Enhanced image-splicing classification: A resilient and scale-invariant approach utilizing edge-weighted local...

Area of Science:

• Digital Forensics
• Computer Vision
• Media Forensics

Background:

• The proliferation of image editing tools raises concerns about malicious digital image manipulation.
• Effective detection of altered high-quality photographs is essential for legal and authentication purposes.

Purpose of the Study:

• To develop an innovative approach for the rapid and accurate detection of spliced regions in digital images.
• To enhance the reliability of media forensics tools for digital content authentication.

Main Methods:

• Color space conversion from RGB to YCBCR to extract luminance and chrominance components.
• Discrete Wavelet Transformation (DWT) applied to chrominance bands (CB and CR) to obtain high-frequency features.
• Feature fusion of histogram vectors derived from DWT bands, followed by Support Vector Machine (SVM) training for binary classification.

Main Results:

• The proposed method achieved high accuracy rates: 98.49% on CASIA v1.0, 97.33% on DVMM, and 98.25% on CASIA v2.0.
• The technique effectively distinguishes between original and spliced images, outperforming existing benchmarks.
• The developed binary SVM provides a reliable tool for identifying image splicing forgeries.

Conclusions:

• The novel approach based on chrominance discontinuities, DWT, and LBP histograms offers a robust solution for splicing forgery detection.
• This method significantly contributes to media forensics, offering a reliable tool for legal investigations and digital content authentication.
• The high accuracy demonstrates the effectiveness of analyzing chrominance features for detecting sophisticated image manipulations.