Research Review: Hierarchical Division Based Method for large non-rigid point sets registration
1430 until 1600
Meeting Room 7th Floor
Assoc. Prof. Dr. Bahari Belaton
Recently, the existing point sets registration algorithms have greatly influenced the field including iterative closest point (ICP), robust point matching based on thin-plate splines (RPM-TPS), coherent point drift (CPD) and Gaussian mixture models based on thin-plate splines (GMM-TPS). However, the registration speed slows down dramatically as the number of points increases, especially in the case of nonrigid point sets.
In the thesis, a hierarchical division-based multi-stage registration framework is designed for large non-rigid point sets registration, where division scheme accounts for the decomposition of the computational cost and hierarchical registrations are responsible for the registration accuracy. Followed by it, three specific division-based point sets registration methods are explored. These are the significant contributions to the thesis.
Considering the transformation and the correspondences in a registration method, a subset-ICP is proposed for rigid point sets registration which shrinks the registration space in iteration by employing subset instead of whole set. Subsets are obtained using clustering method and the aligned subset pair is registered finely by ICP. Furthermore, to improve registration performance, a streaming division associated with the refined CPD method denoting S-CPD-B is presented for large non-rigid point sets registration. A heuristic search method tunes the width parameter of Gaussian kernel in CPD to refine the registration accuracy. In addition, only consider the transformation in a registration method, a two stages GMM-TPS based on clustering scheme denoting bi-GMM-TPS is proposed to handle large non-rigid point sets registration problem. While at the same time, it can address the issues of topological ambiguity in the data. Coarse alignment is performed on the cluster center sets and fine registration is carried out on the tuned cluster pairs.
Extensive experiments are conducted to validate the efficiency of three proposed methods on the publicly available point sets including very large point sets from USF database. The proposed division-based algorithms can reduce the computational complexity whilst maintain comparable registration accuracy especially for large non-rigid point sets.
Proposal Review: Membrane-Inspired Bat Algorithm Approach to Recognize Faces in the Wild
1430 until 1600
Viva Room 7th Floor
Bisan A. N. Salibi
Dr. Ibrahim Venkat
Due to the non-intrusive nature, human analogy, social acceptability and universal applicability, Face Recognition (FR) biometric has been gaining significant attention from the biometrics community in the last few decades. The main motivation of the growing interest of FR stems from various commercial applications such as security and law enforcement pertaining to border control, face logon towards securing mobile devices and human-computer interaction. Notwithstanding the impressive developments of FR algorithms under controlled laboratory conditions, their performances may degrade considerably when moving to the wild where face images exhibit the same amount of variability that one would encounter in everyday life. The essential problem encountered when modeling FR systems under unconstrained scenarios is the complex nature and the high-dimensionality of the data. Furthermore, the intraclass variations such as pose, illumination and expression can result in significant appearance change. An increased number of bio-inspired face recognition systems have emerged in the recent decades owing to their intelligent problem solving ability, flexibility, scalability and adaptive nature. Thus, the main motivation of this research is to propose a novel bio-inspired FR algorithm based on bat algorithm to recognize faces under typical uncontrolled settings. In the first part, an up-to-date overview of various state-of-the-art bio-inspired face recognition models have been systematically synthesized, presented and clearly demonstrated where they stand after being progressively evolved and improved over the past few years. To overcome the shortcomings of typical bio-inspired algorithms such as premature convergence and to further enhance the feature selection mechanism of the proposed approach, a Membrane Computing (MC) framework will be further investigated. Inspired by the biological structure and functioning of living cells, MC has the rigor and sound theoretical development as well as providing a parallel distributed framework and flexible evolution rules. On the other hand, bio-inspired algorithms have outstanding features, such as simplicity, robustness, convenient use for real-world problems and large scope of applications. Motivated by these findings, we will explore the possible interaction between membrane computing and evolutionary computing to propose a membrane-inspired bat approach to recognize faces in the wild and validate it with de facto standard experiments and datasets.
Proposal Review: Towards Effective Retrieval of Software Design Specification
1530 until 1630
Meeting Room 7th Floor
Dr. Wan Mohd Nazmee Wan Zainon
Dr. Hamza Onornoiza Salami
Software reuse is the use of existing software artifacts to build new system rather than building from scratch. It has the potential in improving the software quality, the effective use of specialist, reduced the process risk and reduces the overall cost of software development. Despites, its clear benefits it suffers from few problems though. The software reuse problems includes lack of tool supports, increase in maintenance cost, not invented here syndrome, difficulty in maintaining reusable library, cost of retrieving and adapting a reusable software artifacts into a new projects.
A step towards the solution to these problems is the reuse of early-stage software products artifacts. Because in early-stage reuse once a matched is found all related later-stage artifacts can also be reuse. Early-stage software artifacts are domain models, requirement specification and software design while later-stage includes source code test case and software documentation. Typically these artifacts are stored in a components library or repository. As this repository increases in size, there is corresponding rise in the retrieval time which can lessen or even outweigh the expected savings in development time. Because one of the anticipated gains in software reuse is reduced the development time.
This research intends to propose a fast way of identifying subset of repository artifacts that are potentially similar to the user query prior to retrieval stage. The shortlisted repository artifacts are then compared with the user query in a subsequent computationally demanding retrieval stage to find their actual degree of similarity with the user query. This technique will leads to significant reduction in retrieval time, especially when the retrieval stage is time consuming.
Proposal Review: Impact of Technological, Organizational and Environmental Factors on Small and Medium Enterprises to Intention to Adopt e-Commerce in Jordan with the Moderating Role of Organizational Competency and External ICT Support