PhD studentship – Automatic graphical pattern spotting in historical documents starting September 2013
The aim of this thesis is to develop robust pattern spotting techniques for historical document images (medieval manuscripts or other types of old documents containing graphical parts, such as ornamental background or decorative letters). Pattern spotting consists in searching in a document image for occurrences of a graphical “object”, i.e. a pattern more or less complex such as a logo, a signature, a medieval letter, a symbol, a coast of arms, etc. The query is formulated by pointing in the image an example of the pattern to search for (image query). The interest of pattern spotting is to ease information indexation and retrieval in complex historical digitized documents such as medieval manuscripts for example. One efficient indexing method consists in describing the image using a bag of visual words, i.e. using a vector aggregating local descriptors according to a predefined vocabulary (codebook). Such a representation of the images is efficient for retrieving very la
rge image databases but spatial organization of the characteristics are lost. For pattern spotting in document images, this spatial organization is crucial, especially the spatial organization of colors in medieval illuminated manuscripts.
In this thesis, we wish to deeper explore the adaptation of the technique to the detection of patterns in document images such as medieval manuscripts. Our goal is to exploit color descriptors and the search for a sparse representation of visual word lexicon as well as integrating some mechanisms that enable to describe the spatial organization of the colors. The flexibility of the pattern spotting approach should eventually allow to generalize these works to the spotting of more complex objects like scenes in medieval manuscripts for instance.
About the LITIS Lab: The LITIS (Computer Science, information processing and systems) laboratory is the research unit in Communication and Information Sciences and Technologies of the Upper Normandy Region. Our lab gathers researchers from the three main Higher Education institutions of the region: Rouen University, Le Havre University and the National Institute of Applied Sciences (INSA) of Rouen. The laboratory has 160 members, half of which are PhD students. The LITIS research topics cover a wide spectrum of Communication and Information Sciences and Technologies, from fundamental researches to applications, in particular to life sciences and humanities.
Candidate Profile: The PhD candidate should hold a Master of Science, in the field of computer science or computer engineering, with a major in signal and image processing. He/she should also have sound knowledge in pattern recognition (feature extraction, learning, and classification). Experience with document image analysis is an advantage.
If you are interested in applying for the position, please send a resume, a letter explaining why you are interested, transcripts of the candidate’s Master degrees, and the contacts of two references to: Laurent.Heutte@univ-rouen.fr, Stephane.Nicolas@univ-rouen.fr, Caroline.Petitjean@univ-rouen.fr Deadline: June 15th, 2013
Location : LITIS EA 4108, Université de Rouen, Technopole du Madrillet, 76821 Saint-Etienne-du-Rouvray, FRANCE
Advisoring : L. Heutte (Professeur), S. Nicolas (Maître de Conférences), C. Petitjean (Maître de Conférences)
Funding: The Upper Normandy Region offers a 3-year studentship of 1374,69 € per month (net income).
Posted by: Dr. NICOLAS Stéphane (stephane.nicolas@univ-rouen.fr).
