This research focuses on extracting road models from stereo video sequences taken from a moving vehicle. The proposed method combines color histogram based segmentation, active contours (snakes) and morphological processing to extract road boundary coordinates for conversion into Matlab or Multigen OpenFlight compatible polygonal representations. Color segmentation uses an initial truth frame to develop a color probability density function (PDF) of the road versus the terrain. Subsequent frames are segmented using a Maximum Apostiori Probability (MAP) criteria and the resulting templates are used to update the PDFs. Color segmentation worked well where there was minimal shadowing and occlusion by other cars. A snake algorithm was used to find the road edges which were converted to 3D coordinates using stereo disparity and vehicle position information. The resulting 3D road models were accurate to within 1 meter.

The undirected graphical model or Markov Random Field (MRF) is one of the more popular models used in computer vision and is the type of model with which this work is concerned. Models based on these methods have proven to be particularly useful in low-level vision systems and have led to state-of-the-art results for MRF-based systems. The research presented will describe a new discriminative training algorithm and its implementation. The MRF model will be trained by optimizing its parameters so that the minimum energy solution of the model is as similar as possible to the ground-truth. While previous work has relied on time-consuming iterative approximations or stochastic approximations, this work will demonstrate how implicit differentiation can be used to analytically differentiate the overall training loss with respect to the MRF parameters. This framework leads to an efficient, flexible learning algorithm that can be applied to a number of different models. The effectiveness of the proposed learning method will then be demonstrated by learning the parameters of two related models applied to the task of denoising images. The experimental results will demonstrate that the proposed learning algorithm is comparable and, at times, better than previous training methods applied to the same tasks. A new segmentation model will also be introduced and trained using the proposed learning method. The proposed segmentation model is based on an energy minimization framework that is iii novel in how it incorporates priors on the size of the segments in a way that is straightforward to implement. While other methods, such as normalized cuts, tend to produce segmentations of similar sizes, this method is able to overcome that problem and produce more realistic segmentations.

Nonresident angler participation in Alaskan sport fisheries has increased at a higher rate than resident participation during the past decade. Popular sport fisheries have become crowded and stakeholder groups are increasingly concerned about the future direction of Alaska sport fisheries management. To address stakeholder concerns in an informed manner, the Alaska Department of Fish and Game (ADF&G) commissioned a market segmentation study to collect baseline information for assessing the impacts of projects and strategies that provide benefits to the angling public. I developed a 24-page mail questionnaire that was sent to a stratified random sample of 15,000 Alaska nonresident fishing license holders. Information was collected on fishing participation, fishing experience, activity-specific attitudes, motivations for fishing, as well as species and locations fished. In addition, information on setting preferences, guide use, fish exportation, and opinions on several management proposals was also collected. The response rate was 54% (exclusive of surveys that were undeliverable). A two-stage empirical clustering approach, employing Ward's method and UPGMA hierarchical clustering followed by k-means partitioning, identified five nonresident angler clusters. A combination of seven specialization and four motivation variables were used to identify angler groups. The angler segments ranged in size from 15% to 24% of the sample and had diverse characteristics including differences in frequency of participation, fishing experience and preferences, as well as motivations for fishing. Significant differences existed among angler segments with regard to Alaska fishing characteristics, such as number of days and locations fished in Alaska, number of fish transported from Alaska, attributes important in fishery site selection, and likelihood of returning to Alaska to fish. Differences in fishing characteristics, resource dependency, and preferences with regard to fishery attributes also were found among anglers participating in selected Alaska sport fisheries, as well as anglers fishing for different species within a fishery location (e.g., Kenai River). Information provided by this study will allow ADF&G to assess the relative nonresident demand for different types of angling experiences in Alaska, estimate nonresident angler response to potential management actions, and focus planning and management activities in ways that are consistent with the interests of these different angler types. Results also demonstrate the potential for fishery-based segmentation to provide fisheries managers with a more detailed understanding of nonresident angler participation at the regional and fishery level. / Master of Science

This research is concerned with examining market segments and regional variations associated with winery visitors in the state of Virginia. The tourism literature published by the state of Virginia for wineries indicates that there are five wine regions. In this research, data were collected from interviews conducted at wineries in each of the five wine regions. The first phase of analysis sought to create market segments using a factor-cluster approach. Segments were created using cluster analysis and multiple discriminant analysis. Three distinct market segments based on benefits sought by the visitor emerged from these data. Regional variations were examined in the second part of this study. The data were classified into individual regions based on the locations of the wineries examined. Distinct differences in the regional profiles were revealed. Weak significant relationships among the segments and regions were also revealed through analysis indicating a spatial component to the segments. This research proposes the use of three regions for market research purposes, while retaining the five existing regions for promoting an organized structure to visiting wineries in the state. / Master of Science

Yes / A confocal microscope provides a sequence of images of the various corneal layers and structures at different depths from which medical clinicians can extract clinical information on the state of health of the patient’s cornea. Preprocessing the confocal corneal images to make them suitable for analysis is very challenging due the nature of these images and the amount of the noise present in them. This paper presents an efficient preprocessing approach for confocal corneal images consisting of three main steps including enhancement, binarisation and refinement. Improved visualisation, cell counts and measurements of cell properties have been achieved through this system and an interactive graphical user interface has been developed.

Yes / Solar imaging is currently an active area of research. A fast hybrid system for the automated detection of filaments in solar images is presented in this paper. The system includes three major stages. The central solar region is detected in the first stage using integral projections. Intensity filtering and image enhancement techniques are implemented in the second stage to enhance the quality of detection in the central region. Local detection windows are implemented in the third stage to detect the positions of filaments and to define various sized arrays to contain them. The extracted arrays are fed later to a neural network for verification purposes.

<p> The experiments presented here investigated the effects of manipulating the visibility of spelling patterns in English print, without concurrent oral segmentation, on the word identification skills of beginning non-readers. Visibility of the orthographic patterns was manipulated by presenting material organized into rime families (blocked) or with rime families distributed throughout training (unblocked), as well as through highlighting common rimes in the same colour of print. Experiment 1 demonstrated that while a program emphasizing the orthographic patterns in the English writing system (without concurrent phonological segmentation) led to rapid improvements in beginning non-readers' on line word identification, the benefits of such training did not persist beyond the training context. Experiments 2A and 2B investigated whether the failure to transfer word reading skills beyond the blocked training context was mitigated by training programs that required increased focus on the letter patterns (2A) and the letter-sound relations (2B). These manipulations did not influence performance; children continued to demonstrate poor transfer beyond the training context. Experiment 3 focused on determining the mechanisms underlying the poor transfer following blocked training. To evaluate performance, this final experiment used a novel measure comprising word identification as well as onset and rime identification. Training materials were blocked either by rimes or onsets. The question of interest was whether training on material that blocks by orthographic units allows children to identify the blocked units during training without actively decoding their letter-sound relations, thus decreasing the probability of forming connections between the graphemes and phonemes comprising them. Results indicated that this is the case when children were trained on material blocked by rimes, but not that blocked by onsets. </p> / Thesis / Doctor of Philosophy (PhD)

The promotion of Electric Vehicles (EV) has become a key measure of the governments to reduce greenhouse gas emissions. However, range anxiety is a big barrier for drivers to choose EVs over traditional vehicles. Installing more charging stations in appropriate locations can relieve EV drivers’ range anxiety. To help decide the location and number of public charging stations, we propose two optimization models for two different charging modes - fast and slow charging, which aim at minimizing the total cost while satisfying certain spatial coverage goals. Instead of using discrete points we employ network and polygons to represent charging demands. Importantly, we resolve the partial coverage problem (PCP) by segmenting the geometric objects into smaller ones using Geographic Information System (GIS) functions. We compare the geometric segmentation method (GS) and the complementary partial coverage method (CP) developed by Murray (2005) to solve the PCP. After applying the models to Greater Toronto and Hamilton Area (GTHA) and to Downtown Toronto, we show that that the proposed models are practical and effective in determining the locations and number of required charging stations. Moreover, comparison of the two methods shows that GS can fully eliminate PCP and provide much more accurate result than CP. / Thesis / Master of Science (MSc)

The contourlet transform is an emerging multiscale multidirection image processing technique. It effectively represents smooth curvature details typical of natural images, overcoming a major drawback of the 2-D wavelet transform. To further exploit its potential, in this research, a statistical model, the contourlet contextual hidden Markov model (C-CHMM), has been developed to characterize contourlet images. A systematic mutual information based context construction procedure has been developed to form an appropriate context for the model. With this contourlet image model, a multiscale segmentation method has also been established for the application to texture images. The segmentation method combines a model comparison approach with a multiscale fusion and a multi-neighbor combination process. It also features a neighborhood selection scheme based on a smoothed context map, for both the model estimation and the neighbor combination. The effectiveness of the image model has been verified through a series of denoising and segmentation experiments. As demonstrated with the denoising performance, this new model for contourlet images is more promising than the state of the art, the contourlet hidden Markov tree (C-HMT) model. The other model being compared with in this work is the wavelet contextual hidden Markov model (W-CHMM). Through the denoising experiments, the presented C-CHMM shows better robustness against noise than the W-CHMM. Moreover, the new model demonstrates its superiority to the wavelet model in the segmentation performance. Through the segmentation experiments, the value of the systematic context construction procedure has been proven. The C-CHMM based segmentation method has also been validated. In comparison with the state of the art methods for the same type, the presented technique shows improved accuracy in segmenting texture patterns of diversified nature. This success in segmentation has further manifested the potential of the newly developed contourlet image model.

A study of the soil characteristics, weather conditions, and effect of management skills on the yield of the agricultural crop requires site-specific details, which involves large amount of labor and resources, compared to the traditional whole field based analysis. This thesis discusses the design and implemention of yield monitor for sweetpotatoes grown in heavy clay soil. A data acquisition system is built and image segmentation algorithms are implemented. The system performed with an R-Square value of 0.80 in estimating the yield. The other main contribution of this thesis is to investigate the effectiveness of statistical methods and neural networks to correlate image-based size and shape to the grade and weight of the sweetpotatoes. An R-Square value of 0.88 and 0.63 are obtained for weight and grade estimations respectively using neural networks. This performance is better compared to statistical methods with an R-Square value of 0.84 weight analysis and 0.61 in grade estimation.