Visual Fingerprint of Material Appearance

Real-world comprises thousands of materials having different surface
appearances. These appearances are a key in our everyday judgements of
material properties, which are based on our past experience,
recognition, and usage of these materials.
Although current technology allows for realistic reproduction of
material appearance for visualization and quality control purposes,
sharing of materials and their properties information across different
measured representations and software platforms is still rather
complicated. This problem relates to digital material appearance
assessment over entire pipeline of its lifetime, i.e. from its
acquisition, measured data modeling, to its visualization.
The goal of this project is to create material identifier encoding its
perceptual visual features. These features obtained by judgements of
observers in psychophysical studies form so called material fingerprint.
Based on sparse material measurements and the qualities judgements, we
plan to provide their generalization and propagation in a form of
ontology of existing material appearances based on methods of adaptive
machine learning. Such a system would assess regular measurements of any
material appearance and provide its perceptual fingerprint allowing its
efficient categorization, retrieval and sharing among users.