Copy-move forgery detection in digital images

Khayeat, Ali 2017. Copy-move forgery detection in digital images. PhD Thesis, Cardiff University. Item availability restricted.

Preview	PDF - Accepted Post-Print Version Download (15MB) | Preview
	PDF - Supplemental Material Restricted to Repository staff only Download (2MB)

Abstract

The ready availability of image-editing software makes it important to ensure the authenticity of images. This thesis concerns the detection and localization of cloning, or Copy-Move Forgery (CMF), which is the most common type of image tampering, in which part(s) of the image are copied and pasted back somewhere else in the same image. Post-processing can be used to produce more realistic doctored images and thus can increase the difficulty of detecting forgery. This thesis presents three novel methods for CMF detection, using feature extraction, surface fitting and segmentation. The Dense Scale Invariant Feature Transform (DSIFT) has been improved by using a different method to estimate the canonical orientation of each circular block. The Fitting Function Rotation Invariant Descriptor (FFRID) has been developed by using the least squares method to fit the parameters of a quadratic function on each block curvatures. In the segmentation approach, three different methods were tested: the SLIC superpixels, the Bag of Words Image and the Rolling Guidance filter with the multi-thresholding method. We also developed the Segment Gradient Orientation Histogram (SGOH) to describe the gradient of irregularly shaped blocks (segments). The experimental results illustrate that our proposed algorithms can detect forgery in images containing copy-move objects with different types of transformation (translation, rotation, scaling, distortion and combined transformation). Moreover, the proposed methods are robust to post-processing (i.e. blurring, brightness change, colour reduction, JPEG compression, variations in contrast and added noise) and can detect multiple duplicated objects. In addition, we developed a new method to estimate the similarity threshold for each image by optimizing a cost function based probability distribution. This method can detect CMF better than using a fixed threshold for all the test images, because our proposed method reduces the false positive and the time required to estimate one threshold for different images in the dataset. Finally, we used the hysteresis to decrease the number of false matches and produce the best possible result.

Item Type:	Thesis (PhD)
Date Type:	Submission
Status:	Unpublished
Schools:	Computer Science & Informatics
Subjects:	Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Date of First Compliant Deposit:	24 November 2017
Last Modified:	18 May 2021 11:45
URI:	https://orca.cardiff.ac.uk/id/eprint/107043

Actions (repository staff only)

Edit Item