Computational and Learning based Human Perception Modelling for Advanced Image Processing

Project Details

Description

Image processing is an important topic in many research fields such as
computer vision, assisted medical diagnosis, communication, etc. In
particular, many of today’s artificial intelligence (AI) systems (e.g. face
recognition, autonomous vehicle, human-like robotics, deepfake, etc.)
rely on advanced image processing. Traditionally, image processing
algorithms rely on heuristically formulated human domain knowledge,
where the design of such algorithms demands both good domain
knowledge and mathematical expertise. In recent years, deep learning
frameworks have been widely adopted for the analysis of images with
different content (document, natural scene, medical scanning, street
view, human face, objects, etc.), and have demonstrated new state-ofthe-art in almost all frontiers. We note that, in both computational and
deep learning based image processing frameworks, human perception
may be (to a rather limited extent) implicitly encoded. In this research
proposal we will explore the modeling of human perception for
advanced image processing. On the one hand, we will investigate the
combination of computational models (e.g. probabilistic graph
modeling) with understandings (e.g. using Gestalt principles) on human
perceptual recognition of complex visual environments. On the other
hand, we will also investigate (new) deep learning frameworks for joint
modeling of different perception primitives (e.g. visual and language) for
advanced (and semantic-aware) images processing..
AcronymOZR3994
StatusActive
Effective start/end date1/10/2230/09/26

Keywords

  • Computational Modeling
  • Deep Learning
  • Human Perception
  • Image Processing

Flemish discipline codes in use since 2023

  • Pattern recognition and neural networks
  • Computer vision
  • Knowledge representation and machine learning
  • Image and language processing

Fingerprint

Explore the research topics touched on by this project. These labels are generated based on the underlying awards/grants. Together they form a unique fingerprint.