1. UKM Learning and Research Repository
2. PHD Thesis / Tesis PHD
3. Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat

Title:	JPEG image classification via DCT coefficient analysis for data carving and bulk image
Authors:	Nadeem N.I. Alherbawi (P61109)
Supervisor:	Zarina Shukur, Prof. Dr.
Keywords:	Computer forensic Image processing
Issue Date:	10-Jul-2017
Description:	Digital or computer forensics can be defined as the practice of identifying, preserving, extracting, analyzing and presenting sound legal evidence from digital media such as computer hard drives. The wide spread of image processing tools and the ability to create forged images increase the need for creating methods and frameworks to deal with this problem, and with the advanced development of storage devices technologies, the digital investigator gets overwhelmed by the bulk data waiting for processing. Therefore, the need for automated content integrity verification comes into the picture. Current algorithms cover the localization (highlighting) process of the forged region for each single image, while the classification (detection) methods focus mostly on one single type of forgery. This requires the investigator involvement in examining each image under investigation or required testing the images using multiple methods to cover the different forgery attacks. This study aims to design a model that handles the case of having media storages filled with bulk images waiting to be processed by digital investigator. The model is an automatic classifier that can detect and classify a group of images with high detection rates for the most known forgery attacks and corruptions. It also propose a reduced feature vector model that has a small size. As a result, the investigator can check the images that have potential forgery or corruptions without looking into those that has no sign of manipulation. The proposed authenticity (JPEGDEF) model classify the JPEG images that contains errors or corrupted data by using Libjpeg library error handling. While using the reduced feature vector based on DCT coefficient analysis produced posterior map as an input with the proposed clipping and padding and PCA reduction to multiple machine learning methods. The study involves conducting Group of experiments to test the proposed model on established datasets includes CASIA V2, MICC-F2000 and MICC-F600. The experiments covers variant forgeries attacks such as splice and copy/move with scaling, rotation, and translation. Also, they cover different quality factor levels of the forged image compared to the source image. Result of the conducted experiments on a set of established datasets yields an excellent performance results on CASIA V2 dataset with 99.6% ,99.8%,97.7%, and 99% accuracies on KNN,SVM,Random forest and decision tree, respectively. On MICCF2000 results were 99%, 98%, 93%, and 99% accuracies on KNN,SVM,Random forest and decision tree, respectively. Additionally, on MICC-F600 results were 98%,97%,92%, and 96% accuracies on KNN,SVM,Random forest and decision tree, respectively. Those results were higher than all compared studies in both splice and copy/move cases. It also has small false positive and negative rates when compared with other studies on both splice and copy/move datasets. As a conclusion, the proposed model with this high-performance results could help investigator in real forensic application.,Certification of Master's/Doctoral Thesis" is not available
Pages:	179
Call Number:	TA1637.A378 2017 3 tesis
Publisher:	UKM, Bangi
Appears in Collections:	Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.