Speaker verification using I-vector features

This PhD research has provided novel solutions to three major challenges which have prevented the wide spread deployment of speaker recognition technology: (1) combating enrolment/ verification mismatch, (2) reducing the large amount of development and training data that is required and (3) reducing the duration of speech required to verify a speaker. A range of applications of speaker recognition technology from forensics in criminal investigations to secure access in banking will benefit from the research outcomes.

Multiscale features and information extraction of online strain for long-span bridges

The strain data acquired from structural health monitoring (SHM) systems play an important role in the state monitoring and damage identification of bridges. Due to the environmental complexity of civil structures, a better understanding of the actual strain data will help filling the gap between theoretical/laboratorial results and practical application. In the study, the multi-scale features of strain response are first revealed after abundant investigations on the actual data from two typical long-span bridges. Results show that, strain types at the three typical temporal scales of 10^5, 10^2 and 10^0 sec are caused by temperature change, trains and heavy trucks, and have their respective cut-off frequency in the order of 10^-2, 10^-1 and 10^0 Hz. Multi-resolution analysis and wavelet shrinkage are applied for separating and extracting these strain types. During the above process, two methods for determining thresholds are introduced. The excellent ability of wavelet transform on simultaneously time-frequency analysis leads to an effective information extraction. After extraction, the strain data will be compressed at an attractive ratio. This research may contribute to a further understanding of actual strain data of long-span bridges; also, the proposed extracting methodology is applicable on actual SHM systems.

An MRF based abnormal event detection approach using motion and appearance features

Abnormal event detection has attracted a lot of attention in the computer vision research community during recent years due to the increased focus on automated surveillance systems to improve security in public places. Due to the scarcity of training data and the definition of an abnormality being dependent on context, abnormal event detection is generally formulated as a data-driven approach where activities are modeled in an unsupervised fashion during the training phase. In this work, we use a Gaussian mixture model (GMM) to cluster the activities during the training phase, and propose a Gaussian mixture model based Markov random field (GMM-MRF) to estimate the likelihood scores of new videos in the testing phase. Further-more, we propose two new features: optical acceleration, and the histogram of optical flow gradients; to detect the presence of any abnormal objects and speed violations in the scene. We show that our proposed method outperforms other state of the art abnormal event detection algorithms on publicly available UCSD dataset.

Evaluation of features for leaf classification in challenging conditions

Fine-grained leaf classification has concentrated on the use of traditional shape and statistical features to classify ideal images. In this paper we evaluate the effectiveness of traditional hand-crafted features and propose the use of deep convolutional neural network (ConvNet) features. We introduce a range of condition variations to explore the robustness of these features, including: translation, scaling, rotation, shading and occlusion. Evaluations on the Flavia dataset demonstrate that in ideal imaging conditions, combining traditional and ConvNet features yields state-of-theart performance with an average accuracy of 97:3%�0:6% compared to traditional features which obtain an average accuracy of 91:2%�1:6%. Further experiments show that this combined classification approach consistently outperforms the best set of traditional features by an average of 5:7% for all of the evaluated condition variations.

Achieving competence: The interactional features of children's storytelling

Early years researchers interested in storytelling have largely focused on the development of children’s language and social skills within constructed story sessions. Less focus has been given to the interactional aspects of storytelling in children’s everyday conversation and how the members themselves, the storytellers and story recipients, manage storytelling. An interactional view, using ethnomethodological and conversation analytic approaches, offers the opportunity to study children’s narratives in terms of ‘members work’. Detailed examination of a video-recorded interaction among a group of children in a preparatory year playground shows how the children managed interactions within conversational storytelling. Analyses highlight the ways in which children worked at gaining a turn and made a story tellable within a round of second stories. Investigating children’s competence-in-action ‘from within’, the findings from this research show how children invoke and accomplish competence through their interactions.

Identifying weak linear features with the "coalescing shortest path image transform"

The detection of line-like features in images finds many applications in microanalysis. Actin fibers, microtubules, neurites, pilis, DNA, and other biological structures all come up as tenuous curved lines in microscopy images. A reliable tracing method that preserves the integrity and details of these structures is particularly important for quantitative analyses. We have developed a new image transform called the "Coalescing Shortest Path Image Transform" with very encouraging properties. Our scheme efficiently combines information from an extensive collection of shortest paths in the image to delineate even very weak linear features. © Copyright Microscopy Society of America 2011.

Using upper bounds on attainable discrimination to select discrete valued features

Selection of features that will permit accurate pattern classification is a difficult task. However, if a particular data set is represented by discrete valued features, it becomes possible to determine empirically the contribution that each feature makes to the discrimination between classes. This paper extends the discrimination bound method so that both the maximum and average discrimination expected on unseen test data can be estimated. These estimation techniques are the basis of a backwards elimination algorithm that can be use to rank features in order of their discriminative power. Two problems are used to demonstrate this feature selection process: classification of the Mushroom Database, and a real-world, pregnancy related medical risk prediction task - assessment of risk of perinatal death.

Life is too short to RTFM : how users relate to documentation and excess features in consumer products

This paper addresses two common problems that users of various products and interfaces encounter— over-featured interfaces and product documentation. Over-featured interfaces are seen as a problem as they can confuse and over-complicate everyday interactions. Researchers also often claim that users do not read product documentation, although they are often exhorted to ‘RTFM’(read the field manual).We conducted two sets of studies with users which looked at the issues of both manuals and excess features with common domestic and personal products. The quantitative set was a series of questionnaires administered to 170 people over 7 years. The qualitative set consisted of two 6-month longitudinal studies based on diaries and interviews with a total of 15 participants. We found that manuals are not read by the majority of people, and most do not use all the features of the products that they own and use regularly. Men are more likely to do both than women, and younger people are less likely to use manuals than middle-aged and older ones. More educated people are also less likely to read manuals. Over-featuring and being forced to consult manuals also appears to cause negative emotional experiences. Implications of these findings are discussed.

Exercise during chemotherapy for ovarian cancer (Echo) : study design features and outcomes of a Cancer Australia and Cancer Council Australia funded randomised, controlled trial

Ovarian cancer is the most common cause of gynaecological cancer death, with an overall 5-year relative survival of 43%. Impaired physical wellbeing and overall quality of life (QoL) represent major concerns for women during and following ovarian cancer treatment, predict survival and are amenable to change through interventions. Exercise, now considered an important part of overall management of a number of cancers, improves short-term outcomes (e.g., function, fatigue, QoL) during chemotherapy...

Staged homicides: An examination of common features of faked burglaries, suicides, accidents and car accidents

A staged crime scene involves deliberate alteration of evidence by the offender to simulate events that did not occur for the purpose of misleading authorities (Geberth, 2006; Turvey, 2000). This study examined 115 staged homicides from the USA to determine common elements; victim and perpetrator characteristics; and specific features of different types of staged scenes. General characteristics include: multiple victims and offenders; a previous relationship be- tween parties involved; and victims discovered in their own home, often by the offender. Staged scenes were separated by type with staged burglaries, suicides, accidents, and car accidents examined in more detail. Each type of scene displays differently with separate indicators and common features. Features of staged burglaries were: no points of entry/exit staged; non-valuables taken; scene ransacking; offender self- injury; and offenders bringing weapons to the scene. Features of staged suicides included: weapon arrangement and simulating self-injury to the victim; rearranging the body; and removing valuables. Examples of elements of staged accidents were arranging the implement/weapon and re- positioning the deceased; while staged car accidents involved: transporting the body to the vehicle and arranging both; mutilation after death; attempts to secure an alibi; and clean up at the primary crime scene. The results suggest few staging behaviors are used, despite the credibility they may have offered the façade. This is the first peer-reviewed, published study to examine the specific features of these scenes, and is the largest sample studied to date.

Employee perceptions of alcohol and drug policy effectiveness: Policy features, concerns about drug testing, and the key role of preventative measures

As negative employee attitudes towards alcohol and other drug (AOD) policies may have serious consequences for organizations, the present study examined demographic and attitudinal dimensions leading to employees’ perceptions of AOD policy effectiveness. Survey responses were obtained from 147 employees in an Australian agricultural organization. Three dimensions of attitudes towards AOD policies were examined: knowledge of policy features, attitudes towards testing, and preventative measures such as job design and organizational involvement in community health. Demographic differences were identified, with males and blue-collar employees reporting significantly more negative attitudes towards the AOD policy. Attitude dimensions were stronger predictors of perceptions of policy effectiveness than demographics, and the strongest predictor was preventative measures. This suggests that organizations should do more than design adequate and fair AOD policies, and take a more holistic approach to AOD impairment by engaging in workplace design to reduce AOD use and promote a consistent health message to employees and the community.

Mapping the regional influence of genetics on brain structure variability - A Tensor-Based Morphometry study

Genetic and environmental factors influence brain structure and function profoundly. The search for heritable anatomical features and their influencing genes would be accelerated with detailed 3D maps showing the degree to which brain morphometry is genetically determined. As part of an MRI study that will scan 1150 twins, we applied Tensor-Based Morphometry to compute morphometric differences in 23 pairs of identical twins and 23 pairs of same-sex fraternal twins (mean age: 23.8 ± 1.8 SD years). All 92 twins' 3D brain MRI scans were nonlinearly registered to a common space using a Riemannian fluid-based warping approach to compute volumetric differences across subjects. A multi-template method was used to improve volume quantification. Vector fields driving each subject's anatomy onto the common template were analyzed to create maps of local volumetric excesses and deficits relative to the standard template. Using a new structural equation modeling method, we computed the voxelwise proportion of variance in volumes attributable to additive (A) or dominant (D) genetic factors versus shared environmental (C) or unique environmental factors (E). The method was also applied to various anatomical regions of interest (ROIs). As hypothesized, the overall volumes of the brain, basal ganglia, thalamus, and each lobe were under strong genetic control; local white matter volumes were mostly controlled by common environment. After adjusting for individual differences in overall brain scale, genetic influences were still relatively high in the corpus callosum and in early-maturing brain regions such as the occipital lobes, while environmental influences were greater in frontal brain regions that have a more protracted maturational time-course.

A new combined surface and volume registration

3D registration of brain MRI data is vital for many medical imaging applications. However, purely intensitybased approaches for inter-subject matching of brain structure are generally inaccurate in cortical regions, due to the highly complex network of sulci and gyri, which vary widely across subjects. Here we combine a surfacebased cortical registration with a 3D fluid one for the first time, enabling precise matching of cortical folds, but allowing large deformations in the enclosed brain volume, which guarantee diffeomorphisms. This greatly improves the matching of anatomy in cortical areas. The cortices are segmented and registered with the software Freesurfer. The deformation field is initially extended to the full 3D brain volume using a 3D harmonic mapping that preserves the matching between cortical surfaces. Finally, these deformation fields are used to initialize a 3D Riemannian fluid registration algorithm, that improves the alignment of subcortical brain regions. We validate this method on an MRI dataset from 92 healthy adult twins. Results are compared to those based on volumetric registration without surface constraints; the resulting mean templates resolve consistent anatomical features both subcortically and at the cortex, suggesting that the approach is well-suited for cross-subject integration of functional and anatomic data.

The roles of shared vs. Distinctive conceptual features in lexical access

Contemporary models of spoken word production assume conceptual feature sharing determines the speed with which objects are named in categorically-related contexts. However, statistical models of concept representation have also identified a role for feature distinctiveness, i.e., features that identify a single concept and serve to distinguish it quickly from other similar concepts. In three experiments we investigated whether distinctive features might explain reports of counter-intuitive semantic facilitation effects in the picture word interference (PWI) paradigm. In Experiment 1, categorically-related distractors matched in terms of semantic similarity ratings (e.g., zebra and pony) and manipulated with respect to feature distinctiveness (e.g., a zebra has stripes unlike other equine species) elicited interference effects of comparable magnitude. Experiments 2 and 3 investigated the role of feature distinctiveness with respect to reports of facilitated naming with part-whole distractor-target relations (e.g., a hump is a distinguishing part of a CAMEL, whereas knee is not, vs. an unrelated part such as plug). Related part distractors did not influence target picture naming latencies significantly when the part denoted by the related distractor was not visible in the target picture (whether distinctive or not; Experiment 2). When the part denoted by the related distractor was visible in the target picture, non-distinctive part distractors slowed target naming significantly at SOA of -150 ms (Experiment 3). Thus, our results show that semantic interference does occur for part-whole distractor-target relations in PWI, but only when distractors denote features shared with the target and other category exemplars. We discuss the implications of these results for some recently developed, novel accounts of lexical access in spoken word production.