Sites Inria

Version française

Image Processing

LH - 27/10/2014

Inria will be participating in the international conference dedicated to image processing

Inria has been involved in organising the 2014 edition of ICIP, the international conference dedicated to image-processing, which is being held in France for the first time, between 27 and 30 October at CNIT, Paris – La Défense. Christine Guillemot, Sirocco team manager and Josiane Zerubia, Ayin team manager are both managers of the organising committee. Guillemot will be chairing the plenary sessions while Zerubia will be chairing the tutorials. Nicholas Ayache, manager of the Asclépios team, who was recently awarded the Inria – French Académie des sciences Grand Prize 2014 will be giving a talk entitled “Medical Image Computing for Computational Medicine” on 28 October. Among the numerous Inria teams conducting research into image-processing, eleven teams are involved in the work chosen for ICIP 2014, namely the Ayin, Asclépios, Géostat, Lagadic, Magrit, Prima, Stars, Serpico, Sirrocco, TexMex and Titane teams. Several teams will also be showing their work on the Inria Stand (N°9). These will be Ayin, Clime, Geostat, Morpheme and Stars.

On the occasion of this important meeting of nearly 1500 attendees, Patrick Gros, Deputy Scientific Director responsible for “Perception, Cognition, Interaction” will be there to answer your questions.

What does research into image-processing represent in terms of Inria’s scientific activity? 

Our institute is very active in this field with more than 20 of our teams involved. Image-processing covers a number of specialisms At Inria, we divide the work into several sub-domains, including medical imaging, satellite and aerial imaging, image processing for videosurveillance and image processing for multimedia and television. Some of these fields bring together increasingly larger communities of researchers who convene at these dedicated conferences. This is the case, for example, with MICCAI, a conference dedicated to medical imaging, the 2012 event having been held in Nice thanks to our deep involvement in the Inria centre at Sophia Antipolis – Méditerranée, and, in particular, the work of Nicholas Ayache, head of the Asclépios team.

Our researchers are considered to be among the finest in the world and have been very successful in winning prestigious grants awarded by the European Research Council. One of our teams, Willow, whose members include three prize-winners, Jean Ponce (the 2010 winner in the “experienced researchers” category, Josef Sivic, (2013 winner in the “young researcher” category) and Ivan Laptev (2012 winner in the “young researcher” category).
Nicholas Ayache, who is one of the three lecturers invited to appear at ICIP 2014 won an ERC grant obtained in 2011. So have the heads of other teams, including Cordélia Schmid of the Lear team.
Several of our women researchers have received another mark of distinction through being awarded a Fellowship by the IEEE (Institute of Electrical and Electronics Engineers), among them Christine Guillemot, Cordelia Schmid and Josiane Zerubia.

In short we have some real talent and genuine expertise in this field.

What are the main issues or challenges with which our researchers are confronted?

This obviously depends on the sub-domains.

In medical imaging , one of the major challenges is to meet the needs of medical practitioners so as to be able to offer them image-processing techniques that are sufficiently interactive and to assist them in surgical operations, for instance. This raises issues of real time recalibration of images of tissue, organs with fluctuating contours, and those with many wet, light-reflecting areas that make them difficult to process. These are difficult conditions in which to work because everything is moving and everything is soft. This means doing one’s utmost to perfect image extraction and filtering solutions and those for analysing the morphology.
Another factor derives from the multimodal aspect of the sources of images. To make diagnoses more accurate and thus help them in their treatment decisions or in deciding which operations to perform, healthcare professional use echography, radiography and MRI scans. So the sources of information are multiple and techniques need to be invented that enable the fusion of all the images produced in order to be able to capitalise on information thus generated.

In satellite imaging , researchers tackle images at higher and higher resolutions, representing a greater quantity of data and originating from a variety of optical, radar, lidar etc. sensors. Data comparison techniques therefore need to be developed, and mathematical models are required that are capable of processing all these types of information, which are very often multi-resolutional and using multiple spectrums, at different times and using a multiplicity of sensors.

In the field of videosurveillance , all these challenges are linked to assistance issues, and keeping people in their homes. Excluding the difficulty of being able to experiment, the major challenge is above all the design models that make it possible to provide the correct information, the right analysis of the situation. For example, when a system analyses someone who is doing the dishes how is it possible to identify the interpersonal variability in the way in which the dishes are done and what may possibly be deterioration in a person’s physical abilities? These involve computer viewing issues that our Inria teams are currently researching.

Finally, in the field of image-processing for television or multimedia , one of the major challenges is that of image compression. The vast majority of videos are watched at low definition and represent nearly 75% of Internet traffic. With high resolution TV standards and the development of 3D, new technical solutions need to be found to match these changes and ensure that videos can still be viewed on the internet without taking up too much space, causing the flow of images to deteriorate or compromise the continuity of traffic.

New imaging techniques are being developed such as HDR (high dynamic range) imaging or plenoptic cameras that capture the amount and direction of the available light in each pixel. These cameras make it possible to show the depth in elements of the image based on a single image and thus enable 3D rendering on the basis of just one image, or to vary the depth of field. All these are new problems to tackle, new fields of research.

The work you have mentioned will have a potential impact in very practical applications. To what extent has the image-processing research conducted by Inria’s scientists been applied to industry, has it resulted in technology or products that we see all around us? 

One of Inria’s strength is its understanding of how to work with the world of industry. Our image-processing teams, like many of our other teams, work with leading industrials such as Technicolor, Orange, Saint-Gobain, Philips, Thomson Video network, Thalès, Alcatel-Lucent, Microsoft, Xerox and more as well as with such SMEs as Technosens, Aldebaran Robotics, Citilog, Envivio, LTU, and others.

They produce software, patented concepts and even help create start-ups.

An example of a recent start-up is Therapixel, the result of work performed by the Asclépios. Team. This start-up offers surgeons a device that they can use to consult data and images without having to touch either a screen or a keyboard. This was a solution that was long-awaited by surgeons who must touch as few items as possible in the operating theatre.
Another example is Iconem, the result of work directed by Jean Ponce of the Paris École Normale Supérieure and Inria in Microsoft Research-Inria Joint Centre. The aim of this research project is to apply the latest image-processing techniques to automatic 3D reconstruction and merely on the basis of photographs, recreated the décor that has disappeared from one of the great patrician villas of Pompei, “Diomedes Villa”. The Iconem start-up aims to enable this technology to be used to preserve the world’s greatest archaeological and historic sites.

Image-processing is a very promising field whether on the scientific level or that of transfers to industry.

Keywords: Modeling Multimedia ERC grant Personally Assisted Living Data Visualization Medical imaging Image processing Video surveillance ICIP