Thesis Title: Improved Deep Hiding / Extraction Algorithm to Enhance Payload Capacity and Security Level of Hidden Information
Author: Marawah Ahmed Abdullah Ail, Supervisor: Prof. Nameer N. EL-Emam, Dr. Ali F. AL-Azawi, Year: 2019
Faculty: Information Technology, Department: Computer Science

Abstract: Data-hiding by using steganography algorithm becomes a significant technique to prevent illegal users from contacting secret data. In this thesis, Deep Hiding/extraction (IDHEA) algorithm has been improved to hide secret data in color images, enhance payload capacity and reduce time complexity as much as possible. This algorithm is based on modified LSB (MLSB). It distributes secret data randomly on cover-image and replaces the number of bit per byte (Nbpb) up to 4 bits and makes it hard to access hidden data by unauthorized users. The proposed IDHEA algorithm specifies number of levels randomly; each level uses a color image and from a level to the next one, the image size is expanded. We start with a small size of the cover-image and increase the size gradually or suddenly according to the enlargement ratio. In addition, the appropriate randomization approach is implemented for pixel selection to hide secret data at each level. Lossless image compression based on run-length encoding algorithm and Gzip to enable the size of hiding at the next level. The Advanced Encryption Standard (AES) is applied at each level to increase security and protection against attackers. Thus, the effectiveness of the proposed IDHEA algorithm has been applied at unlimited levels to confuse attackers, and compress a deep sequence of images in one image. The main square error (MSE), the signal to noise ratio (SNR), and the peak signal to noise ratio (PSNR) are calculated to check the performance of the proposed algorithm. The experimental results are discussed regarding different performance measures; these metrics are demonstrated the effectiveness of the proposed algorithm in terms of its embedding capacity and imperceptible level. Comparisons between the proposed approach and the other works have been implemented, and they appear that the proposed algorithm is better than the MDLSB & LSB algorithms and it is working against statistical and visual attacks. Furthermore, the results confirm that the stego-image with high imperceptibility has been reached even if the stego-image holds a large amount of data that reach twelve bits per pixels (12-bpp) at certain pixels. In addition, it is confirmed that the proposed algorithm can embed secret data efficiently with better visual quality.

Keywords: Data-hiding, IDHEA, MLSB, Nbpb, AES

Thesis Title: Extracting Arabic Text Summarization on Social Media for Trending Topics
Author: Alaa Salim Malkawi, Supervisor: Dr. Mouyad A. Fadhil Al-Athami, Year: 2019
Faculty: Information Technology, Department: Computer Science

Abstract: Automatic text summarization (ATS) is one of the most important challenges in the computer science and Natural Language Processing (NLP). In this research we aim to build a dynamic summarization approach on social media using twitter. The aim of the proposed approach is to extract summarizations from the interesting topics for twitter' users using recent tweets, where the proposed approach will be connected directly with twitter using twitter' API. The proposed approach will be hybrid using each of Latent Dirichlet Allocation (LDA) algorithm and N-gram, where LDA algorithm will be used to extract the most important keywords from collected dataset through calculating the frequencies for words in each topic. While N-gram will be used to find the best ordering for words in the suggested topic. The proposed system produced best results compared to other summarization techniques. Generally, ROUGE evolution metric (F-measure) = 65% which indicate that the proposed system achieves good level of summarization performance compared to the manual summarization.

Keywords: Arabic text summarization (ATS), Twitter API, Latent Dirichlet Allocation (LDA), N-gram.

Thesis Title: Controlling a Robotic Arm Using Image-Based Feedback and Deep Reinforcement Learning
Author: Abdullah Al-Zabt, Supervisor: Dr. Tarek A. Tutunji, Year: 2019
Faculty: Engineering, Department: Mechatronics Engineering

Abstract: This work presents a methodology for implementing a robotic arm task using feedback images and Deep Reinforcement Learning (DRL). The task is for the robotic arm to reach a desired target without having information about the target position nor the mathematical model of the robot (i.e. model-free). The methodology employs information from images to represent both: the state of the environment and the reward function. The agent (i.e. controller) is based on a convolutional neural network structure, with raw pixels as inputs and value function estimating future rewards as outputs. The agent type is value-based that uses a modified version of the famous Deep Q-Network (DQN) algorithm to gain knowledge and control over the environment (i.e. plant). The original DQN algorithm uses Temporal-Difference (TD) target along with experience replay while the modified DQN algorithm uses the same principle, but adds actions-sequence capture feature. This feature improves the agent’s performance by memorizing the highest rewarded sequence of actions episode and replay it over whenever the agent experiences a sequence of failed episodes. Several simulation runs are used to validate the proposed methodology as the modified version of DQN is tested on five cases with three different robot structures. This work used Python for coding the algorithm and V-Rep for providing the graphical modelling of the robotic arm. Results show that the agent succeeded in controlling the robot’s position and reaching the desired target using only images as inputs.

Keywords: Deep Reinforcement Learning, convolutional neural network, DQN

Thesis Title: The Feminist Response to the Traditional Advertisement
Author: Aseel Nizam Muhammad Jaber, Supervisor: Dr. Areen Khalifeh, Year: 2019
Faculty: Arts, Department: English Language and Literature

Abstract: This study will examine the language used to address women in both traditional magazines and feminist magazines. Specifically, it examines the difference between the two types of magazines and the effect their languages have on women. Furthermore, it aims at showing how some magazines like Glamour and Cosmopolitan claim to be feminist. The results show that these magazines are not, because they trigger low self-esteem in women and set high standards that average women cannot keep up with. In contrast with traditional magazines, Wear your voice and Bust proved there are magazines that advocate women's empowerment and equality. Both magazines shed light on race related issues women face and sisterhood. Adding to that, the research provides word count tables that show the most frequent and repeated words in each article or advertisement. The analysis is divided into two chapters. Each is dedicated to two different magazines. The first chapter analyzes the rhetorical and the linguistic features used in traditional magazines and how they use language to persuade their audience into buying products or adapt to new ideas. And in chapter two, the study will investigate the rhetorical features used in the feminist magazines to see how both types of magazines differ in the way they address their audience. In the light of the conclusion, a summary of the findings will be included: Traditional magazines aim at selling products and make money from their audience, whereas feminist magazines aim at empowering women and helping them stand up against all forms of discrimination.

Keywords: traditional, magazines, advertisement

Thesis Title: Real-Time Monitoring and Fault Detection of Photovoltaic System
Author: Hudefah Zuhair Al-Kashashneh, Supervisor: Prof. Kasim Al-Aubidy, Year: 2018
Faculty: Engineering, Department: Mechatronics Engineering

Abstract: This thesis presents design and implementation of monitoring and fault detection for photovoltaic renewable energy system. A real-time monitoring unit, based on wireless sensor networks, is designed to monitor the required variables related to each module and unit in the PV system. An intelligent algorithm based on fuzzy logic has been developed to detect and locate faults. A mathematical model together with an experimental prototype is designed to evaluate the performance of the proposed PV system under normal and faulty conditions such as normally shading and one faulty PV module, normally shading and two faulty PV modules, one faulty PV module only, and two faulty PV modules only. The obtained results from both simulated and experimental systems confirm that the implemented monitoring and fault detection system perform the required tasks with acceptable accuracy and cost. Applying real-time remote monitoring is easier to specify faults and observe the overall performance of any solar PV system.

Keywords: Real-Time, Monitoring, Fault Detection, Photovoltaic System