University of Northern British Columbia Institutional Repository

Search results

Search Results

Displaying (1 - 17 of 17)

Title: Inverse scale invariant feature transform models for object recognition and image tagging.
Contributors: Md. Kamrul Hasan (author), Liang Chen (Thesis advisor), University of Northern British Columbia (Degree granting institution)
Abstract: This thesis presents three novel image models based on Scale Invariant Feature Transform (SIFT) features and the k-Nearest Neighbors (k-NN) machine learning methodology. While SIFT features characterize an image with distinctive keypoints, the k-NN filters away and normalizes the keypoints with a two-fold goal: (i) compressing the image size, and (ii) reducing the bias that is induced by the variance of keypoint numbers among object classes. Object recognition is approached as a supervised machine learning problem, and the models have been formulated using Support Vector Machines (SVMs). These object recognition models have been tested for single and multiple object detection, and for asymmetrical rotational recognition. Finally, a hierarchical probabilistic framework with basic object classification methodology is formulated as a multi-class learning framework. This framework has been tested for automatic image annotation generation. Object recognition models were evaluated using recognition rate (rank 1) whereas the annotation task was evaluated using the well-known Information Retrieval measures: precision, recall, average precision and average recall.
Discipline: Computer Science
Content Model: info:fedora/ir:thesisCModel
Date added: 2010

Title: Experimental study of multi-level regional voting scheme and its application in human face recognition.
Contributors: Jian Zhang (author), Liang Chen (Thesis advisor), University of Northern British Columbia (Degree granting institution)
Abstract: Based on previous regional voting scheme model studies and assumptions, we extend studies from two-level voting scheme models to multi-level voting scheme models. Due to mathematical complexity, we use the Monte Carlo approach in studying the stability of the multi-level regional voting scheme with respect to region sizes and levels. Using this model, we are able to obtain the stability characteristics of the national voting scheme, the two-level regional voting scheme, and the multi-level regional voting scheme, and apply it in FERET human face database. According to our study, we verify again that the regional voting scheme (including the two-level regional voting scheme and multi-level regional voting scheme) is always more stable than the national voting scheme. We find that the stability of the multi-level regional voting scheme is not as good as the stability of the two-level regional voting scheme when region size is within a certain range. Out of this range, the multi-level regional voting scheme may compete with the two-level voting scheme. We conclude that the multi-level regional voting scheme may be comparable to the two-level regional voting scheme and prove our conclusion in the face recognition application.
Discipline: Computer Science
Content Model: info:fedora/ir:thesisCModel
Date added: 2009

Title: A security architecture for IPv6 enabled wireless medical sensor networks.
Contributors: M. Abdul Alim (author), Liang Chen (Thesis advisor), University of Northern British Columbia (Degree granting institution)
Abstract: We present the design of an IPv6 enabled wireless sensor network based on the IEEE 802.15.4 standard for medical monitoring. We design a routing mechanism for efficient flooding, a hop-by-hop error recovery and congestion control mechanism for reliable packet delivery and a lightweight security architecture for the medical monitoring system. We extend the widely used Extensible Authentication Protocol (EAP) to employ the Generalized Pre-shared Key (GPSK) authentication method with some optimizations for securing the system. We use the 3-party EAP model with the Personal Area Network Coordinator (PAN coordinator) of IEEE 802.15.4 standard as the EAP authenticator for authenticating sensor nodes within the radio range of the PAN coordinator. In order to use EAP authentication for a sensor node several hops away from the PAN coordinator, we define a new role (relay authenticator) for its coordinator which tunnels EAP messages to the PAN coordinator securely. We define EAP message encapsulation for IEEE 802.15.4 networks and a key hierarchy for the security architecture. We have simulated the system and shown that EAP based authentication is feasible in wireless sensor networks.
Discipline: Computer Science
Content Model: info:fedora/ir:thesisCModel
Date added: 2008

Title: Improving Sonar Sensor Fidelity in a Robot Simulator.
Contributors: Allan Edward Kranz (author), Charles Brown (Thesis advisor), Liang Chen (Thesis advisor), University of Northern British Columbia (Degree granting institution)
Abstract: It is slow and expensive to develop robot control systems using real robots. Simulation can provide the benefits of lowering the time and cost. In order to take advantage of the benefits of development in a simulator we need high fidelity representations of actual sensors. Sensors do not provide perfect data and simulations that use either perfect models or models that are too simple will not translate well into the real world. This research introduces a sensor model that overcomes some of the existing limitations in current simulations and provides a methodology for developing both new models and corresponding testing regimes. An actual sensor is used in realistic situations to create authentic models that more closely match the performance of the robot in the real world. A simple sonar sensor is tested against three generic obstacles and a realistic software simulation model of its capabilities is created. The Simbad robot simulator is modified to use this model, a testing regime is created to validate the results, and improved performance over the existing model is achieved. --Leaf iii.
Discipline: Computer Science
Content Model: info:fedora/ir:thesisCModel
Date added: 2015

Title: Face recognition using Dual Linear Regression based Classification
Contributors: Yuan Wang (author), Liang Chen (Thesis advisor), University of Northern British Columbia (Degree granting institution)
Abstract: This project discusses a face recognition algorithm that uses Dual Linear Regression based Classification (DLRC) and a voting approach. Each face image used is first converted into a cluster of images each image in the cluster is obtained by shifting the original image a few pixels. The similarity of a pair of face images can be measured by comparing the distance between the corresponding image clusters, which is calculated using DLRC approach. To further improve performance, each cluster of images, representing a single face image, is then partitioned into a union of clusters of sub images. DLRC is then used to measure similarities between corresponding sub-image clusters to provide temporary identity decisions a voting approach is applied to make final conclusions. We have carried out experiments on a benchmark dataset for face recognition. The result demonstrates that the proposed approach works best in certain simple situations, while its performance is also comparable to known algorithms in complicated situations. --Leaf ii.
Discipline: Computer Science
Content Model: info:fedora/ir:thesisCModel
Date added: 2016

Title: Local binary pattern network: a deep learning approach for face recognition
Contributors: Meng Xi (author), Liang Chen (Thesis advisor), University of Northern British Columbia (Degree granting institution)
Abstract: Deep learning is well known as a method to extract hierarchical representations of data. This method has been widely implemented in many fields, including image classification, speech recognition, natural language processing, etc. Over the past decade, deep learning has made a great progress in solving face recognition problems due to its effectiveness. In this thesis a novel deep learning multilayer hierarchy based methodology, named Local Binary Pattern Network (LBPNet), is proposed. Unlike the shallow LBP method, LBPNet performs multi-scale analysis and gains high-level representations from low-level overlapped features in a systematic manner. The LBPNet deep learning network is generated by retaining the topology of Convolutional Neural Network (CNN) and replacing its trainable kernel with the off-the-shelf computer vision descriptor, the LBP descriptor. This enables LBPNet to achieve a high recognition accuracy without requiring costly model learning approach on massive data. LBPNet progressively extracts features from input images from test and training data through multiple processing layers, pairwisely measures the similarity of extracted features in regional level, and then performs the classification based on the aggregated similarity values. Through extensive numerical experiments using the popular benchmarks (i.e., FERET, LFW and YTF), LBPNet has shown the promising results. Its results out-perform (on FERET) or are comparable (on LFW and FERET) to other methods in the same categories, which are single descriptor based unsupervised learning methods on FERET and LFW, and single descriptor based supervise learning methods with image-restricted no outside data settings on LFW and YTF, respectively. --Leaves i-ii.
Discipline: Computer Science
Content Model: info:fedora/ir:thesisCModel
Date added: 2015

Title: Packing equal circles in a damaged square using simulated annealing and greedy vacancy search.
Contributors: Xinyi Zhuang (author), Liang Chen (Thesis advisor), Desanka Polajnar (Thesis advisor), University of Northern British Columbia (Degree granting institution)
Abstract: This thesis defines and investigates a generalized circle packing problem, called Packing Equal Circles into a Damaged Square (PECDS). We introduce a new heuristic algorithm that enhances and combines the Greedy Vacancy Search (GVS) and Stimulated Annealing (SA), and demonstrate, through a series of experiments, its ability to find better solutions than either GVS or SA alone. The synergy between the enhanced GVS and SA, along with explicit convergence detection, makes the algorithm robust in escaping the points of local optimum. --Leaf ii.
Discipline: Computer Science
Content Model: info:fedora/ir:thesisCModel
Date added: 2015

Title: Heuristic Path Finding Method for Online Game Environment.
Contributors: Jia-jia Tang (author), Liang Chen (Thesis advisor), University of Northern British Columbia (Degree granting institution)
Abstract: Pathfinding is one of the main problems for computer gaming. It has long been a bottleneck for system performance in the online game industry. Due to the vast amount of pathfinding requests and attributes of game maps, many pathfinding methods that work well for the console game environment have failed the challenges of online games. In order to obtain a satisfactory performance, the background processing system has to sacrifice either efficiency or accuracy otherwise it would require a hardware improvement. Therefore, after investigating possible solutions to resolve these common issues of pathfinding, we have designed a Heuristic Path Finding Method. Under this method, designers analyze the game map structure and build area information first. The online game system will then generate path templates for in-game usage based on the map information. As the templates are being generated, the system's pathfinding Artificial Intelligence (AI) will pick a path from the templates and adjust it accordingly to produce a real path. This method improves pathfinding tasks with higher accuracy, is less time consuming and requires fewer resources from the game system. We have also created a testing system as a tool for testing and evaluating pathfinding related work. We carried out a series of experiments with the testing system on the online game service, and showed us that our method is a better solution than a few known algorithms.--P. i.
Discipline: Computer Science
Content Model: info:fedora/ir:thesisCModel
Date added: 2011

Title: Study of document retrieval using Latent Semantic Indexing (LSI) on a very large data set.
Contributors: A. N. K. Zaman (author), Liang Chen (Thesis advisor), Charles Brown (Thesis advisor), University of Northern British Columbia (Degree granting institution)
Abstract: The primary purpose of an information retrieval system is to retrieve all the relevant documents, which are relevant to the user query. The Latent Semantic Indexing (LSI) based ad hoc document retrieval task investigates the performance of retrieval systems that search a static set of documents using new questions/queries. Performance of LSI has been tested for several smaller datasets (e.g., MED, CISI abstracts etc) however, LSI has not been tested for a large dataset. In this research, we concentrated on the performance of LSI on large dataset. Stop word list and term weighting schemes are two key parameters in the area of information retrieval. We investigated the performance of LSI by using three different set of stop word lists and, also, without removing the stop words from the test collection. We also applied three different term-weighting (raw term frequency, log-entropy, and tf-idf) schemes to measure retrieval performance of LSI. We observed that, firstly, for a LSI based ad hoc information retrieval system, a tailored stop word list must be assembled for every unique large dataset. Secondly, the use of tf-idf term weighting scheme shows better retrieval performance than log-entropy and raw term frequency weighting schemes even when the test collection became large. --P. ii.
Discipline: Computer Science
Content Model: info:fedora/ir:thesisCModel
Date added: 2010

Title: A weighted regional voting based ensemble of multiple classifiers for face recognition.
Contributors: Jing Cheng (author), Liang Chen (Thesis advisor), University of Northern British Columbia (Degree granting institution)
Abstract: Face recognition is heavily studied for its wide range of application in areas such as information security, law enforcement, surveillance of the environment, entertainment, smart cards, etc. Competing techniques have been proposed in computer vision conferences and journals, no algorithm has emerged as superior in all cases over the last decade. In this work, we developed a framework which can embed all available algorithms and achieve better results in all cases over the algorithms that we have embedded, without great sacrifice in time complexity. We build on the success of a recently raised concept - Regional Voting. The new system adds weights to different regions of the human face. Different methods of cooperation among algorithms are also proposed. Extensive experiments, carried out on benchmark face databases, show the proposed system's joint contribution from multiple algorithms is faster and more accurate than Regional Voting in every case. --P. ix.
Discipline: Computer Science
Content Model: info:fedora/ir:thesisCModel
Date added: 2012

Title: Information retrieval by multi-perspective representation.
Contributors: Jia Zeng (author), Liang Chen (Thesis advisor), University of Northern British Columbia (Degree granting institution)
Abstract: No abstract available.
Discipline: Computer Science
Content Model: info:fedora/ir:thesisCModel
Date added: 2006

Title: Architecture for automatic poetry generation through pattern recognition
Contributors: Kimberley Scofield (author), Tina Fraser (Thesis advisor), Liang Chen (Thesis advisor), Charles Brown (Thesis advisor), University of Northern British Columbia College of Science and Management (Degree granting institution), David Casperson (Committee member), Robert Budde (Committee member), Han Li (Committee member)
Abstract: Document representation and topic modelling are important problems for artificial intelligence researchers, with applications ranging from education technology to bioinformatics. Many approaches have been proposed, the majority falling broadly into categories of Statistical Analysis and Natural Language Processing (NLP). This thesis proposes an architecture that optimizes a combination of statistical and linguistic analysis in an unsupervised machine learning environment. The proposed architecture is a design for agile, stable, document modelling. By clustering within the statistical inference algorithm, it reduces the computational cost of time and space associated with conventional classifying algorithms such as K-means, increasing the threshold for size and frequency of aggregate data analysis. This translates to an increased stability for evolution of learning. The architecture builds on the concept of socio-linguistic connections as an inherent combination of statistics and linguistics, and employs well-researched concepts of statistical and linguistic analysis, including embedded sub-manifold analysis. ...
Discipline: Interdisciplinary Studies
Content Model: info:fedora/ir:thesisCModel
Date added: 2017

Title: A copula based method for fish species classification
Contributors: Raj Singh Dhawal (author), Liang Chen (Thesis advisor), University of Northern British Columbia (Degree granting institution)
Abstract: The purpose of this thesis is to develop a method for classification of the species of a fish given in an image. This method uses the state of the art multi-dimensional image descriptor HOG (Histogram of Oriented Gradients), and colour histograms to create representative feature vectors. In this work copula theory have been used to summarize the multi-dimensional features. Copula theory has been used extensively for analysing the bivariate data however, not much exploration has been done to find its application for analysing multivariate data. This work is one of the few attempts where copulas have been used to analyse multivariate data. The classification accuracy of this method is comparable with other reported methods. --Leaf ii.
Discipline: Computer Science
Content Model: info:fedora/ir:thesisCModel
Date added: 2015

Title: Face Recognition Through Regret Minimization.
Contributors: Daniel Yule (author), Liang Chen (Thesis advisor), University of Northern British Columbia (Degree granting institution)
Abstract: Face Recognition is an important problem for Artificial Intelligence Researchers, with applications to law enforcement, medicine and entertainment. Many different approaches to the problem have been suggested most approaches can be categorized as being either Holistic or Local. Recently, local approaches have shown some gains. This thesis presents a system for embedding a holistic algorithm into a local framework. The system proposed builds on the concept of Regional Voting, to create Weighted Regional Voting which divides the face images to be classified into regions, performs classification on each region, and finds the final classification through a weighted majority vote on the regions. Three different weighting schemes taken from the field of Regret Minimization are suggested, and their results compared. Weighted Regional Voting is shown to improve upon unweighted Regional Voting in every case, and to outperform or equal many modern face recognition algorithms. --P. ii.
Discipline: Computer Science
Content Model: info:fedora/ir:thesisCModel
Date added: 2011

Title: Scale-space and wavelet decomposition based scheme for face recognition using nearest linear combination.
Contributors: Farhana Hoque (author), Liang Chen (Thesis advisor), University of Northern British Columbia (Degree granting institution)
Abstract: Face recognition has attracted much attention from Artificial Intelligence researchers due to its wide acceptability in many applications. Many techniques have been suggested to develop a practical face recognition system that has the ability to handle different challenges. Illumination variation is one of the major issues that significantly affects the performances of face recognition systems. Among many illumination robust approaches, scale-space decomposition based methods play an important role in reducing the lighting effects in facial images. This research presents a face recognition approach for utilizing both the scale-space decomposition and wavelet decomposition methods. In most cases, the existing scale-space decomposition methods perform recognition, based on only the illumination-invariant small-scale features. The proposed approach uses both large-scale and small-scale features through scale-space decomposition and wavelet decomposition. Together with the Nearest Linear Combination (NLC) approach, the proposed system is validated on different databases. The experimental results have shown that the system outperforms many recognition methods in the same category. --Leaf ii.
Discipline: Computer Science
Content Model: info:fedora/ir:thesisCModel
Date added: 2014

Title: Noise reduction for face identification in videos
Contributors: Negar Hassanpour (author), Liang Chen (Thesis advisor), University of Northern British Columbia (Degree granting institution)
Abstract: The wealth of information extracted from a sequence of frames in a video provides samples of the subject in different illuminations, head poses, and facial expressions. However, various sources can impose noise on data (e.g., occlusion, low resolution, and face detection failures). In this thesis, a novel framework is proposed that employs the well-studied concepts in quantum probability theory to design a representation structure capable of making inferences with multiple sources of uncertainty. The dual extension of this framework is aimed at reducing the effect of noisy frames in a video. It is also used to guide the sampling process in a novel learning scheme, called specialization generalization, which is designed to support efficient learning, as well as neutralizing the effect of noisy samples in the identification process. The contributions of this thesis are not method-specific and can be utilized for enhancement of other face identification approaches in the literature. --Leaf i.
Discipline: Computer Science
Content Model: info:fedora/ir:thesisCModel
Date added: 2015

Title: Initial investigation into using a two-level regional voting approach for face verification.
Contributors: Jun Ma (author), Liang Chen (Thesis advisor), University of Northern British Columbia (Degree granting institution)
Abstract: Face verification is defined as a person whose identity is claimed a priori will be compared with the person's individual template in database, and then the system checks whether the similarity between pattern and template is sufficient to provide access. In this thesis we introduce a new procedure of face verification with an embedding Electoral College framework, which has been applied successfully in face identification. The approaches are evaluated by experiments on benchmark face databases applying the Electoral College framework embedded with standard baseline PCA algorithm and newly developed algorithm S-LDA. The results demonstrate that the proposed face verification systems improve the performance of these holistic algorithms. --P. i.
Discipline: Computer Science
Content Model: info:fedora/ir:thesisCModel
Date added: 2010

Usage statistics