Luca is a research scientist in the Computer Vision group. At Xerox, he applies machine learning and pattern recognition techniques to model preference in photographic images. He also studies data driven techniques for selecting, combining and using colours in creative design. He led a three year ANR project called OMNIA that terminated in 2011. In his spare time, he enjoys creating prototype applications for image search and manipulation.
You can read some of his recent publications , patent applications and reviews of his work on gizmodo.com or MIT technology review .
Short Bio
Luca joined XRCE in 2006 after getting his Ph.D. in Computer Science from the University of Genova where he studied methods for analyzing multicamera video sequences in surveillance applications. From the same institution he also received a Master in Telecommunications Engineering in 2001. In 2000 he was a visiting scholar at Kingston University London, studying agent based architecture for computer vision applications. Luca also spent one year at the University of Glasgow with an Erasmus scholarship back in 1997.
Luca is an amateur runner and he also enjoys snowboarding, travelling and music .
Features for photographic preference assessment
When dealing with media objects like photographic images, quality can be defined in terms of descriptors capturing the aesthetic properties of an image. Such descriptors can be defined at low level (e.g. distortions such as blur, noise, exposure etc.) or at a higher level of abstraction ( layout, colour harmonies, object position etc.).
In analogy with the most recent state of the art methods for quality assessment, we tackle quality assessment as a binary classification problem: is this image ``good'' or ``bad''? This is an intrinsically challenging problem because visual data is very rich and ambiguous and when judging photographs, people are often confronted with personal tastes. However, novel databases annotated with user judgements and comments were recently published. This is the reason why we are currently interested in developing new robust and scalable features able to correlate well with human preferences.
Image Search and Visualization
All the time, we end up looking for the right image for our web page, album or document and we care about many different aspects such as semantics, aesthetics and ultimately quality. At Xerox, we create web applications that leverage the techniques for image categorization and similarity retrieval developed by my colleague Florent and I mash them up with other methods for quality and aesthetic analysis. The result is a unique swiss-army toolkit for image search and visualization composed by several widgets and tools that can be used by graphic designers and by non skilled users to create multimedia content. Check out the Omnia project for more information on this research line.
Colour Computer Vision
In this project we study methods for accessing and using colours in graphic design applications. In particular, we are interested in data-driven approaches leveraging large-scale datasets of colours selected and annotated in unconstrained environments. Our aim is to develop, supervised and non-supervised models to organize, search and combine colours. We are also interested in practical uses of colours and colour schemes for multimedia document personalization, transfer and retrieval.
