Patrick LAMBERT

Professor Emeritus, Polytech Annecy-Chambéry

Contact

E-mail: patrick.lambert -@- univ-smb.fr

Telephone: +33(0) 4 50 09 65 50 - fax: +33(0) 4 50 09 65 09

Office: A225

Address: LISTIC - Polytech Annecy Chambéry, BP 80439, 74944 Annecy le Vieux Cedex, France

Main duties / responsibilities

  • 07/2014 - 06/2018: Director of LISTIC
  • 09/2010 - 06/2014: Deputy Director of Polytech Annecy-Chambéry, in charge of research
  • 2012: General Cair CBMI 2012
  • 2011 - 2017: member of the scientific committee of ARC6 (regional research community)

Research activity

Keywords: Multivariate image filtering (especially color), Automatic understanding of image or video content

The research I've carried out throughout my career can be divided into two main periods.

Multi-component image filtering and segmentation (1989 - 2003)

This work corresponds to the supervision of three theses: W. Alshatti (1994), TH. Carron (1995), J. Chanussot (1998). The common denominator of these works is the multi-component nature of the images considered (multi-spectral images, with in particular the case of color, multi-temporal images, multi-modal images, etc.). Two themes have been addressed.

Multi-component image filtering :

The work envisaged was dedicated to extending order filters to the case of multi-component images where each pixel is characterized by a vector of components. The situation is made tricky by the absence of a natural way of ordering vectors. The originality of the work carried out was to envisage new methodologies for defining a vector order. Among the solutions proposed, the use of a distance measure dependent on two real-valued parameters α and β, conferring enhancement properties in certain configurations, and a technique, dubbed "bit interleaving", the latter solution enabling the definition of a total order between vectors.

Segmentation image multi-component :

Much of this work has been devoted to the special case of color images. In particular, specific segmentation methodologies have been developed for Hue-Luminance-Saturation representations. We used Saturation as a measure of Hue relevance and developed a segmentation methodology based on fuzzy logic tools. In a second stage, segmentation was considered as a problem of merging the results of marginal segmentations. This study was developed as part of a PNTS (Projet National de Télédétection Spatiale) project, and focused on the analysis of fusion mechanisms enabling efficient segmentation of fine structures.

More recently, work on color image filtering has been revived through a collaboration with a colleague at the University of Oran. Together, we extended his work on PDE filtering to the case of color images. We were thus able to propose a vector filter combining smoothing and enhancement (anisotropic diffusion filtering combined with a Shock filter) and introducing no false colors.

Automatic understanding of image content (2004 - ...)

This work has been developed since 2004 and corresponds to the supervision of 6 defended theses: B. Ionescu (2007), G. Païs (2010), A. Simac-Lejeune (2011), T. Strat (2013), N. Voiron (2015), R. Outach-Raoui (2019) and 2 theses in progress (A. Ben Hamida and M. Jacquemont).

Initially, we focused on the analysis of image sequences. The specificity of the work carried out is the search for features of a semantic nature, a delicate area where we have to deal with the "semantic gap". Animated films, in the context of the Annecy Animation Festival, were used as an application support for the methods implemented.

In B. Ionescu's thesis, we sought to obtain symbolic descriptors that globally characterized the sequence in terms of color, rhythm and movement. These features were exploited to automatically produce summaries and perform automatic genre classification. We then sought to enrich these descriptions by exploiting, alongside the images, information of a textual nature: the synopses of animated films. The difficulty lies in the difference in semantic level between information extracted from images and text (G. Païs thesis). Since 2008, participation in the TrecVid challenge (through the IRIM/ISIS group) has opened up new avenues of research. In particular, the theses of A. Simac-Lejeune and especially T. Strat focused on "concept" recognition in video sequences, with a wide variety of concepts to be found (school bus, applause...). Using machine learning approaches, the specificity of T. Strat's thesis work consisted in developing an approach based on spatial-temporal bags of words. With the aim of classifying videos, the thesis of N. Voiron (associate professor at the Annecy IUT) focused on the implementation of semi-supervised approaches based on spectral clustering.

More recently, with Alexandre Benoit, we have turned our attention to deep learning neural network approaches. In this very "effervescent" field, we have chosen to develop original architectures for multi-spectral satellite imagery. For example, as part of Amina Ben Hamida's thesis, we proposed an architecture based on 3D convolutional networks (2 spatial + 1 spectral) that outperformed the state of the art on certain image bases.

Rizlène Outach-Raoui's CIFRE thesis (in partnership with the start-up AboutGoods), on the subject of automatic reading of sales receipts, is part of this general framework of automatic understanding of image content. In particular, we have developed multitasking deep learning methodologies for the semantic segmentation of tickets.

Finally, a thesis in cooperation with LAPP (Laboratoire annécien de Physique des Particules) began in October 2017. The subject of this thesis concerns the use of deep arrays to analyze astrophysical images and characterize γ radiation.

Publications

referenced in HAL

Thesis directions

  • Wael ALSHATTI

"Vector approaches to filtering and edge detection in multi-spectral images".

Dec. 1990 to Oct. 1994 (ministerial allocation)

  • Thierry CARRON

"Color image segmentation in the Hue-Luminance-Saturation database: a digital and symbolic approach".

Oct. 1992 to Dec. 1995 (ministerial allocation)

  • Jocelyn CHANUSSOT

"Vector or marginal approaches to multi-component image processing".

Oct. 1995 to Nov. 1998 (ministerial allocation)

  • Bogdan IONESCU

"Symbolic Characterization of Image Sequences: Application to Animated Films".

Cotutelle Université de Savoie/Université POLITEHNICA de Bucarest

March 2004 to May 2007 (Allocation from the laboratory's own funds)

  • Gregory PAÏS

"Joint analysis of text and image to characterize animated films".

Oct. 2006 to Apr. 2010 (Allocation Assemblée des Pays de Savoie)

  • Alain SIMAC-LEJEUNE

"Modeling and management of concepts, in particular temporal concepts, to support the characterization of image sequences".

Oct. 2007 to June 2011 (Rhône-Alpes region grant)

  • Tiberius STRAT

"Analysis and interpretation of visual scenes using collaborative approaches".

Cotutelle Grenoble Alpes University / POLITEHNICA University Bucharest

Oct. 2010 to Dec. 2013 (ministerial allocation)

  • Nicolas VOIRON

"Structuring multimedia databases for visual exploration".

From Feb. 2012 to Dec. 2015 (associate professor of mathematics in the computer science department of the Annecy IUT)

  • Rizlène OUTACH-RAOUI

"Automatic reading of receipts from a mobile device".

May 2014 to July 2019 (CIFRE contract with start-up AboutGoods)

  • Amina BEN HAMIDA

"Analysis of satellite images using deep neural network approaches".

Began in Oct. 2015 (ministerial allocation)

Cotutelle University of Grenoble Alpes / University of Sfax

  • Mikaël JACQUEMONT

"Using deep neural networks for 3D analysis of gamma radiation".

Started Oct. 2017 (Project H2020 + Foundation Université Savoie Mont Blanc)

Teaching at Polytech Annecy-Chambery

  • IT, industrial computing
  • Signal and image processing
  • Signal mathematics

Activities

Events / research groups / contracts

  • Co-leader, with R. Mullot and R. Clouard, of the SCATI action of the GdR ISIS from 2005 to 2007.
  • Co-leader, with K. Chedhi and F. Truchetet, of the Multi-Composants action of the GdR ISIS from 1999 to 2003.
  • Active participation in the Color group of GdR ISIS from 1999 to 2006 (four times participation in a thematic school on Digital Color Imaging - Saint-Étienne 1999, Pau 2001, Dijon 2003, Lille 2005).
  • Co-leader of the regional LIMA project (Loisir et IMAges) of the ISLE research cluster - 2007/2010 - 60 people and 14 teams
  • Head of LISTIC's Information Processing team (13 permanent staff) 2007 - 2014
  • Organizer of the ENERSI summer school dedicated to Renewable Energies and Intelligent Systems (August 30 - September 3, 2010)
  • Scientific manager of the research contract with Schneider Electric (July 2005 to July 2008)
  • Scientific manager for research contract with SNR (April 2019 to September 2019)
  • Scientific manager of the research contract with AboutGoods (2014 to 2017)

International cooperation with :

  • Politehnica University of Bucharest
  • University of Sfax
  • University of Oran
  • University of Ottawa

Course

  • Professor at Polytech Annecy-Chambery since 2007
  • Lecturer at the Annecy IUT from 1984 to 2007
  • 2-year secondment in industry from 1988 to 1990
  • Doctor of Engineering thesis, 1983 INP Grenoble.
  • IEG engineer, 1978 INP Grenoble.