• Speaker : Professor Ronan Fablet, Institut Telecom Bretagne

• Title : Characterizing the spatial distribution of objects and local signatures in images: application to ocean sensing data

• Date: Thursday 15 April
• Time: 15:00-15:45
• Place: Meeting Room num 2, 13th floor, CASIA

• Abstract
  

In recent years important advances have been gained in the definition of robust local image signatures for computer vision applications, such as image indexing, scene categorization, object and texture recognition,…. Relevant classification and recognition performances have been reported including invariances to contrast changes and geometric image transform.

Though it may me convey relevant information, the spatial distribution of the local signatures is most often ignored and only the local visual information is exploited. Here we investigate how to combine the characterization of the visual information of local image signatures and the associated spatial patterns. Viewing a set of local image signatures as the realization of a marked point process, second-order spatial summary statistics are proposed. Two case-studies are reported to exemplify this approach: texture recognition from the spatial patterns formed by SIFT keypoints and image categorization for fisheries acoustics data based on from the spatial patterns formed by segmented foreground objects.

• Biography
  

Graduated from SUPAERO in 1997 and holding a postgraduate in applied mathematics from the University Paul Sabatier (Toulouse), Ronan Fablet was in 1998 at the Signal Processing Lab of the French Naval Academy (Brest) as a research assistant. He received in PhD from INRIA Rennes in 2001. His Phd was concerned with complex dynamic scene analysis and content-based video indexing. He then held a postdoc position for one year in the Dpt of Computer Science of Brown University (USA). In 2002 he was appointed at Ifremer (the French Research Institute for Marine Science) as a full time researcher in the fisheries science department and developed multidisciplinary projects on signal processing and computer vision applied to marine ecology. Since 2008 he is with the Signal and Communication department at Institut Telecom/Telecom Bretagne as an associated professor and develop research activities in the field of computer vision with a particular emphasis on ocean sensing applications. He was awarded in 2002 the AFRIF price from ? Association Fran?aise de Reconnaissance et d’Interprétation des Formes ? and in 2004 the ? Herve Troadec ? award at the International Otolith Research Symposium. He is the principal investigator of several national and European research projets. His main research interests include probabilistic models for signal and image processing, dynamic scene analysis, texture analysis, statistical and variational inference, geometric image structures, similarity-based indexing and retrieval, applications to ocean sensing data.