Enhanced Block-Based Copy-Move Image Forgery Detection Using K-Means Clustering Technique
Loading...
Date
2018-06-01
Authors
Mohammed Abdo Al-Qershi, Osamah
Journal Title
Journal ISSN
Volume Title
Publisher
Universiti Sains Malaysia
Abstract
In this thesis, the effect of feature type and matching method has been analyzed by
comparing different combinations of matching method – feature type for copy-move
image forgery detection. The results showed an interaction between some of the
features and some of the matching methods. Due to the importance of matching
process, this thesis focused on improving the matching process by proposing an
enhanced block-based copy-move forgery detection pipeline. The proposed pipeline
relied on clustering the image blocks into clusters, and then independently
performing the matching of the blocks within each cluster which will reduce the time
required for matching and increase the true positive ratio (TPR) as well. In order to
deploy the proposed pipeline, two combinations of matching method - feature type
are considered. In the first case, Zernike Moments (ZMs) were combined with
Locality Sensitive Hashing (LSH) and tested on three datasets. The experimental
results showed that the proposed pipeline reduced the processing time by 73.05% to
84.70% and enhanced the accuracy of detection by 5.56% to 25.43%. In the second
case, Polar Cosine Transform (PCT) was combined with Lexicographical Sort (LS).
Although the proposed pipeline could not reduce the processing time, it enhanced the
accuracy of detection by 32.46%. The obtained results were statistically analyzed,
and it was proven that the proposed pipeline can enhance the accuracy of detection
significantly based on the comparison with other two methods.