The main research lines within the Textual and Visual Pattern Analysis (TVPA) area at XRCE are categorization and retrieval of text and images; multimodal and hybrid pattern analysis (text, images, cross-lingual); image clustering and visualization; machine learning; document object detection and image aesthetics. Much of our research along these lines has been delivered into innovative solutions with proven scientific and commercial performance. Examples of this are Xerox GVC (image categorization engine), CategoriX (text categorization engine) and Xerox AIE (Automatic Image Enhancement).

The assessment of the aesthetic value of a photographic image still remains an open issue in the state of the art. Indeed, features catching such properties are in most part of the cases subjective and they can hardly be translated into quantitative metrics. Most of the methods now developed are in fact based on low and medium level features (chromatic content, spectrum of the image, exposure, blur, rule of thirds) and they neglect more complex content-dependant features (e.g. technical perspective, shading, etc). In addition, the estimation of the quality is rough and it is reduced to a categorization of images in few classes (e.g. point-and-shoot vs. professional photo). Typically, Bayesian classifiers and Support Vector Machine are trained on databases that contain a large number of pictures for which preference was sampled (www.photo.net). In this scenario, the internship we propose aims at developing features that can successfully sample the aesthetic quality of a picture. In particular the candidate will have the possibility to:

• Creating a database containing images and associated user preference obtained by sampling a population of users.
• Design and develop state-of-the-art features for no reference quality evaluation of photographic images
• Integration of features in a classifier and existing properties' of XRCE
• Develop new features for improving performance of the classifier

The successful candidate will combine competences in C/C++, java, matlab and a strong theoric background in digital image processing. Knowledge and practice of pattern recognition will be a plus.

XRCE provides an informal and relaxed working environment situated in the Parc de Maupertuis in Meylan. The successful students will be given the freedom and flexibility to find their own solutions and to work in a way that suits them but will have the guidance and support of experienced full-time Xerox researchers and thereby gain an introduction to the field of commercial research in a world-class research laboratory.