Department of Computer Science and Engineering

Browse

Now showing 1 - 18 of 18

A new approach of bangla news document summarization
(University of Dhaka, 2018-09-24) Haque, Md. Majharul
The object of this research work is to propose a new method of automatic Bangla news document summarization. It is noticeable that the existing English text summarization systems may not be directly applicable for Bangla for the complexities of Bangla language in grammatical rules, structure of sentences, diﬀerent placement of subject and object, etc. Again, the research work for Bangla language processing is diﬃcult because there is hardly any automated tool to facilitate research work. In this challenging situation, a new approach for Bangla news document summarization has been presented here by introducing pronoun replacement and an improved version of sentence ranking. Major parts of this approach are (i) preprocessing the input document, (ii) word tagging, (iii) replacement of pronoun, and (iv) sentence ranking. Replacement of pronoun has been accomplished here for the rst time to minimize the dangling pronoun in summary. After replacing pronoun, sentences are ranked by considering (i) term frequency, (ii) sentence frequency, (iii) numerical gures (presented in words and digits), and (iv) title words. If two sentences has at least 60% cosine similarity, frequency of larger sentence is increased and remove smaller sentence which eliminates redundancy. Moreover, the rst sentence has been specially considered for containing any title word. Again, numerical gure has beenidenti ed from words and digits to assess the importance of sentences despite the variety of forms for any numerical gure in Bangla. For achieving the target of this proposed method, 3000 news documents have been analyzed and some Bangla grammar books have been studied. The eﬀect of each incorporated feature has been demonstrated with step by step performance analysis. From the evaluation results of the proposed method, the F-measure scores for ROUGE-1 and ROUGE-2 have been found as 0.6003 and 0.5708 respectively and the accuracy of pronoun replacement has been found as 71.80%. The proposed method has minimized the dangling pronoun in summary for 89.75% than the latest Bangla text summarization system. Again, the text summarization performance of the proposed method has been observed as 9.39% (based on ROUGE-1 F-measure score) and 12.52% (based on ROUGE-2 F-measure score) better than the latest existing method.
Analysis and Synthesis of Bangla Phonemes for Computer Speech Recognition
(© University of Dhaka, 2025-05-27) Hossain, Syed Akhter
AnEfficient Worker Selection and Task Allocation Scheme for Mobile Crowdsourcing System
(© University of Dhaka, 2025-04-10) Huq, Farhana
The rapid proliferation of mobile computing devices has facilitated mobile crowd sourcing systems (MCS) to emerge as a powerful model for allocating tasks to a distributed workforce. The efficiency of such systems is significantly dependent on two critical factors: the selection of appropriate workers and the distribution of tasks to workers. Addressing these challenges is crucial for optimizing task accom plishment, minimizing task completion time, and improving satisfaction for both workers and task requesters. Online Food delivery (OFD), a specialized applica tion of mobile crowdsourcing, represents a rapidly evolving e-business application that leverages cloud computing data centers, playing a crucial role in meeting the demands of urban lifestyles. Assigning food delivery orders to workers in a way that optimizes service quality by maximizing workers’ profit while minimizing order completion time to simultaneously enhance customer satisfaction is a challenging problem. The growing complexity of order fulfillment features and rising expecta tions for service quality have made the task of efficiently assigning riders for long distance, cross regional deliveries a major engineering challenge. Existing studies in the literature are limited as they tend to focus solely on either reducing order completion time or minimizing order delivery costs, failing to meet both objectives thoroughly. Prior research frequently depended on conventional order allocation i Abstract ii methods that either failed to notice varying capacities, or utilized non-intelligent systems that inadequately addressed fluctuating order demands and service delays. In this dissertation, we aim to develop a framework for optimal assignment of food delivery orders for both regional and cross regional online food delivery sys tem. At first, we propose a framework for the optimal assignment of food delivery orders to workers, formulated as a multi-objective linear programming (MOLP) problem, which balances the trade-off between maximizing worker profit and en hancing customer satisfaction, providing a comprehensive solution that addresses both objectives simultaneously. A Water Wave Optimization based metaheuristic assignment algorithm is developed for the online Food Delivery system that bal ances worker’s profit and customer satisfaction by selecting appropriate workers to complete the orders. The experiment results show the assignment significantly improves the performance of the OFD system in terms of average worker profit, customer satisfaction, average service time. The second contribution of this thesis is the development of the system compo nents and functional architecture of a cross regional online food delivery (XROFD) system, designed to facilitate real-time deliveries across regions efficiently. A Mixed Integer Linear Programming (MILP) optimization framework has been designed to minimize the total service time and delivery cost for cross regional orders. This framework divides a large OFD area into multiple regions and utilizes both transfer vehicles and riders to optimize deliveries. In the XROFD system, food orders are primarily transported by transfer vehicles, such as pickup vans, from restaurants to meeting point locations. To enhance the predictive accuracy of the XROFD sys tem, we incorporate advanced machine learning techniques. Specifically, we employ the Long Short-Term Memory (LSTM) model to forecast regional order demands Abstract iii accurately, reflecting the dynamic nature of the marketplace. Additionally, Ex treme Gradient Boosting (XGBoost) is tailored to dynamically predict travel times from restaurants to customer locations, facilitating more precise scheduling and re source allocation within the MILP framework. These machine-learning techniques significantly bolster the MILP framework by providing detailed, accurate predic tions that improve decision-making processes and adaptability to real-time condi tions. Proven that the above MILP is an NP-hard problem, we further enhance our approach by integrating a metaheuristic algorithm, Adaptive Large Neighbor Search (ALNS), which efficiently assigns orders to the appropriate transfer vehi cles and riders within polynomial time. Our Cross Regional Online Food Delivery (XROFD) system is meticulously designed to optimize both customer satisfaction and rider incentives. Simulation experiments confirm that the XROFD system not only reduces service times and delivery costs but also markedly enhances customer satisfaction and provides superior incentives for riders
Computer vision based Bangla sign language recognition
(University of Dhaka, 2018-10-11) Rahaman, Muhammad Aminur
The recognition of computer vision-based natural Bangla sign language in real-time is challenging due to occlusion, illumination variation, cluttered background and huge computational cost. This thesis presents a computer vision-based Bangla sign language recognition system. The proposed system contains three major modules: Bangla hand-signs segmentation and classification; Bangla sign words recognition; and hand-sign-spelled Bangla words, composite numerals and sentences recognition. In the first module, the system uses Haar-classifiers to initialize region of interest (ROI) by detecting predefined hand-signs. Then the system segments human skin-color using proposed Fuzzy rule based RGB (FRB-RGB) model. The system extracts probable binary hand-signs based on segmented skin-color pixels with specific motions. To remove noise morphological closing operation are done on the binary image. Then the system applies Gaussian smoothing filter to smooth the binary image. The system applies specific rotation followed by clipping and normalization on the binary image to make it rotation and scale invariant. Then the system generates Window-Grid Vector (WGV) by applying a 5×5 window-grid mask on the image. The system forms a feature vector based on the WGV, area and mean height of the probable hand-sign. The system classifies the hand-signs by calculating the maximum Inter-Correlation Coefficient (ICC) between test and pre-trained feature vectors. The system is trained using 4600 (10×10×46) images for 46 hand-signs of two-handed 36 Bangla alphabet and one-handed 10 numeral signs (Set-1); 3800 (10×10×38) images for 38 one-handed Bangla alphabet signs (Set-2); and 3000 (10×10×30) images for 30 hand shapes of main elements in BdSL (Set-3). The system is tested using 27600 (4600×6) images for Set-1, 22800 (3800×6) images for Set-2 and 18000 (3000×6) images for Set-3 in six different environments. The system achieves mean recognition accuracy of 95.67% for Set-1, 95.60% for Set-2 and 95.10% for Set-3 of hand-signs with average computational cost of 8.01 ms/f. In the second module, the system segments skin-like area using previous FRB-RGB model from the ROI. After noise removing and Gaussian smoothing, the system converts it into gray image. Then the system extracts Outer Boundaries (contours) using Canny edge detector and encodes it into Normalized Outer Boundary Vector (NOBV). The system recognizes sign words based on maximum Inter-Correlation Coefficient (ICC) between test NOBVs and pre-trained NOBV. The system is trained using 1800 (18×10×10) outer boundary templates for 18 BdSL words from ten signers and the system is tested using another 1800 images of the 18 BdSL words from another ten signers. The system achieves mean recognition accuracy of 90.11% with average computational cost of 26.063 ms/f. The third module interprets hand-sign-spelled BdSL into Bangla words, composite numerals and sentences using Bangla Language Modeling Algorithm (BLMA). The system tracks the ROI using Adaptive Kalman Filter (AKF) and extracts probable binary hand-signs. The system initially classifies hand-signs using NOBV. If classification score does not satisfy specific threshold of ICC then the system uses WGV based classifier. The system is trained using 5200 images for 52 hand-signs for hand-signs classification phase. The system is tested using 31200 (5200×6) images for 52 hand-signs in six different environments. The system achieves mean accuracy of 95.83% for hand-signs classification with average computational cost of 39.972 ms/f. Then the system is tested for BLMA using the hand-sign-spelled of 500 words, 100 composite numerals and 80 sentences in BdSL. For this experiment the system achieved mean accuracy of 93.50% for words, 95.50% for composite numerals and 90.50% for sentences recognition in BdSL.
Design and implementation of a reversible central processing unit
(University of Dhaka, 2018-04-05) Jamal, Laﬁfa
Reversible logic is an emerging technology. Conventional logic dissipates more power by losing bits of information whereas reversibility recovers bit loss from the unique input-output mapping. Reversible computing spans computational models that are both forward and backward deterministic. Reversible logic circuits require some constant ancilla inputs to generate the required functions. It produces some unused outputs, called garbage outputs, to maintain the reversibility of a circuit. It is important to minimize the number of ancilla inputs, number of garbage outputs and quantum cost in the design of reversible circuits. Reversible logic has promising applications in emerging computing paradigm such as quantum computing, DNA computing, optical computing, quantum dot cellular automata etc. The central processing unit (CPU) is regarded as the brain of a computer. It is a piece of hardware that carries out the instructions needed to run a computer program. Processor speed determines the performance of a computer. It performs the basic arithmetical, logical, and input/output operations of a computer system. Advanced conventional processors have inherently higher power dissipation. The arithmetic logic unit of CPU performs simple arithmetic and logical operations and the control unit of CPU manages the various components of the computer. Power dissipation is the main constrain when it comes to portability. Reversible logic gates are designed as a method to reduce the energy dissipation of logic circuits based on Launder ′ s concept. Central processing unit designed using reversible logic have a signiﬁcant contribution in low power computing. Launder proved that a reversible circuit produces less power than the irreversible circuit. Moreover, by using power optimization algorithm, the power consumption of a reversible circuit can be optimized than their existing reversible counterparts. For example, for a 64-bit comparator, the proposed circuit achieves the improvement of 55.67% in terms of power over the existing reversible one. Reversible circuit has some disadvantageous properties. It requires a large number of constant ancilla inputs. It also produces a large number of garbage outputs to maintain the reversibility which eventually increases the number of gates, area, power and delay. By using eﬃcient design techniques and suitable algorithmsthe number of garbage outputs should be minimized. For example, for a 64-bit comparator, the proposed circuit achieves the improvement of 33.1% in terms of garbage outputs over the existing reversible one. As reversible computing is a new era in logic synthesis, the existing reversible central processing units have the lack of completeness. However, researchers proposed various basic components of reversible central processing unit. In this thesis, the basic elements of the reversible circuits are newly proposed or improved in terms of numbers of gates, garbage outputs, quantum cost and delay compared to the existing counterparts. The proposed reversible basic elements are further used to construct the complex circuits, such as memory circuits, arithmetic logic unit, control unit etc., in order to design a complete reversible central processing unit. The main contribution of this thesis is to design a new reversible central processing unit which is eﬃcient in terms of numbers of gates, garbage outputs, quantum cost, ancilla inputs and delay. The reversible central processing unit is designed using novel modularization approach by presenting architecture of a logically reversible processor based on the Von Neumann architecture that can operate with very low power consumption, protection of power analysis attack and long span of life due to less heat dissipation. The organization and architecture of the proposed processor is designed from scratch. Algorithms are proposed to produce the components of the reversible processor and to calculate area and power consumption. The capabilities of the new processor is determined, the datapath layout is designed and the necessary logic is constructed to control the datapath. The computational complexity is considered to estimate the execution time of the algorithm. Existing component designs are compared with the proposed components and theorems are presented to prove the superiority of the proposed architecture. The comparative results show that the proposed circuit requires less power than the existing circuits in terms of numbers of gates, garbage outputs, quantum cost, area and power. For example, for a 64-bit comparator, the proposed design achieves the improvement of 66.6% in terms of number of gates, 33.1% in terms of garbage outputs, 27.6% in terms of quantum cost, 52.39% in terms of area and 55.67% in terms of power over the existing reversible counterpart. The proposed components are simulated and the simulation results verify the correctness of the proposed design. The proposed reversible central processing unit can make a signiﬁcant contribution in the ﬁeld of low power reversible computing and quantum computing.
Design and Implementation of Reversible Programmable Devices
(University of Dhaka, 2020-03-15) Tara, Nazma
Reversible logic is a computing design, where the ideal implementation would produce zero entropy gain. This unique feature causes the prominent use of reversible computing. More integration capability and regular structure for synthesizing a large number of logic functions made programmable devices enthusiastic to use. In this thesis, we describe efficient design procedures of two programmable devices namely Programmable Logic Array (PLA) and Field Programmable Gate Array (FPGA) with reversible logic gate. In the first part of this thesis, we design the reversible Programmable Logic Array (RPLA). Here, we propose an efficient algorithm to design the RPLA with a newly proposed 3 × 3 reversible TB (Tara-Babu) gate, which can realize multioutput ESOP (Exclusive-OR Sum of Product) functions. We present a heuristic algorithm to sort and realize the product terms of ESOP functions to share the internal sub-products to reduce the number of gates in the proposed RPLA. Proposed algorithms make the RPLA more efficient with improvement of 9.83% in terms of the number of gates, 21.3% in terms of the number of garbage outputs and 14.75% in terms of quantum cost than the counter metrics of the existing RPLA averagely. Moreover, we compute the area requirement and the power consumption of the proposed RPLA. We also analyze the performances by using MCNC benchmark functions. In the last part of this thesis, we design the most significant part of a Field Programmable Gate Array, the Plessey Logic Block with reversible gate. On the way to design the proposed reversible Plessey Logic Block, we design each component such as reversible D-Latch, reversible Decoder, reversible Multiplexer, reversible Master-Slave Flip-Flop, and reversible RAM separately. The proposed design of the individual component is primarily made efficient in terms of the number of gates, garbage outputs, quantum cost, and delay. In addition, area and power are reduced to ensure the power efficiency of the circuits. Two 4 × 4 reversible gates, namely HNF (Hafiz-Naz-Flip-Flop) gate and HND (Hafiz-Naz-Decoder) gate are proposed to achieve the optimization goal. Moreover, proposed algorithms, lemmas and theorems certify the novelty of the proposed design. Compared to previous works, the proposed counter-parts of Reversible Plessey Logic Block require less number of gates, garbage outputs, quantum cost, and delay. Finally, the proposed Reversible Plessey (4 × 2) Logic Block is compared with existing designs. The Comparative results prove the efficacy and novelty of the proposed design showing improvement of 51.62% in terms of number of Transistor, 73.57% in terms of area requirement and 34.12% in terms of power consumption with respect to the corresponding metrics of the best existing design in the literature
Duty-cycle Medium Access Control for Directional Wireless Sensor Networks
(University of Dhaka, 2018-10-27) Nur, Fernaz Narin
Directional communication in wireless sensor network minimizes interference and thereby increases reliability and throughput performances of the network. Such advantages of Directional Wireless Sensor Networks (DSNs) have attracted the interests of researchers and industry experts around the globe. Furthermore, the sensor nodes with directional antennas provide extended network lifetime and better coverage performances. However, designing a communication protocol for wireless networks with directional antennas is a challenging problem due to lack of synchronization, asymmetry-in-gain, hidden terminal and deafness problems. Our endeavor in this dissertation is to address the aforementioned challenges in neighbor discovery and medium access control in Directional Sensor Networks. One of the key challenges of a directional node is to discover its neighbors due to diﬃculty in achieving synchronization among directed transmissions and receptions. Existing solutions suﬀer from high discovery latency and poor percentage of neighbor discovery either due to lack of proper coordination or centralized management of the discovery operation. In this thesis, we develop a collaborative neighbor discovery (COND) mechanism for DSNs. Using polling mechanism, each COND node directly discovers its neighbors in a distributed way and collaborates with other discovered nodes so as to allow indirect discovery. It helps to increase the neighbor discovery performance signi cantly. A Markov chain-based analysis has been carried out to quantify theoretical performances of the proposed COND system. The performance of the COND system is evaluated in NetworkSimulator Version-3 (NS-3), and the results reveal that it greatly reduces the discovery latency and increases neighbor discovery ratio compared to state-of-the-art approaches. The second contribution of this thesis is to develop a low duty-cycle directional medium access control protocol, termed as DCD-MAC, where, each pair of (parent and child sensor) nodes performs synchronization with each other before data communication. Each parent node in the network schedules data transmissions of its childs in such a way that the number of collisions occurred during transmissions from multiple nodes is minimized. A sensor node remains active only when it needs to communicate with others; otherwise, it goes to sleep state. The DCD-MAC exploits localized information of nodes in a distributed manner and it gives weighted-fair access of transmission slots to the nodes. As a nal point, we have studied the performances of our proposed MAC protocol through extensive simulations in NS-3 and the results show that the DCD-MAC gives better reliability, throughput, end-to-end delay and network lifetime compared to the related directional MAC protocols.
Dynamic Traffic Engineering for high-Throughput Data Delivery III Wireless Mesh Networks
(© University of Dhaka, 2025-05-27) Islam, Maheen
Dynamic traffic engineering for high-throughput data delivery in wireless mesh networks
(University of Dhaka, 2018-04-08) Islam, Maheen
Wireless Mesh Network (WMN) has recently been emerged as a promising technology for wireless Internet infrastructure development because of its low cost, ease of deployment and installation facilities. The increasing number of users and diversi ed application usages as well as the incorporation of sensors and Internet of Things (IoT) devices with the WMNs have caused exponential growth in traffic ows. This increased volume of traffic causes congestion in the network and degrades application throughput, reliability and delay performances. Therefore, providing satisfactory network performance using the limited bandwidth resources, has emerged as a challenging problem. Our endeavour in this dissertation is to address high-throughput data delivery chal- lenges in WMNs. Many state-of-the-art works address ow performance improvements in WMNs in many ways, ranging from routing, scheduling, channel allocation to rate control. However, none of these approaches merely addresses the instantaneous network conditions and sudden surge of huge data traffic from diverse user applications, that cause network to become congested. To optimize network performance, a dynamic traffic engi- neering mechanism requires to consider underlying network topology, available resources and traffic demand. Furthermore, traffic forwarding should act upon network dynamics, e.g., link error, link failure, neighborhood interference, path congestion, etc. Considering the aforementioned issues, in this thesis, we rst develop an optimization framework for Dynamic Traffic Engineering, namely O-DTE, assuming that xed channels are allocated to different links. O-DTE aims to minimize neighborhood interference and backlogged traffic, and explores the least congested next-hop nodes so that the overall throughput of the network is maximized. The O-DTE belongs to mixed integer nonlinear programming (MINLP) problem and involves both combinatorial and continuous constraints, making it an NP-hard problem. A greedy heuristic alternate solution G-DTE is then developed that produces near-optimal results. Motivated by the enhanced capacity offered by dynamic channel allocation in WMNs, the second part of our thesis focus on developing a joint link-channel selection and power allocation optimization framework (OLCP), which follows hop-by-hop traffic splitting approach and exploits single-hop information to forward traffic over least-congested and minimally-interfered link-channel pairs, which in turn improves spatial reuse and thus helps to improve overall network throughput. As nding a real-time solution of OLCP is intractable in a typical mesh router, we develop a greedy heuristic solution for the problem, GLCP, to achieve a sub-optimal solution. Recently, cognitive radio (CR) enabled mesh routers have proven to mitigate spec- trum scarcity by opportunistic licensed spectrum utilization. Thus, to boost up flow throughput in Cognitive Radio Wireless Mesh Network (CR-WMNs), we present a cen- tralized optimization framework, called COTE, in the third part of this dissertation. The COTE aims at maximizing aggregated network throughput by selecting an optimal set of link-channel pairs, power allocation over those and fair traffic splitting after considering channel idle probability, link interference and path congestion. Further, a centralized Suboptimal Traffic Engineering (SOTE) solution is proposed by employing Lagrangian dual decomposition to the COTE problem, to ensure a resolution in polynomial time. Finally, a Distributed Greedy Traffic Engineering (DGTE) method is proposed to ensure fast convergence to the dynamic changing network behavior and to improve scalability. The effectiveness of our proposed dynamic traffic engineering methods are evaluated via ns-3 simulations. The simulation results demonstrate that the proposed solutions outperform the state-of-the-art works in terms of throughput, delay, reliability, fairness and convergence cost.
Maximizing Coverage Quality Using Minimum Number of Nodes in Clustered Directional Sensor Networks
(University of Dhaka, 2018-09-24) Sharmin, Selina
The invention of directional sensor nodes has given birth of a special kind of network namely Directional Sensor Network (DSN), which provides better network lifetime and sensing coverage compared to its omni-directional counterpart. These two cutting-edge features help DSNs attracting interests of research and industrial communities, particularly for the areas of high quality sensing in Smart City applications including healthcare, infrastructure security, traﬃc and access monitoring, etc. In this thesis, we focus on two fundamental reasearch topics of DSNs - area coverage and target coverage. The area coverage problem in Directional Sensor Networks (DSNs) presents great research challenges including minimization of number of active sensors and overlapping sensing coverage area among those, determination of their active sensing directions in an energy-eﬃcient way, etc. Existing solutions permit to execute coverage enhancement algorithms at each individual sensor nodes, leading to high communication and computation overheads, loss of energy and reduced sensing coverage. In this thesis, we ﬁrst formulate the problem of maximizing area coverage with minimum number of active nodes as a mixed-integer linear programming (MILP) optimization problem for a clustered DSN. Due to its NP-completeness, we then develop a greedy alternate solution, namely α-overlapping area coverage (α-OAC). In α-OAC, each cluster head (CH) takes the responsibility of determining the active member nodes and their sensing directions; where, each sensing node is allowed to have at most α% coverage overlapping with its neighbors. The α-OAC CHs activate a sensor node if and only if the later has suﬃcient residual energy and send other member nodes to the sleep state. The proposed α-OAC system is distributed and scalable since it requires single-hop neighborhood informationonly. Results from extensive simulations, done in Network Simulator version 3 (ns-3), reveal that the α-OAC system outperforms state-of-the-art works in terms of area coverage, network lifetime and operation overhead. Conventional researches on target coverage in Directional Sensor Networks (DSNs) mainly focus to increase the network lifetime, overlooking the coverage quality of targets; especially, they don’t consider the targets that have heterogeneous coverage requirements. Increasing sensing quality is of utmost importance to ensure comfort living in Smart Cities. In this dissertation, we have designed a generalized framework, namely MQMS-DSN (Maximizing coverage Quality with Minimum number of Sensors in DSN), that has the ability to maximize the target coverage quality or the network lifetime or to make an eﬃcient tradeoﬀ between the two following an application demand. Using a probabilistic model for measuring the sensing coverage quality, we have developed optimal, suboptimal and greedy solutions for MQMS problem. Empirical evaluations of the proposed MQMS systems have been carried out in ns-3. The results show the eﬀectiveness of the proposed systems compared to state-of-the-art-works in terms of sensing quality and network lifetime.
Optimal Algorithms for Stereo Correspondence Estimation
(2024-04-24) Mondal, Md. Abdul Mannan
Stereo correspondence has attained a position of overwhelming dominance in Computer Vision for long days for determining three-dimensional depth information of objects using a pair of left and right images from a stereo camera system. In this thesis we propose four novel ideas for improving the efficiency and accuracy of stereo correspondence estimation in stereo vision. First idea presents a “Real Time Approximation (RTA)” algorithm for computing the disparity of the stereo image sequences. The algorithm has been organized to make it dedicated for real timeapplications. To do this, the original image is scaled down and obtained highest speed to compute the stereo correspondences. The second idea is a searching algorithm titled “Two Dimensional Real Time Spiral Search Algorithm (2DRTSSA)” to compute the stereo correspondence two dimensionally. The 2DRTSSA thus increases the speed and accuracy over the existing state-of-the-art methods of one dimensional and left-right searching strategy. The third idea is a new and significant searching method, is explored by the name “Self-Adaptive Algorithm (SAA)” for computing stereo correspondence or disparity of stereo image. According to the SAA method, stereo matching search range can be selected dynamically until finding the best match. The searching speed is almost doubled by reducing the search range half of its original, by dividing the searching range into two regions. First one is –d to 0 and second one is 0 to +d max max . To determine the correspondence of a pixel of the reference image (left image), the window costs of the right image are computed either for –d to 0 region or for 0 to +d region depending on the result of previous matching. The speed and accuracy are further improved by introducing the fourth idea entitled “Self-Guided Stereo Correspondence (SGSC) Estimation” algorithm. The SGSC algorithm is directed by photometric properties of the candidate-pixels. Searching performance is slightly improved by utilizing this photometric property of the candidate-pixels as well as by implanting the pioneer threshold technique. These two key techniques reduced the computational costs with further improvement of accuracy. The achievements of the SGSC method are testified on Middlebury standard stereo datasets of 2001, 2003, 2006 and Middlebury latest Optical Flow Datasets. Moreover, the newly invented algorithms RTA, 2DRTSSA, SAA and SGSC have been justified on real images which are acquisitioned in our laboratory in complex environment. The overall performances of max all algorithms are satisfactory in case of real stereo images. Finally, the proposed methods are compared with present state-of-the-art methods and our 2DRTSSA, SAA and SGSC outperforms the latest methods in terms of speed, visualization of hidden ground truth, 3D reconstruction and accuracy.
Quality of Service Aware Data Delivery Protocol in Narrow Band Internet of Things Enabled Healthcare Systems
(© University of Dhaka, 2025-04-10) Sultana, Nahar
The need for establishing smart hospitals is becoming increasingly evident due to a number of reasons driven by modern healthcare and technological breakthroughs. Internet of Things (IoT), Medical Sensors, Low Power Wide Area Network Technol ogy (LPWAN), Artificial intelligence (AI), Digital technologies, and Reliable Data Transmission Techniques are used by smart hospitals to improve patient care, opti mize resource management and streamline hospital operations. In order to handle the increasing complexity of healthcare delivery, smart hospitals are essential for various purposes. They improve automation in patient demand, increase oper ational effectiveness, lower the costs, and increase accessibility to healthcare by leveraging advanced technologies. In this dissertation, a suitable licensed LPWAN technology, namely Narrow band Internet of Things (NB-IoT) is chosen as a promising technology for health care applications since it reduces end to end latency. Due to the interference, limited bandwidth, and heterogeneity of generated data packets, developing a data transmission framework that offers differentiated Quality of Services (QoS) to the critical and non-critical data packets is challenging. The existing literature studies suffer from insufficient access scheduling considering heterogeneous data packets and relationship among them in healthcare applications. The first contribution of i Abstract ii this thesis is to develop an optimal resource allocation framework for NB-IoT that maximizes a user’s utility through event prioritization, rate enhancement, and in terference mitigation. The proposed Priority Aware Utility Maximization (PAUM) system ensures weighted fair access to resources. In second contribution, the utilization of Device-to-Device (D2D) communi cation among Narrowband Internet of Things (NB-IoT) devices offers significant potential for advancing intelligent healthcare systems by extending its superior data rates, low power consumption. In D2D communication, strategies to miti gate interference and ensure coexistence with cellular networks are crucial. These strategies are aimed at enhancing user data rates by optimally allocating spectrum and managing the transmission power of D2D devices, presenting a complex engi neering challenge. Existing studies are limited either by the inadequate integration of NB-IoT D2D communication methods for healthcare, lacking intelligent, dis tributed, and autonomous decision-making for reliable data transmission, or by in sufficient healthcare event management policies during resource allocation in smart healthcare systems. In this work, we introduce an Intelligent Resource Allocation for Smart Healthcare (iRASH) system, designed to optimize D2D communication within NB-IoT environments. The iRASH innovatively integrates the Density based Spatial Clustering of Applications with Noise (DBSCAN) and Ant Colony Optimization (ACO) algorithms to effectively address the unique requirements of healthcare applications. The proposed system utilizes Belief-Desire-Intention (BDI) agents for dynamic and intelligent clustering of D2D devices, facilitating autonomous decision-making and efficient resource allocation. This approach not only enhances data transmission rates but also reduces power consumption, and is formulated as a Multi-objective Integer Linear Programming (MILP) problem. Abstract iii Given the NP-hard nature of this problem, iRASH incorporates a polynomial-time meta-heuristic-based ACO algorithm, which provides a suboptimal solution. This algorithm adheres to the principles of distributed D2D communication, promoting equitable resource distribution and substantial improvements in utility, energy effi ciency, and scalability. Finally, its performances are validated through simulations on the Network Simulator version 3 (NS-3) platform, demonstrating significant ad vancements over state-of-the-art solutions in terms of utility, delay, fair resource distribution, data rate, power efficiency,and system adaptability. As high as im provements of 65% in utility, 45% in fair sharing of resources, 25% in delay, 15% in packet delivery ratio observed by PAUM system and 35% in utility cost and 50% in energy cost are demonstrated by the iRASH system compared to the benchmark, proving their effectiveness
Quality-of-Experience and Reputation Aware Incentive Mechanism for Workers in Mobile Device Cloud
(©University of Dhaka, 2022-04-20) Saha, Sajeeb
Mobile Device Cloud (MDC) is a collaborative mobile cloud computing platform in which neighboring smart devices form an alliance of shared resources to mitigate resource-scarcity of an individual user device. It unfolds an improved computing opportunity for hand-held mobile devices to run compute-intensive applications like visual text translation, face recognition, augmented reality, and real-time health monitoring etc. exploiting code offloading mechanism. However, the sustainability of such a distributed platform depends on spontaneous participation of the in- volved mobile devices, i.e., resource-requester (buyer) and resource-provider (seller or worker). A fundamental challenge in such a resource-trading system is the se- lection of reliable worker mobile devices that enhances the computation quality of user applications. Moreover, participation of the worker mobile devices greatly depends on their compensations provided for the used resources. In this thesis, we focus on incentivizing mobile worker devices based on their task execution qualities to materialize a sustainable MDC system. Selection of worker mobile devices for task offloading imposes great research challenges including computation quality and worker reliability. Unfortunately, these two performance parameters often oppose each other. In this thesis, we rst develop an optimization framework that trades-off in between application execution speedup and reliability while maintaining device energy within a prede ned range. We also design an algorithm for developing a dependency tree among the modules of a software application so as to allow higher number of parallel executions, wherever and whenever it is possible. The emulation results of the proposed algorithm outperform the relevant state-of-the-art works in terms of application completion time, communication latency and rescheduling overhead. The second contribution of this thesis is to maximize user Quality-of-Experience (QoE) at minimum cost while providing attractive incentives to mobile worker devices. In literature works, mobile devices are assumed either to take part in execution voluntarily or aim to optimize one objective parameter (quality or cost) only. In this thesis, the aforementioned challenging problem is formulated as a multi-objective linear programming (MOLP) optimization function that exploits reverse-auction bidding policy. Practical application scenarios have been considered to trade-off between the cost and quality of execution. Due to NP-hardness of the MOLP, we offer two greedy worker selection algorithms for maximizing user QoE and minimizing execution cost. In both the algorithms, the amount of incentive awarded to a worker is determined following the QoE offered to a user. Theoretical proofs on holding desirable properties of the proposed incentive mechanisms have been presented. Simulation results depict effectiveness of our incentive algorithms compared to the state-of-the-art approaches.
Securing Graphical Authentication Using Keystroke Dynamics
(© University of Dhaka, 2026-03-03) Roy, Indrani
Account recovery is a critical aspect of web application security, often overlooked despite its importance. Traditional account recovery methods, such as sending a password reset link or a new username to the user’s registered email, are vulnerable to impostors who may have access to the user’s email and other credentials. This vulnerability makes account recovery a potential weak point in the overall security of a web application. Recent applications of behavioral biometrics, such as keystroke dynamics, for attack detection and user authentication bear similarities to biometric authentication. Adding keystroke dynamics analysis to the account recovery process significantly increases the difficulty for an impostor to successfully recover and take over a user’s account. To enhance user authentication effectiveness and raise account recovery requirements through keystroke dynamics, this study adds one additional measure of keystroke patterns to the already-existing features. Compared to other access control systems based on biometric features like face or fingerprint, keystroke analysis has attained a respectable level of accuracy. In this aim, this study uses experimental data and statistical analysis to show how the unique keystroke measure provided may be utilized in conjunction with the current authentication mechanism to greatly improve the authentication and security of sensitive applications. It may be beneficial to recognize the intruders and expel them from the system as long as this job can accommodate their typing rhythm. In this study, generative adversarial networks (GAN) are utilized to generate keyboard dynamics data with a focus on impersonating a user at the identification step in both fixed text and fixed sentence contexts. Three distinct architectures have been devised, implemented, and validated with the aid of machine learning and deep learning: vanilla-GAN based on simple neural networks NN, LSTM-GAN based on recurrent neural networks using long short-term memories (LSTM), CNN-GAN based on convolutional neural networks. The developed Conditional Generative Adversarial Networks have shown that these architectures can successfully replicate a user’s keystroke dynamics by learning about the user’s typing style and generating keyboard dynamics data using different GANs with different architectural styles. Findings show that keystroke dynamics patterns can be efficiently produced by the GAN and utilized to trick keystroke authentication systems.
Starfish Routing to Maximize Lifetime of Sensor Networks with Mobile Sink
(©University of Dhaka, 2022-08-29) Habib, Md. Ahsan
Nowadays, the development of smart cities has become a dire necessity to improve the quality of human life. Wireless sensors and necessary information and commu- nication technology (ICT) infrastructures are the fundamental building blocks of smart cities. The number of connected sensor nodes worldwide is jumping to 125 bn by 2030. We are continuously witnessing the rapid development of sensor net- works for a vast range of real-time applications. It becomes challenging to collect data on time from these versatile sensor nodes. Moreover, the energy criticality of these autonomous nodes is a crucial constraint to maximize lifetime of sensor net- works. Existing works in the literature suffer from imbalance energy consumption, reducing the network lifetime, and many of those don't explore efficient methods for on-time data collection from a network having obstacles. In this dissertation, we aim to explore real-time data routing and collection strategies both for obstacle-free and obstructed networks. At rst, we focus on de- veloping a novel data routing backbone to maximize network lifetime in an obstacle- free sensor network. The rst contribution of this thesis is introduction of a Star sh routing (SFR) backbone for sensor networks with a mobile sink that spreads the backbone nodes over the different regions of the network in such a manner that any source node can directly access at least one of the backbone nodes. The con-struction of the SFR backbone is motivated by the water vascular system of a sea Star sh. Following this, the backbone nodes are placed on a central ring-canal that helps to alleviate the hot-spot problem around the network center, and on the radial-canals that facilitate faster data delivery towards the mobile sink from any corner of the network. These canals jointly help to distribute data routing loads and uniform energy consumption throughout the network, resulting in extending network lifetime. The second contribution of this thesis is the development of a data collection strategy for a mobile sink in an obstructed network to reduce data delivery delay and to maximize network lifetime. Due to the presence of obstacles (e.g., building, forest, etc.) and heterogeneous data generation rates, the data collection strate- gies become more challenging and have not yet been well-studied. Therefore, it necessitates developing a data collection schedule to maximize network lifetime for real-time applications in an obstructed network. A high-speed data routing back- bone and data collection schedule of a mobile sink in an obstructed network have been developed as a mixed-integer linear programming problem that maximizes network lifetime while meeting delay-deadline requirements of real-time applica- tions. It nds an optimal travel plan of the mobile sink and corresponding sojourn durations at sojourn locations. The proposed Star sh data collection schedule also guarantees loop-free travel scheduling among the sojourn locations, ensuring bal- anced energy consumption throughout the network. Finally, the performances of these works have been carried out in Network Simulator-2, and signi cant improve- ments are observed for both obstacle-free and obstructed sensor networks in terms of network lifetime, end-to-end data delivery delay, throughput, etc. compared to the state-of-the-art works.
User Authentication from Mouse Movement Data Using Multi Classifier
(University of Dhaka, 2019-10-07) Karim, Masuda
User authentication is a process to verify the identity of someone who connects to system or resource. There are many technologies to authenticate a user. The biometric authentication system is becoming very popular due to its unique characteristics. This thesis presents a user authentication system from the mouse movement data. The mouse movement data are captured using our own developed user interface and two tools named Jitbit macro reader and Recording User Input (RUI). Raw data are sampled into blocks in two ways: one is based on specific number of action and another is based on specific duration. These data blocks are stored in database. From each blocks, twelve features are generated: Number of Points in the Trajectory, Delay Time, Number of Delay, Number of Action, Standard Deviation of Trajectory Length, Total Length of Trajectory, Standard Deviation of Slope, Standard Deviation of Difference Between Each of Slopes, Number of Curvatures, Curvature of Trajectory, Number of Changes in Horizontal Position and Number of Changes in Vertical Position. This system uses three classifiers: Support Vector Machine, K-Nearest Neighbor and Naïve Bayes separately to verify the proposed authentication system. The system is trained and tested using our captured dataset of 10 users and a benchmark dataset of 28 users. The experimental result shows that K-Nearest Neighbor based classifier performs better in terms of Average Receiver Operating Characteristic (ROC) Area, False Acceptance Rate (FAR) and False Rejection Rate (FRR). We have found FAR=2.78 and FRR=0 by using our collected own data and FAR=1 and FRR=1.2 by using benchmark data. This system is compared with S. Suganya, G. Muthumari, and C. Balasubramanian’s research and found that both FAR and FRR is improved.