Image-format-independent Tampered Image Detection Based on Overlapping Concurrent Directional Patterns and Neural Networks

doi:10.1007/s10489-017-0893-4

機構典藏 > College of Engineering > Graduate Institute & Department of Computer Science and Information Engineering > Journal Article > Item 987654321/120008

Please use this identifier to cite or link to this item: https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/120008

Title:	Image-format-independent Tampered Image Detection Based on Overlapping Concurrent Directional Patterns and Neural Networks
Authors:	Wu, M. L.;Fahn, C. S.;Chen, Y. F.
Keywords:	Digital image forensics;Digital image authentication;Tampered image detection;Artificial neural network
Date:	2017-03-13
Issue Date:	2021-03-04 12:13:11 (UTC+8)
Publisher:	Springer Netherlands
Abstract:	With the advancement of photo editing software, digital documents can easily be altered, which causes some legal issues. This paper proposes an image authentication method, which determines whether an image is authentic. Unlike many existing methods that only work with images in the JPEG format, the proposed method is image format independent, implying that it works with both noncompressed images and images in all compression formats. To improve the authentication accuracy, some strategies, such as overlapping image blocks only on concurrent directions, using a two-scale local binary pattern operator, and choosing the mean deviation instead of the standard deviation, are applied. A back-propagation neural network (BPNN) is used instead of support vector machines (SVMs) for classification to make online learning easier and achieve higher accuracy. In our experiments, we used the CASIA Database (CASIA TIDE v1.0) of compressed images and the Columbia University Digital Video Multimedia (DVMM) dataset of uncompressed images to evaluate our image authentication method. This benchmark dataset includes two types of image tampering, namely image splicing and copy–move forgery. Experiments were performed using both the SVM and BPNN classifiers with various parameters. We determined that the BPNN achieved a higher accuracy of up to 97.26 %.
Relation:	Applied Intelligence 47(2), pp. 347-361
DOI:	10.1007/s10489-017-0893-4
Appears in Collections:	[Graduate Institute & Department of Computer Science and Information Engineering] Journal Article

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	40	View/Open

Loading...