Image splicing manipulation location by multi-scale dual-channel supervision

The swift growth of diverse editing software has resulted in image splicing manipulation becoming more complex, the discovery of a meticulously crafted splice forgery in digital images poses a significant challenge for both humans and machines. Existing image splicing manipulation detection algorith...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Multimedia tools and applications 2024-03, Vol.83 (11), p.31759-31782
Hauptverfasser: Hu, Jingyun, Xue, Ru, Teng, Guofeng, Niu, Shiming, Jin, Danyang
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:The swift growth of diverse editing software has resulted in image splicing manipulation becoming more complex, the discovery of a meticulously crafted splice forgery in digital images poses a significant challenge for both humans and machines. Existing image splicing manipulation detection algorithms have low localization accuracy and poor detection of small manipulation areas. In this paper, we proposed an end-to-end effective image manipulation location method based on a multi-scale and dual-channel model, MD_Unet. First, a dual-channel encoding network model is constructed. Adding a high-pass filtering branch containing SRM filters and Gabor filters at the input of the model and helps it to learn the manipulation traces of the image. Secondly, the dual-channel features are fused using an improved multi-scale pyramid pooling module. Then, Squeeze-Excitation is introduced to recalibrate the fused features so that the network pays more attention to splicing manipulation-related features. Finally, the fused feature map is input to the decoder, and the predicted image is decoded layer by layer to segment the manipulation region. We have performed extensive experimental validation and powerfully demonstrate the efficacy of the proposed approach.
ISSN:1573-7721
1380-7501
1573-7721
DOI:10.1007/s11042-023-16705-y