Exploring duplicated regions in natural images

Area of Science:

• Computer Vision
• Image Forensics
• Digital Signal Processing

Background:

• Image region duplication is a prevalent image manipulation technique.
• Existing detection methods often rely on principal component analysis (PCA).

Purpose of the Study:

• To propose a robust image duplication detection approach using discrete wavelet transform (DWT) and kernel principal component analysis (KPCA).
• To develop a technique for detecting flip and rotation forgeries.
• To compare the proposed method's performance against conventional PCA-based approaches.

Main Methods:

• Utilizing DWT and KPCA for robust image block feature extraction.
• Employing lexicographic sorting of block features for similarity matching.
• Implementing an automatic segmentation technique for duplicated regions.
• Extending the algorithm with global geometric transformation for flip/rotation forgery detection.

Main Results:

• The proposed method demonstrates promising results on natural images.
• Wavelet-based features outperform PCA/KPCA in noiseless/uncompressed domains regarding precision and recall.
• KPCA-based features show excellent performance in noisy and JPEG compressed environments.

Conclusions:

• The DWT and KPCA-based approach offers a robust solution for image duplication detection.
• The extended method effectively identifies flip and rotation forgeries.
• The choice between DWT and KPCA features depends on the image's condition (noiseless vs. compressed/noisy).