Search

The purpose of this study was to evaluate the human capability to use selective auditory attention across the two different hearing pathways, air conduction and bone conduction. Four two channel communication systems were used as a basis for comparison: (1) both channels presented via headphones (left ear and right ear each represent a different channel), (2) two channels presented via loudspeakers placed at 135° and 225°, (3) one channel presented via headphones (both ears) and one channel to bone vibrator (placed at the condoyle), and (4) one channel presented via bone vibrator and one channel via loud speakers (placed at 135° and 225°). The SATASK sentence database was used as auditory stimuli. For the four systems chosen, each channel presented a different sentence, in a different voice, simultaneously. Listeners were assigned a target sentence and asked to record various elements of the SATASK sentence for the target presentation.The results showed that when bone conduction is used as a channel in the communication systems chosen, the speech intelligibility scores were slightly decreased, but the differences were not statistically significant. The only systems where the differences in the scores were significant were the loudspeakers and the system that combined bone conduction and air conduction.

This thesis describes a Byzantine fault tolerant coordination framework for Web services atomic transactions. In the framework, all core services, including transaction activation, registration, completion, and distributed commit, are replicated and protected by Byzantine fault tolerance mechanisms. The traditional two-phase commit protocol is extended by a Byzantine fault tolerant version that can tolerate arbitrary faults on the coordinator and the initiator sides, and some types of malicious faults on the participant side. To achieve Byzantine fault tolerance in an efficient manner, and to limit the types of malicious behaviors of the coordinator, a novel decision certificate is introduced. The decision certificate includes a signed copy of the participants’ vote records, and it is piggybacked with all decision notifications to the participants for each participant to verify the legitimacy of the decision. The Byzantine fault tolerance mechanisms, together with the extended two-phase commit protocol, have been incorporated into an open-source framework supporting the standard Web services atomic transactions specification. Performance characterizations of the framework show that the implementation is fairly efficient. Such a Byzantine fault tolerant coordination framework can be useful for many transactional Web services that require a high degree of security and dependability.

Large document repositories need to be organized and summarized to make them more accessible and understandable. Such needs exist in many applications, including web search, e-rulemaking (electronic rulemaking) and document archiving. Even though much has been done in the areas of document clustering and summarization, there are still many new challenges and issues that need to be addressed as the repositories become larger, more prevalent and dynamic. In this dissertation, we investigate more informative ways to organize and summarize large document repositories, especially e-rulemaking feedback repositories (ERFRs), so that the large repositories can be managed and digested more efficiently and effectively. Specifically, we mainly consider the following four tasks: 1) identifying important aspects of ERFR, 2)constructing cluster descriptions for document clustering, 3) clustering of ERFR with simultaneous construction of succinct cluster descriptions, and 4) selecting representative arguments for ERFR clustering. We propose to organize and summarize e-rulemaking feedbacks based on three different major aspects of the rulemaking process, in order to meet the different needs of the rule-writers or analysts; the three aspects are: opinions (O), issues (I) and stakeholders (S). We introduce an OIS-based approach to producing informative summaritive digest (SD) for given ERFRs. In addition, several novel concepts, approaches and algorithms are introduced, including the CDD measure, active feature selection (AFS), Pagoda search algorithms, etc. An SD, simply put, consists of a document clustering, along with certain succinct cluster descriptions (SCDs) and representative arguments (RAs) for each cluster in the clustering. The clustering of an SD can be constructed in either a flat or hierarchical manner. For hierarchical clustering, each level of the hierarchy can be constructed by emphasizing one of the O, I, and S aspects. Different orders of O, I and S can be used for the levels of the hierarchy. Different clusterings could be used to meet the needs of different users. Given a goodness measure, a "best" clustering can be recommended to the user. An SCD consists of a set of carefully selected terms along with some statistics, and the RAs are some typical arguments selected from each cluster. An RA should be a statement where certain major stakeholders have expressed opinions on some of the important issues. Collectively, an SD provides an informative navigation aid for the rule-writers and analysts to manage and digest large ERFRs. We conduct an experimental evaluation on our approaches by using some publicly available ERFRs. The results suggest that the SD not only helps user for "browsing" the feedbacks, but also gives the users some high-level sense about the feedbacks before they dig into each individual comment. The results also show that our approaches are efficient and scalable for managing large document repositories. Even though we devoted special attention to the application of e-rulemaking, we believe that most of the ideas are very generic and can be easily applied to other types of repositories, including digital archives.

Synthesis of realistic human faces is still one of the biggest challenges in computer graphics. Similarly, analysis of human faces in images has been a traditional topic of research in computer vision. Both research areas related to human faces have many practical applications including face modeling, face tracking, face relighting, and face recognition in diverse industry fields such as games, virtual reality, digital photography, biometrics and security. In this dissertation we explore the problem of model-based 3D face synthesis and analysis using various kind of image ensembles including multi-view image silhouettes, sequential video frames, and single photographs. We design and develop various image-model difference metrics that can be used in an optimization framework to recover the optimal model parameters of our 3D face model from the input images. In the first part of this dissertation we present a novel method for 3D face reconstruction from a set or sequence of 2D binary silhouettes. Experiments with a multi-camera rig as well as monocular video sequences demonstrate the advantages of our 3D modeling framework. In order to determine the set of optimal views required for silhouette-based shape reconstruction, we build on our modeling framework and extend it by aggressive pruning of the view-sphere with view clustering and various imaging constraints. In the second part of this dissertation we present a novel framework that acquires the 3D shape, texture, pose and illumination of a face from a single photograph. Using a custom-built face scanning system which is equipped with 16 digital cameras and 146 directional LED light sources, a large-scale dataset was collected, which consists of 3D face scans and light reflection images of a diverse group of human subjects. From this dataset, we derive a novel measurement-based illumination model that implicitly incorporates cast-shadows and specularities. We also propose a novel fitting framework that estimates the parameters of the morphable model by minimizing the distance of the input image to the dynamically changing illumination subspace. We leverage our modeling and fitting methods to solve two challenging problems in computer graphics and computer vision - face relighting and face recognition.