A5054724020- Video Analysis and Summarization
- Speech Recognition and Synthesis
- Speech and Audio Processing
- Advanced Image and Video Retrieval Techniques
- Music and Audio Processing
- Authorship Attribution and Profiling
- Face recognition and analysis
- Handwritten Text Recognition Techniques
- Image Retrieval and Classification Techniques
- Natural Language Processing Techniques
- Video Surveillance and Tracking Methods
- Advanced Data Compression Techniques
- Biometric Identification and Security
- Names, Identity, and Discrimination Research
- Topic Modeling
- Algorithms and Data Compression
- Diverse Cultural and Historical Studies
- 3D Surveying and Cultural Heritage
- Speech and dialogue systems
- Histone Deacetylase Inhibitors Research
- Artificial Intelligence in Healthcare
- Ubiquitin and proteasome pathways
- Human Pose and Action Recognition
- Multimodal Machine Learning Applications
- Text and Document Classification Technologies
Université Grenoble Alpes
2011-2021
Centre National de la Recherche Scientifique
2011-2021
Institut pour l'avancée des biosciences
2021
Inserm
2021
Laboratoire d'Informatique de Grenoble
2011-2017
Université Paris-Sud
2015-2017
Laboratoire d'Informatique pour la Mécanique et les Sciences de l'Ingénieur
2014-2017
Université Joseph Fourier
2011-2012
Institut polytechnique de Grenoble
2011-2012
We present in this article a video OCR system that detects and recognizes overlaid texts as well its application to person identification documents. proceed several steps. First, text detection temporal tracking are performed. After adaptation of images standard system, final post-processing combines multiple transcriptions the same box. The semi-supervised particular type (video broadcast from French TV) is proposed evaluated. efficient it runs 3 times faster than real time (including step)...
Identifying speakers in TV broadcast an unsupervised way (i.e., without biometric models) is a solution for avoiding costly annotations. Existing methods usually use pronounced names, as source of identifying speech clusters provided by diarization step but this too imprecise having sufficient confidence. To overcome issue, another names can be used: the written title block image track. We first compared these two sources on their abilities to provide name broadcast. This study shows that it...
This article presents a demo of person search in audiovisual broadcast using only the text available video and resources external to video. We also present different steps used recognize characters for multi-modal recognition systems. Text detection is realized features (texture, color, contrast, geometry, temporal information). The itself performed by Google Tesseract free software. method was successfully evaluated on news corpus that contains 59 videos from France 2 French TV channel.
Growing evidence is showing that acetylation plays an essential role in cancer, but studies on the impact of KDAC inhibition (KDACi) metabolic profile are still their infancy. Here, we analyzed, by using iTRAQ-based quantitative proteomics approach, changes proteome KRAS-mutated non-small cell lung cancer (NSCLC) A549 cells response to trichostatin-A (TSA) and nicotinamide (NAM) under normoxia hypoxia. Part this was further validated molecular biochemical analyses correlated with...
In this paper an approach to human annotation propagation for person identification in the multimodal context is proposed. A system used, which combines speaker diarization and face clustering produce clusters. The whole clusters are later annotated rather than just single tracks, done by propagation. Optical character recognition systems provides initial annotation. Four different strategies, select candidates annotation, tested. results of promising. With use a proper active learning...
L'identification de personnes dans les émissions télévision est un outil précieux pour l'indexation ce type vidéos mais l'utilisation modèles biométriques n'est pas une option viable sans connaissance a priori des présentes vidéos.Les noms prononcés ou écrits peuvent nous fournir liste hypothèses.Nous proposons comparaison du potentiel ces deux modalités (noms écrits) afin d'extraire le nom parlant et/ou apparaissant.Les proposent plus grand nombre d'occurrences citation erreurs...
Classification quality criteria such as precision, recall, and F-measure are generally the basis for evaluating contributions in automatic speaker recognition. Specifically, comparisons carried out mostly via mean values estimated on a set of media. Whilst this approach is relevant to assess improvement w.r.t. state-of-the-art, or ranking participants context an annotation challenge, it gives little insight system designers terms cues improving algorithms, hypothesis formulation, evidence...