Detection and localization of splicing on remote sensing images using image-to-image transformation

Remote sensing images can be easily tampered with user-friendly tools to hide important information. It necessitated the development of automatic splicing detection methods. The existing few methods concentrate on the semantic content in images for tamper detection and are not robust. On the contrar...

Full description

Saved in:
Bibliographic Details
Published in:Applied intelligence (Dordrecht, Netherlands) Netherlands), 2023-06, Vol.53 (11), p.13275-13292
Main Authors: Alsughayer, Rawan, Hussain, Muhammad, Saeed, Fahman, AboalSamh, Hatim
Format: Article
Language:English
Subjects:
Artificial Intelligence
Blurring
Computer Science
Generative adversarial networks
Localization
Machines
Manufacturing
Mechanical Engineering
Processes
Random noise
Remote sensing
Remote sensors
Satellite imagery
Semantics
Transformations (mathematics)
Citations: Items that this one cites
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Remote sensing images can be easily tampered with user-friendly tools to hide important information. It necessitated the development of automatic splicing detection methods. The existing few methods concentrate on the semantic content in images for tamper detection and are not robust. On the contrary, we hypothesize that residual noise is independent of the semantic content and embeds the tampering traces; it is helpful for splice detection. In view of this, we focus on residual noise and formulate the problem as an image-to-image transformation, and model it using a U-net architecture. To suppress semantic content and extract the residual noise, we introduce a constrained convolutional layer in the U-net model. The model processes the input image and yields a map that localizes tampering in case of splicing. The model is trained using the conditional generative adversarial network (cGAN) framework. The loss function is composed of the cross-entropy, the Jaccard, and the end-point error (EPE) loss functions to enhance the detection and localization of tampered regions. To evaluate the proposed method, we develop a new dataset containing remote sensing images from different satellites and aerial sensors. The model detects splicing at pixel and image levels with high accuracy. It shows good robustness against well-known post-processing operations, including Gaussian blurring (GB) and white additive Gaussian noise (WAGN).
ISSN:0924-669X
1573-7497
DOI:10.1007/s10489-022-04126-7