Vojkan Mihajlovic

Multi-modal extraction of highlights from TV Formula 1 programs

As amounts of publicly available video data grow, the need to automatically infer semantics from raw video data becomes significant. In this paper, we focus on the use of dynamic Bayesian networks (DBN) for that purpose, and demonstrate how they can be effectively applied for fusing the evidence obtained from different media information sources. The approach is validated in the particular domain of Formula I race videos. For that specific domain we introduce a robust audiovisual feature extraction scheme and a text recognition and detection method. Based on numerous experiments performed with DBN, we give some recommendations with respect to the modeling of temporal and atemporal dependences within the network. Finally, we present the experimental results for the detection of excited speech and the extraction of highlights, as well as the advantageous query capabilities of our system.

Soft, Comfortable Polymer Dry Electrodes for High Quality ECG and EEG Recording

Conventional gel electrodes are widely used for biopotential measurements, despite important drawbacks such as skin irritation, long set-up time and uncomfortable removal. Recently introduced dry electrodes with rigid metal pins overcome most of these problems; however, their rigidity causes discomfort and pain. This paper presents dry electrodes offering high user comfort, since they are fabricated from EPDM rubber containing various additives for optimum conductivity, flexibility and ease of fabrication. The electrode impedance is measured on phantoms and human skin. After optimization of the polymer composition, the skin-electrode impedance is only ∼10 times larger than that of gel electrodes. Therefore, these electrodes are directly capable of recording strong biopotential signals such as ECG while for low-amplitude signals such as EEG, the electrodes need to be coupled with an active circuit. EEG recordings using active polymer electrodes connected to a clinical EEG system show very promising results: alpha waves can be clearly observed when subjects close their eyes, and correlation and coherence analyses reveal high similarity between dry and gel electrode signals. Moreover, all subjects reported that our polymer electrodes did not cause discomfort. Hence, the polymer-based dry electrodes are promising alternatives to either rigid dry electrodes or conventional gel electrodes.

A novel self-guided approach to alpha activity training

Fifty healthy participants took part in a double-blind placebo-controlled study in which they were either given auditory alpha activity (8-12Hz) training (N=18), random beta training (N=12), or no training at all (N=20). A novel wireless electrode system was used for training without instructions, involving water-based electrodes mounted in an audio headset. Training was applied approximately at central electrodes. Post-training measurement using a conventional full-cap EEG system revealed a 10% increase in alpha activity at posterior sites compared to pre-training levels, when using the conventional index of alpha activity and a non-linear regression fit intended to model individual alpha frequency. This statistically significant increase was present only in the group that received the alpha training, and remained evident at a 3 month follow-up session, especially under eyes open conditions where an additional 10% increase was found. In an exit interview, approximately twice as many participants in the alpha training group (53%) mentioned that the training was relaxing, compared to those in either the beta (20%) or no training (21%) control groups. Behavioural measures of stress and relaxation were indicative of effects of alpha activity training but failed to reach statistical significance. These results are discussed in terms of a lack of statistical power. Overall, results suggest that self-guided alpha activity training using this novel system is feasible and represents a step forward in the ease of instrumental conditioning of brain rhythms.

Spatial filters to detect steady-state visual evoked potentials elicited by high frequency stimulation: BCI application.

Abstract Brain-computer interfaces (BCIs) based on steady-state visual evoked potentials (SSVEPs) require minimal user training and can offer higher information throughput compared to other BCI modalities. We focused on SSVEPs elicited by high-frequency stimuli (>30 Hz) because they cause minimal fatigue/annoyance and reduce the risk of inducing photoepileptic seizures. This paper presents an approach that analyzes electroencephalographic activity to automatically obtain the optimum spatial filter for detecting the SSVEP at a given stimulation frequency from a short signal where the stimulation is presented at intermittent periods interspersed with breaks. A vector space generated by sinusoidal signals at the stimulation frequency and harmonics is defined. The spatial filter coefficients result from maximizing the ratio between the energy of the spatially filtered signal and that of its orthogonal component with regard to the vector space. The spatial filters are customized for each BCI user through a short calibration procedure taking into account individual specificity. Our experiments on six subjects applying the spatial filters resulted in an average transfer rate ranging from 20.9 to 22.7 bits/min.

TIJAH: Embracing IR Methods in XML Databases

This paper discusses our participation in INEX (the Initiative for the Evaluation of XML Retrieval) using the TIJAH XML-IR system. TIJAH’s system design follows a ‘standard’ layered database architecture, carefully separating the conceptual, logical and physical levels. At the conceptual level, we classify the INEX XPath-based query expressions into three different query patterns. For each pattern, we present its mapping into a query execution strategy. The logical layer exploits score region algebra (SRA) as the basis for query processing. We discuss the region operators used to select and manipulate XML document components. The logical algebra expressions are mapped into efficient relational algebra expressions over a physical representation of the XML document collection using the ‘pre-post numbering scheme’. The paper concludes with an analysis of experiments performed with the INEX test collection.

Score region algebra: building a transparent XML-R database

A unified database framework that will enable better comprehension of ranked XML retrieval is still a challenge in the XML database field. We propose a logical algebra, named score region algebra, that enables transparent specification of information retrieval (IR) models for XML databases. The transparency is achieved by a possibility to instantiate various retrieval models, using abstract score functions within algebra operators, while logical query plan and operator definitions remain unchanged. Our algebra operators model three important aspects of XML retrieval: element relevance score computation, element score propagation, and element score combination. To illustrate the usefulness of our algebra we instantiate four different, well known IR scoring models, and combine them with different score propagation and combination functions. We implemented the algebra operators in a prototype system on top of a low-level database kernel. The evaluation of the system is performed on a collection of IEEE articles in XML format provided by INEX. We argue that state of the art XML IR models can be transparently implemented using our score region algebra framework on top of any low-level physical database engine or existing RDBMS, allowing a more systematic investigation of retrieval model behavior.

TIJAH at INEX 2004 modeling phrases and relevance feedback

This paper discusses our participation in INEX using the TIJAH XML-IR system. We have enriched the TIJAH system, which follows a standard layered database architecture, with several new features. An extensible conceptual level processing unit has been added to the system. The algebra on the logical level and the implementation on the physical level have been extended to support phrase search and structural relevance feedback. The conceptual processing unit is capable of rewriting NEXI content-only and content-and-structure queries into the internal form, based on the retrieval model parameter specification, that is either predefined or based on relevance feedback. Relevance feedback parameters are produced based on the data fusion of result element score values and sizes, and relevance assessments. The introduction of new operators supporting phrase search in score region algebra on the logical level is discussed in the paper, as well as their implementation on the physical level using the pre-post numbering scheme. The framework for structural relevance feedback is also explained in the paper. We conclude with a preliminary analysis of the system performance based on INEX 2004 runs.

TIJAH scratches INEX 2005: vague element selection, image search, overlap, and relevance feedback

Retrieving information from heterogeneous data sources in a flexible manner and within a single (database) framework is still a challenge. In this paper we present several extensions of our prototype database system TIJAH developed for structured retrieval. The extensions are aimed at modeling vague selection of XML elements and image retrieval. All three levels (conceptual, logical, and physical) of the TIJAH system are enhanced to support the extensions. Additionally, we analyze different ways of removing overlap and explain how structural information can be used for relevance feedback.

Phase synchrony analysis for SSVEP-based BCIs

Brain-computer interfaces (BCI) based on Steady State Visual Evoked Potential (SSVEP) can provide higher throughput than other BCI modalities. For the sake of safety and comfort, the frequencies of the stimulus should be higher than 30 Hz. However, only a limited number of these frequencies can elicit SSVEPs that are strong enough for BCI purposes. In order to increase the number of available stimuli, the SSVEP phase can be taken into account. In this study, we used phase synchrony analysis to extract the phase difference between SSVEP and stimuli as a feature to identify a subjects intention. This analysis can mitigate the adverse effect brought by the phase deviation that may occur in the stimuli. Furthermore, the classification accuracy when using a single lead signal (Oz-Cz) is compared to a spatial filtered signal. The result shows that the phase synchrony analysis can effectively extract the phase difference and that spatial filtering can significantly increase the classification accuracy.

Dry and Water-Based EEG Electrodes in SSVEP-Based BCI Applications

This paper evaluates whether water-based and dry contact electrode solutions can replace the gel ones in measuring electrical brain activity by the electroencephalogram (EEG). The quality of the signals measured by three setups (dry, water, and gel), each using 8 electrodes, is estimated for the case of a brain-computer interface (BCI) based on steady state visual evoked potential (SSVEP). Repetitive visual stimuli in the low (12 to 21Hz) and high (28 to 40Hz) frequency ranges were applied. Six people, that had different hair length and type, participated in the experiment. For people with shorter hair style the performance of water-based and dry electrodes comes close to the gel ones in the optimal setting. On average, the classification accuracy of 0.63 for dry and 0.88 for water-based electrodes is achieved, compared to the 0.96 obtained for gel electrodes. The theoretical maximum of the average information transfer rate across participants was 23bpm for dry, 38bpm for water-based and 67bpm for gel electrodes. Furthermore, the convenience level of all three setups was seen as comparable. These results demonstrate that, having optimized headset and electrode design, dry and water-based electrodes can replace gel ones in BCI applications where lower communication speed is acceptable.