الفهرس 
Chapter 1: Introduction To Image Forgery DetectionOnly 14 pages are availabe for public view
 |  | from | 120 |  |  |
| | | from | 120 | | |
Abstract
Digital images are widely used in today’s technological era because the internet, magazines, newspapers or scientific journals are the most common means of conveying information. They are used as strong evidence against different crimes and as evidence for a variety of purposes. Capturing, creating, and modifying images has become easier and simpler due to the improvement of image processing and editing tools. There are different types of image forgery, but copy-move forgery is the most common. In this type, a part of an image is copied, then pasted into the original image with the goal of hiding something significant or presenting a fake scene. Because the copied areas are identical to the original, essential components like noise and brightness will merge with the background, making it more challenging for experts to detect the change. Traditional methods of detecting copy-move forgery are extremely time-consuming.
In this thesis, we have explored the performance of the deep learning (DL) model, which uses CNN (convolutional neural networks) with CASIA v1.0, CASIA v2.0 and Columbia as training and test datasets, respectively. These results clearly confirm that the CNN model is accurate with different data sets.
Then, we have compared the test of pre-trained CNN model with individual ML algorithms in Copy-Move Forgery Detection (CMFD) and compared between them. The ML classifiers, namely Support Vector Machines (SVM), K-Nearest Neighbor (KNN), Decision Tree (DT), Random Forest (RF), and Naïve Bayes (NB), with pre-trained CNN model for feature extraction. The experiments showed that using CNN with KNN classifier achieved better CMFD accuracy, precision, recall and F-score with three data sets. They also indicated that features extracted by pre-trained model are effective with the three data sets comparing with [Yuan et al., 2016].
Based on our exploration of the performance the CNN model with KNN in CMFD, we conclude that the performance of DL model with KNN in CMFD is significantly better than that of other classifiers.