SS2: Biologically-inspired multimedia processing and quality assessment

Motivation and Objectives

Biologically-inspired mechanisms are increasingly being applied into the design of multimedia processing technologies. When it comes to visual data processing for example, the knowledge of human visual systems (HVS) properties have been widely deployed to perceptually optimize imaging systems and to assess visual quality in a more HVS-aligned manner. Convolutional neural networks for vision tasks, such as image classification, have also shown resemblances with biological vision in terms of their hierarchical organization and increasing abstract ‘representations’ during the processing steps; however, it remains unclear whether such networks do actually closely mimic biological vision. Biologically-inspired spiking neural networks, on the other hand, are now able to provide flexible, low-latency, and energy efficient solutions thanks to advances in neuromorphic processors. While bio-inspired models are becoming increasingly popular in multimedia processing, they do not fully imitate biological systems, thereby presenting a gap between computational models and brain data analysis. This calls for new research results that bridge biological mechanisms and computational models for multimedia processing by integrating neural biology, computer science and mathematics. Research is also required towards developing more transparent and interpretable models, especially for neural networks, by examining the instantiation of a certain theoretical model of a biological system in a network, and determining which parameters, and architectural properties, of the network correspond to relevant (biologic) model properties.

Topics of interest

This special session targets research related to biologically-inspired systems to promote the deployment of biological mechanisms in computational models for multimedia processing. Topics of interest include, but are not limited to:

• Biologically-inspired models for multimedia quality assessment

• Biologically-inspired networks for multimedia compression and restoration

• Neuromorphic and perception-based analysis

• Advanced methods for spiking neural networks, including out-of-distribution detection, interpretability and explainability

• Visual neural coding and biologically-inspired deep neural networks

• Biologically-inspired virtual reality and human-computer interaction

• Biologically-inspired statistical learning model for multimedia processing

• Research on deep neural networks and biological mechanism correspondence

• Neurophysiological experiments to study neuronal activity in visual cortex.

• Computational models incorporating visual mechanism characteristics (such as selective attention, feature selectivity, noise resistance, contrast sensitivity, etc.) for optimized multimedia analysis and quality assessment.

• Perception-based saliency models, foveated rendering, foveated displays, eyetracking, gaze-contingent rendering

Organizers

• Hantao Liu (LiuH35@cardiff.ac.uk), Cardiff University, UK

• Effrosyni Doutsi (edoutsi@ics.forth.gr), Foundation for Research and Technology - Hellas (FORTH), Greece

• Marc Antonini (am@i3s.unice.fr ), Université de Nice Sophia Antipolis, France

• Saeed Mahmoudpour (Saeed.Mahmoudpour@vub.be ), Vrije Universiteit Brussel, Belgium