A5058767281- Music and Audio Processing
- Speech and Audio Processing
- Video Analysis and Summarization
- Image Retrieval and Classification Techniques
- Video Surveillance and Tracking Methods
- Topic Modeling
- Image Processing Techniques and Applications
- Biomedical Text Mining and Ontologies
- Gaze Tracking and Assistive Technology
- Text and Document Classification Technologies
- Advanced Image and Video Retrieval Techniques
- Neural Networks and Applications
- Speech Recognition and Synthesis
- Semantic Web and Ontologies
- Natural Language Processing Techniques
- Data Management and Algorithms
- Spectroscopy and Chemometric Analyses
- Handwritten Text Recognition Techniques
- Image Processing and 3D Reconstruction
- Human Pose and Action Recognition
- Face and Expression Recognition
- Image and Object Detection Techniques
- Constraint Satisfaction and Optimization
- Network Security and Intrusion Detection
- Social Robot Interaction and HRI
National Centre of Scientific Research "Demokritos"
2005-2019
Institute of Informatics & Telecommunications
2000-2014
Institute of Informatics of the Slovak Academy of Sciences
2007-2010
This article provides an overview of the first BIOASQ challenge, a competition on large-scale biomedical semantic indexing and question answering (QA), which took place between March September 2013. assesses ability systems to semantically index very large numbers scientific articles, return concise user-understandable answers given natural language questions by combining information from articles ontologies.The 2013 comprised two tasks, Task 1a 1b. In participants were asked automatically...
The volume of spam e-mails has grown rapidly in the last two years resulting increasing costs to users, network operators, and e-mail service providers (ESPs). E-mail users demand accurate filtering with minimum effort from their side. Since distribution non-spam is often different for a single filter trained on general corpus not optimal all users. question asked by ESPs is: How do you build robust scalable automatic personalized filters? We address this presenting PSSF, novel statistical...
This paper proposes a method for learning ontologies given corpus of text documents. The identifies concepts in documents and organizes them into subsumption hierarchy, without presupposing the existence seed ontology. uncovers latent topics terms which document is being generated. These form new done language neutral way, using probabilistic space reduction techniques over original term corpus. Given multiple sets (latent topics) discovered, proposed constructs hierarchy by performing...
Given an audio signal with unknown number of people speaking, speaker diarization aims to automatically answer the question “who spoke when.” Crucial success is distance metric between speech segments, a factor depending on choice feature space: distances should be low for segments same and high different speakers. Starting from Mel-frequency cepstrum coefficient (MFCC)-based space, algorithm proposed that finds Fisher near-optimal linear discriminant subspace, adapted particular speakers...
In this paper, we present an off-line methodology for isolated Greek handwritten character recognition based on efficient feature extraction followed by a suitable vector dimensionality reduction scheme. Extracted features are (i) horizontal and vertical zones, (ii) the projections of profiles, (Hi) distances from boundaries (iv) profiles edges. The combination these types leads to 325- dimensional vector. At next step, technique is applied, according which dimension space lowered down...
This paper describes a methodology for detecting when human has changed clothes. Changing clothes is basic activity of daily living which makes the valuable tracking functional status elderly people, in context non-contract unobtrusive monitoring system. Our approach uses Kinect and OpenNI SDK, along with workflow image analysis steps. Evaluation been conducted on set real recordings under various illumination conditions, publicly available source code proposed system at...
Abstract. This paper proposes a method for learning ontologies given corpus of text documents. The identifies conceptsin documents and organizes them into subsumption hierarchy, without presupposing the existence seed ontology. Themethod uncovers latent topics generating document text. discovered form concepts new ontology.Concept discovery is done in language neutral way, using probabilistic space reduction techniques over original term spaceof corpus. Furthermore, proposed constructs...
The paper is motivated by the need to handle robustly uncertainty of temporal intervals, e.g. as it occurs in automated event detection video streams. introduces a two-dimensional mapping Allen's relations, based on orthogonal characteristics interval namely relative position and size. hourglass-shaped also represents limit cases that correspond durationless intervals. Based this mapping, we define two sets primitive relations terms positioning size These primitives are then used derive...
We present an approach for grouping single-speaker speech segments into speaker-specific clusters. Our is based on applying the K-means clustering algorithm to a suitable discriminant subspace, where euclidean distance reflect speaker differences. A core feature of our approximating speaker-conditional statistics, that are not available, with which can be evaluated, thus making possible apply LDA finding optimal discriminative using unlabeled data. To illustrate method, we examples clusters...
Feature detection and tracking is an important problem in Computer Vision. Corners image are a good indication of features to track. Original algorithms may be expensive even on multicore architectures because they require full convolutions performed. Although these can performed real time modern GPUs CPUs, faster solutions needed for embedded systems complex algorithms, given that corner detections just step the analysis process. In this paper we evaluate performance energy efficiency...
We propose a method for user-driven recognition of events in audio streams, aiming to assist journalists towards easily annotate unedited audiovisual content. Nonlocal information provided by the user, as example that sound applause exists within video, is used adapting event classifiers so detect exact position these video. Towards this end, each class modeled using Support Vector Machine (SVM) and final automatic decision taken on mid-term basis, an alternative One Vs All architecture. A...
Unobtrusive every day health monitoring can be of important use for the elderly population. In particular, pupil size may a valuable source information, since, apart from pathological cases, it reveal emotional state, fatigue and ageing. To allow unobtrusive to gain acceptance, one should seek efficient methods using com- mon low-cost hardware. This paper describes method sizes common web camera in real time. Our works by first detecting face eyes area. Subsequently, optimal iris sclera...
Spatial relations between image regions are used in this paper for classification a rule-based fashion. In the particular case where correspond to semantically interpretable objects rules provide means justifying human-familiar manner. work presented here instances of object classes detected combining bottom-up (learnable models based on simple features) and top-down information (object consisting primitive geometric shapes such as lines). The system acts model spatial configuration objects....