second, a hybrid learning framework, Adaptive Uniform and Self-Matching Refinement (AUR-SMURef), is proposed that combines handcrafted geometric interpretability with deep feature refinement. It combines AKAZE with Adaptive Uniform Refinement (AKAZE-AUR) detection, Root Normalized Scale-Invariant Feature Transform (RootSIFT) + Accelerated KAZE Features with the Modified Local Difference Binary Descriptor (AKAZE-MLDB) descriptors, and theSelf-Matching Union (SMU) mechanism to ensure that geometric consistency is strong. At the same time, a lightweight Refine module achieves coarse-to-fine pixel-level optimization using Top-K correlation and GrabCut refinement. Finally, a source target distinction and restoration framework based on physical priors and statistical evidence fusion is introduced. It establishes a physically interpretable model for differentiating donor and target regions and reconstructs tampered contents through a patch-based optimization process. Extensive studies on the FAU, CVIP, GRIP, and CoMoFoD datasets show that the proposed methods are superior to current CMFD methods in object detection, robustness, and edge precision. The final integrated framework also establishes a single, clear pipeline from forgery detection to semantic restoration. This makes it a dependable and flexible solution for digital image forensics.

Item Type:	Thesis (PhD)
Additional Information:	Call No.: TA1637 .J53 2026
Uncontrolled Keywords:	Image processing—Digital techniques
Subjects:	T Technology > TA Engineering (General). Civil engineering (General) > TA1501-1820 Applied optics. Photonics
Divisions:	Faculty of Computing and Informatics (FCI)
Depositing User:	Ms Nurul Iqtiani Ahmad
Date Deposited:	05 Oct 2026 06:03
Last Modified:	05 Oct 2026 06:03
URII:	http://shdl.mmu.edu.my/id/eprint/16878

Downloads

Edit (login required)