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

The spread of image editing tools demonstrates how modern mixed-media technology enables changes in digital images. Such easy access raises severe moral and legal concerns around the potential for malicious image editing. Overcoming this difficulty will need the development of innovative approaches for the quick detection of changes in high-quality photographs. This paper proposes a new way to solve this problem by analyzing chrominance discontinuities in spliced regions, DWT, and unique histograms based on local binary patterns. To start extracting the luminance and chrominance components, we change the input image's color space from RGB to YCBCR. Then, using discrete wavelet transformation, the blue and red chromaticity levels were converted into wavelet bands. We compute histograms using the CB and CR DWT high-frequency bands. The next step is to use feature fusion methods to merge the CB and CR feature vectors from each high-frequency band after we change the histograms into vectors. Finally, we train a Support Vector Machine (SVM) using the combined color characteristics. A binary SVM trained to identify spliced images between original and spliced images has been produced. Improving upon existing methods, the proposed method achieved up to 98.49% accuracy on CASIA v1.0, outperforming existing benchmarks, that is, 97.33% on DVMM and 98.25% on Casiav2.0, thereby enhancing splicing forgery detection. This method contributes to media forensics by providing a reliable tool for detecting tampered images, which holds significant relevance in legal investigations and digital content authentication.