Please use this identifier to cite or link to this item:
|Title:||An Adaptive Hybrid Algorithm For Digital Image Copy-Move Forgery Detection|
|Authors:||Lamba, Amanjot Kaur|
|Keywords:||image forgery;Block based;Keypoint based|
|Abstract:||Due to the development of sophisticated cameras and image editing tools, digital image tampering techniques are frequently used without leaving visual cues behind. Digital image copy-move forgery is said to be an image manipulation which involves copying and pasting of certain section (or sections) within the same digital image. Generally, this is done with intention of hiding important information or providing false information in an image. This motivates a need for forgery detection systems that are transparent to such manipulations and can disclose whether a given image has been morphed just by investigating the dummy image. Several methods have been presented for the copy-move forgery detection in recent years. Nearly all of the existing block-based methods are computationally expensive and robust to noise addition, JPEG compression, but are susceptible to geometrical attacks like rotation, translation and scaling. On the other hand, keypoint-based detection techniques are computationally efficient as well as perform better under geometrical attacks in comparison with block-based methods but suffer from low recall rate. The proposed technique is a hybrid one which incorporates both block-based and keypointbased schemes in order to deal with their drawbacks. Focus of the proposed thesis work is on achieving 100% precision and recall at image level copy move forgery detection using adaptive algorithm. Firstly, adaptive image segmentation is performed on the test image resulting in image patches followed by detection and extraction of features of these patches. These features are matched patch-wise to obtain suspected keypoint pairs. An adaptive keypoint matching algorithm is used to extract matched keypoint pairs from the suspected keypoint pairs. Finally, an adaptive forgery region extraction is used to locate similar areas in the test image. The evaluation results demonstrate that the proposed hybrid scheme is more robust under plain as well as various challenging situations such as down-sampling, up-scaling, downscaling and JPEG compression than the prior state-of-the-art techniques. The proposed scheme achieved improved results with 100% precision, 100% recall and 100% F1 score at image level, while 95.01% precision, 87.18% recall and 90.92% F1 score at pixel level under plain copy-move attack. The proposed adaptive scheme can be extended to videos in future.|
|Appears in Collections:||Masters Theses@ECED|
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.