Thomas Sikora of TU Berlin receives Google Faculty Research Award – recognition of joint work with IDLab-MEDIA (UGent-imec)

Prof. Thomas Sikora of Technical University Berlin received one of the prestigious 2016 Faculty Research Awards in the area of Machine Perception. The award was given to TU Berlin to assist future work on Steered Mixtures-of-Experts (SMoE) for Video Coding.

The award is also a recognition of the fruitful collaboration between the Communication Systems Lab of TU Berlin (Germany) and IDLab-MEDIA of Ghent University – imec (Belgium) in this field and their pioneering work on SMoE. Ruben Verhack of IDLab is currently pursuing a double PhD degree on SMoE between UGent and TU Berlin under the supervision of Prof. Peter Lambert (IDLab) and Prof. Thomas Sikora. The awarded grant will be used to continue the successful collaboration between the two groups towards the development of novel enhanced SMoE texture coding algorithms.

Google Faculty Research Awards [1] are one-year awards structured as unrestricted gifts to support the research of world-class permanent faculty members at top universities around the world pursuing cutting-edge research in areas of mutual interest to Google. Out of around one thousand of nominees, 37 awards were granted with a Google Faculty Awards 2016. The full list of 2016 Google Faculty Awards recipients can found under [2].

SMoE aims to be the disruptive technology in the field of image data coding. The last 30 years were primarily dominated by hybrid DCT/DPCM technologies in image and video coding. However, these technologies are reaching their limits given that the characteristics of modern image data are changing (think virtual reality, immersion, light fields, …). As such, we believed that the whole coding paradigm needed to be rethought from scratch. This lead to the development of a new data-driven coding approach that unifies a range of image modalities. Instead of storing the views directly captured by the cameras, the underlying illuminance model that gave rise to these views is stored. This model can then be used to generate arbitrary views, and to deliver extra image information (such as segmentation, edges, and others) directly from the coded file.

The award clearly illustrates that large industry players show interest in this fundamental research. Both IDLab and Communication Systems Group are continuing their joint work on SMoE. The grant will be used for the further development of the methodology.