Multi-Level Dense Descriptor and Hierarchical Feature Matching for Copy-Move Forgery Detection

Xiuli Bi, Chi Man Pun, Xiao Chen Yuan

Research output: Contribution to journalArticlepeer-review

78 Citations (Scopus)

Abstract

In this paper, a Multi-Level Dense Descriptor (MLDD) extraction method and a Hierarchical Feature Matching method are proposed to detect copy-move forgery in digital images. The MLDD extraction method extracts the dense feature descriptors using multiple levels, while the extracted dense descriptor consists of two parts: the Color Texture Descriptor and the Invariant Moment Descriptor. After calculating the MLDD for each pixel, the Hierarchical Feature Matching method subsequently detects forgery regions in the input image. First, the pixels that have similar color textures are grouped together into distinctive neighbor pixel sets. Next, each pixel is matched with pixels in its corresponding neighbor pixel set through its geometric invariant moments. Then, the redundant pixels from previously generated matched pixel pairs are filtered out by the proposed Adaptive Distance and Orientation Based Filtering method. Finally, some morphological operations are applied to generate the final detected forgery regions. Experimental results show that the proposed scheme can achieve much better detection results compared with the existing state-of-the-art CMFD methods, even under various challenging conditions such as geometric transforms, JPEG compression, noise addition and down-sampling.

Original languageEnglish
Pages (from-to)226-242
Number of pages17
JournalInformation Sciences
Volume345
DOIs
Publication statusPublished - 1 Jun 2016
Externally publishedYes

Keywords

  • Color Texture Descriptor
  • Copy-Move Forgery Detection (CMFD)
  • Hierarchical Feature Matching
  • Invariant Moment Descriptor
  • Multi-Level Dense Descriptor (MLDD)

Fingerprint

Dive into the research topics of 'Multi-Level Dense Descriptor and Hierarchical Feature Matching for Copy-Move Forgery Detection'. Together they form a unique fingerprint.

Cite this