No description available.

The World Wide Web is a rich source of information and continues to expand in size and complexity. Mainly because the data on the web is lack of rigid and uniform data models or schemas, how to effectively and efficiently manage web data and retrieve information is becoming a challenge problem. Discovering web page communities, which capture the features of the web and web-based data to find intrinsic relationships among the data, is one of the effective ways to solve this problem. A web page community is a set of web pages that has its own logical and semantic structures. In this work, we concentrate on the web data in web page format and exploit hyperlink information to discover (construct) web page communities. Three main web page communities are studied in this work: the first one is consisted of hub and authority pages, the second one is composed of relevant web pages with respect to a given page (URL), and the last one is the community with hierarchical cluster structures. For analysing hyperlinks, we establish a mathematical framework, especially the matrix-based framework, to model hyperlinks. Within this mathematical framework, hyperlink analysis is placed on a solid mathematic base and the results are reliable. For the web page community that is consisted of hub and authority pages, we focus on eliminating noise pages from the concerned page source to obtain another good quality page source, and in turn improve the quality of web page communities. We propose an innovative noise page elimination algorithm based on the hyperlink matrix model and mathematic operations, especially the singular value decomposition (SVD) of matrix. The proposed algorithm exploits hyperlink information among the web pages, reveals page relationships at a deeper level, and numerically defines thresholds for noise page elimination. The experiment results show the effectiveness and feasibility of the algorithm. This algorithm could also be used solely for web-based data management systems to filter unnecessary web pages and reduce the management cost. In order to construct a web page community that is consisted of relevant pages with respect to a given page (URL), we propose two hyperlink based relevant page finding algorithms. The first algorithm comes from the extended co-citation analysis of web pages. It is intuitive and easy to be implemented. The second one takes advantage of linear algebra theories to reveal deeper relationships among the web pages and identify relevant pages more precisely and effectively. The corresponding page source construction for these two algorithms can prevent the results from being affected by malicious hyperlinks on the web. The experiment results show the feasibility and effectiveness of the algorithms. The research results could be used to enhance web search by caching the relevant pages for certain searched pages. For the purpose of clustering web pages to construct a community with its hierarchical cluster structures, we propose an innovative web page similarity measurement that incorporates hyperlink transitivity and page importance (weight).Based on this similarity measurement, two types of hierarchical web page clustering algorithms are proposed. The first one is the improvement of the conventional K-mean algorithms. It is effective in improving page clustering, but is sensitive to the predefined similarity thresholds for clustering. Another type is the matrix-based hierarchical algorithm. Two algorithms of this type are proposed in this work. One takes cluster-overlapping into consideration, another one does not. The matrix-based algorithms do not require predefined similarity thresholds for clustering, are independent of the order in which the pages are presented, and produce stable clustering results. The matrix-based algorithms exploit intrinsic relationships among web pages within a uniform matrix framework, avoid much influence of human interference in the clustering procedure, and are easy to be implemented for applications. The experiments show the effectiveness of the new similarity measurement and the proposed algorithms in web page clustering improvement. For applying above mathematical algorithms better in practice, we generalize the web page discovering as a special case of information retrieval and present a visualization system prototype, as well as technical details on visualization algorithm design, to support information retrieval based on linear algebra. The visualization algorithms could be smoothly applied to web applications. XML is a new standard for data representation and exchange on the Internet. In order to extend our research to cover this important web data, we propose an object representation model (ORM) for XML data. A set of transformation rules and algorithms are established to transform XML data (DTD and XML documents with DTD or without DTD) into this model. This model capsulizes elements of XML data and data manipulation methods. DTD-Tree is also defined to describe the logical structure of DTD. It also can be used as an application program interface (API) for processing DTD, such as transforming a DTD document into the ORM. With this data model, semantic meanings of the tags (elements) in XML data can be used for further research in XML data management and information retrieval, such as community construction for XML data.

Numerous theories have been developed in explanation of object perception, such as Feature Integration Theory, which posits that an object is perceived after two stages: a pre-attentive stage and a focused attention stage. It is during the focused attention stage that a representation of the perceived object is formed. Theories such as object file theory account for the maintenance of these object representations following their creation. Evidence for object file theory has been provided by studies of the object specific preview benefit. This thesis seeks to examine the effect that dividing attention has on the maintenance of object representations. Using the tenets of object file theory and the cortical field hypothesis for dual task interference, it is hypothesized that by presenting participants with two simultaneous tasks which make use of overlapping cortical areas the object representation initially formed will be lost resulting in the loss of the object specific preview benefit. Whereas presenting participants with two simultaneous tasks which are associated with spatially separate, or non-overlapping, cortical regions will not result in the loss of the object specific preview benefit.

We present a novel scheme ("Categorical Basis Functions", CBF) for object class representation in the brain and contrast it to the "Chorus of Prototypes" scheme recently proposed by Edelman. The power and flexibility of CBF is demonstrated in two examples. CBF is then applied to investigate the phenomenon of Categorical Perception, in particular the finding by Bulthoff et al. (1998) of categorization of faces by gender without corresponding Categorical Perception. Here, CBF makes predictions that can be tested in a psychophysical experiment. Finally, experiments are suggested to further test CBF.

An improved object tracking algorithm based Kalman filtering is developed in this thesis. The algorithm uses a median filter and morphological operations during tracking. The problem created by object shadows is identified and the primary focus is to incorporate shadow detection and removal to improve tracking multiple objects in complex scenes. It is shown that the Kalman filter, without the improvements, fails to remove shadows that connect different objects. The application of the median filter helps the separation of different objects and thus enables the tracking of multiple objects individually. The performances of the Kalman filter and the improved tracking algorithm were tested on a highway video sequence of moving cars and it is shown that the proposed algorithm yields better performance in the presence of shadows.

A novel approach to represent the profile of objects using Gaussian models is presented. The profile is a representation of the object and its surrounding regions. The object profile can be viewed as a comprehensive feature of that object and its surrounding regions. Different algorithms to estimate the profile are described. Geometric descriptors of the model are also proposed. The profile model is empirically shown to be effective and easily applicable to certain object recognition and segmentation tasks. Application experiments include modeling thin and thick objects as straight-lines, curves, and blobs using different primitives such as gray-level intensities, RGB, and HSV color. The datasets used for empirical validation are quite challenging. Datasets include images and videos corresponding to outdoor video, most of them with moving cameras. Some of the typical problems faced with the used datasets are: digital scaling, compression artifacts, camera jitter, weather effects, and cluttered backgrounds. We demonstrate the potential of leveraging the context of objects of interest as a part of an online detection process. Sample applications including detection of wires, sea horizon, street, and vehicles in outdoor videos are considered.

Recent research makes use of biologically inspired computation and artificial intelligence as efficient means to solve real-world problems. Humans show a significant performance in extracting and interpreting visual information. In the cases where visual data is not available, or, for example, if it fails to provide comprehensive information due to occlusions, tactile exploration assists in the interpretation and better understanding of the environment. This cooperation between human senses can serve as an inspiration to embed a higher level of intelligence in computational models. In the context of this research, in the first step, computational models of visual attention are explored to determine salient regions on the surface of objects. Two different approaches are proposed. The first approach takes advantage of a series of contributing features in guiding human visual attention, namely color, contrast, curvature, edge, entropy, intensity, orientation, and symmetry are efficiently integrated to identify salient features on the surface of 3D objects. This model of visual attention also learns to adaptively weight each feature based on ground-truth data to ensure a better compatibility with human visual exploration capabilities. The second approach uses a deep Convolutional Neural Network (CNN) for feature extraction from images collected from 3D objects and formulates saliency as a fusion map of regions where the CNN looks at, while classifying the object based on their geometrical and semantic characteristics. The main difference between the outcomes of the two algorithms is that the first approach results in saliencies spread over the surface of the objects while the second approach highlights one or two regions with concentrated saliency. Therefore, the first approach is an appropriate simulation of visual exploration of objects, while the second approach successfully simulates the eye fixation locations on objects. In the second step, the first computational model of visual attention is used to determine scattered salient points on the surface of objects based on which simplified versions of 3D object models preserving the important visual characteristics of objects are constructed. Subsequently, the thesis focuses on the topic of tactile object recognition, leveraging the proposed model of visual attention. Beyond the sensor technologies which are instrumental in ensuring data quality, biological models can also assist in guiding the placement of sensors and support various selective data sampling strategies that allow exploring an object’s surface faster. Therefore, the possibility to guide the acquisition of tactile data based on the identified visually salient features is tested and validated in this research. Different object exploration and data processing approaches were used to identify the most promising solution. Our experiments confirm the effectiveness of computational models of visual attention as a guide for data selection for both simplifying 3D representation of objects as well as enhancing tactile object recognition. In particular, the current research demonstrates that: (1) the simplified representation of objects by preserving visually salient characteristics shows a better compatibility with human visual capabilities compared to uniformly simplified models, and (2) tactile data acquired based on salient visual features are more informative about the objects’ characteristics and can be employed in tactile object manipulation and recognition scenarios. In the last section, the thesis addresses the issue of transfer of learning from vision to touch. Inspired from biological studies that attest similarities between the processing of visual and tactile stimuli in human brain, the thesis studies the possibility of transfer of learning from vision to touch using deep learning architectures and proposes a hybrid CNN that handles both visual and tactile object recognition.

No description available.

Tracking the identity of occluded objects requires binding an object’s features to its location to represent exactly which objects are located where, relying heavily on capacity-limited visual working memory. This dissertation aims to examine the capacity and stability of object working memory, and the complexity and fidelity of object working memory representations, in toddlers and young children. A series of four experiments used a novel task to examine 28- to 40-month-old toddlers’ and 5- to 6-year-old children’s visual working memory recall of specific objects in specific locations. I predicted capacity limits would vary with age, presentation/occlusion type, and complexity, and that older children would be able to monitor these limits successfully. Children observed arrays of featurally-distinct objects that were hidden from view either simultaneously (Chapter 2, Experiment 1 and Chapter 3, Experiments 1 & 2) or sequentially (Chapter 2, Experiment 2) and were asked to recall an object’s location. When objects were hidden simultaneously, toddlers showed a capacity of 3 feature-location bindings (Chapter 2, Experiment 1) and 5- to 6-year-old children showed a capacity of 4 feature-location bindings (Chapter 3, Experiment 1), and both showed capacity development, supporting the hypotheses. When objects were hidden sequentially, toddlers’ performance was impacted by whether they had the easier (set size 2) or harder (set size 3) block first, suggesting the structure of the task may have influenced how children divided attention between maintaining and encoding of representations in working memory. Additionally, in Chapter 3, the number of feature bindings that children had to maintain was varied. Children could remember more single-feature objects than multi-feature objects (limit of 4 vs. 3, respectively), suggesting that binding additional features to a representation taxes cognitive resources, as hypothesized. Finally, the study in Chapter 3 explored children’s ability to monitor the fidelity of their visual working memories by asking them to gauge their confidence by placing bets with tangible, at-risk resources. Children modulated their bets appropriately, betting more after providing correct answers and fewer after incorrect answers, as hypothesized. Together, these data help to inform our understanding of visual working memory for feature-location bound objects across early development.

<p>This thesis explored the development of infants’ ability to preserve spatio-temporal continuity of moving objects in situations where they disappeared completely (Study I & II) or partially (Study III) behind other objects (occluders). We recorded infants gaze direction with the help of two different techniques: 1) infants’ gaze shifts in Study I were measured with electro-oculogram (EOG) in combination with a motion analyzing system (Qualisys) that recorded the reflected infrared light from markers placed on the infant’s head and the moving object; 2) in Studies II and III a cornea reflection eye tracker was used (Tobii 1750) . </p><p>The results presented in this thesis demonstrate that 4-month-old infants are able to represent the temporal aspects of object motion during different periods of complete occlusion (Study I). At 6 months of age infants are able not only to predict the time when a moving object will reappear after complete occlusion but they are also capable to extrapolate pre-occlusion trajectory of the moving object and, thus, to accurately predict its reappearance (Study II). Moreover, in the situation where a linear pre-occlusion trajectory of the moving object is violated (the object turns by 90 degrees behind the occluder), infants at this age are capable of rapidly learning this new experience and base their future gaze shifts over occluder on the newly acquired knowledge. They are also able to preserve this new experience over a 24-hour period. </p><p>In the situations where occlusion is not complete and some visual information is still available (Study III), 9-month-old infants and to a lesser extent 5-month-old infants are able to reconstruct the moving pattern and to follow its direction of motion with the smooth eye movements. Moreover, 9-month-olds are capable to produce such smooth pursuit at an adult-like level.</p>