I2S–Geostat: an innovation lab for digital imaging

What area do you operate in?

Xavier Datin.  With more than 40 years’ worth of experience, I2S specialises in image capture and processing. Imaging is applied to three main sectors:

• heritage digitisation (book scanners - Digibooks)
• industrial vision, with various different control systems for use in industry (online control of industrial processes, etc.)
• health and wellbeing, with cellular analysis tools (controlling the development of embryos in medically-assisted procreation, for example) and portable 2D or 3D vision tools (for modelling dentures in 3D in order to develop implants). 

Hussein Yahia. Our team, Geostat (Statistical geometry in acquisition data) deals in signal processing. Here, signal is to be understood in the broadest sense of the term, including sound, images, cardiac signals - all natural and complex signals from the real world. 
In our work, we use advanced ideas from statistical physics to tackle the concept of scale: the way in which events at a small scale are translated to a larger scale, i.e. how the world as we perceive it is organised based on these constituent parts. Take the perception of temperature, for example, a macroscopic concept that comes from the speed of molecules [MOU1]. Signals can be described using tools from physics capable of handling complexity and large quantities of data. 

What brought I2Sand Geostat together?

Xavier Datin.  There is a significant amount of progress still to be made in the field of imaging, particularly in terms of image quality or the digitisation techniques employed. We can still go much further when it comes to image fidelity and the quality of the information that we seek from an image. This is what makes research work so important for a company like I2S, for whom the ability to stand out from the crowd through technological innovation is crucial. We wanted to link up with a research laboratory with experience in basic research who might also be keen to link up with a SME such as ours and to share their expertise. After exploring the options available to us in Bordeaux and the surrounding area, it didn’t take us long to decide on Inria. They are based locally, and one of their PhD students had just completed a thesis on a subject linked to what we do. 

Hussein Yahia. Geostat have developed tools capable of processing large quantities of data. Such tools include non-convex analysis, i.e. the analysis of functions with a vast number of variables and irregular shapes. In 2015, one of our PhD students, Hicham Badri, was awarded the thesis prize from the Association Française pour la Reconnaissance et l'Interprétation des Formes(AFRIF - the French Association for Shape Recognition and Interpretation) for his work on non-convex image analysis, and I2S became aware of this work. We gave them a presentation and they expressed their interest in this work for one of their applications, DigiBook - seeking out high quality scans translates into significant computational restrictions. This resulted in a first research contract being signed in 2016.

What is the focus of your research? 

Xavier Datin.  We decided to research algorithm-based methods for improving image quality and fidelity at an information level that could be used in our imaging solutions. There are a number of different themes within the project: improving the 1st level, particularly in relation to colours, image corrections, supra-resolution (the level of detail for objects being analysed), 3D.

Hussein Yahia. Our research is focused on the use of non-convex optimisation methods in obtaining optimal image quality, filtering (eliminating noise), HDR (High Dynamic Range) imaging, supra-resolution and 3D image reconstruction. We began by looking to improve rapid images for large scanners.

Why did you choose the Inria Innovation Lab as the setting for your collaboration?

Xavier Datin.  Unlike a research contract where there are limits on time or where the focus is on one project in particular, the Inria Innovation Lab was launched as a genuine shared laboratory. At I2S, we have two research & development teams: an Innovation team - the R in R&D - for new concepts and for helping to develop our skills, and an Operations team tasked with developing new products - the D. The Inria project team work with our Innovation team in advance in order to boost our capacity for high-level research, enabling us to move from basic research to highly-applied research. We then make use of this in our products, which helps us to break down technological barriers, all while speaking the same language. 

Hussein Yahia. The Inria Innovation Lab has made a commitment to closer cooperation on shared research subjects, with no limits in terms of time, helping to build trust-based relationships. Research carried out within the laboratory touches on a range of areas: transferring algorithms to improve scanner capacity; more complex problems in the field of super resolution; and artificial intelligence, leading to transfers further down the line. 

What avenues have opened up for you through the Inria Innovation Lab? 

Xavier Datin.  Some avenues have already opened up to us in terms of converting the work carried out within the Inria Innovation Lab over to our activity in heritage digitisation. In this particular field, our intention is to continue to play a major role in the process of developing standards governing image quality, as we did before with the ISO standard 19264. The book scanner market is now dominated by us and a rival company from Germany. We are aiming for pole position in this market, specialising in large format book scanners, and we’re confident that the work undertaken with our Inria partners within the Innovation Lab will help us get there.

Hussein Yahia. This laboratory gives us the opportunity to work on real-life cases that we find interesting, including in the field of deep learning, for example. This also helps establish ties with other scientific fields - applied research on the theme of signals is interconnected with a range of other disciplines, including biology, astronomy, etc.

Launched by Inria in 2010, the goal of the Inria Innovation Lab (formerly the I-Lab) is to promote technology transfer towards SMEs. This involves bringing together an Inria project team and a partner SME within a shared laboratory. The two entities then work together to outline a joint work programme lasting two to three years, under a joint governance structure. Long term, this should help enhance the innovation capacity of these SMEs, boosting recruitment in R&D and technology transfer towards the commercial sector.