Purchase this article with an account.
Mitchell van van Zuijlen, Paul Upchurch, Sylvia Pont, Maarten Wijntjes; Material property space analysis for depicted materials. Journal of Vision 2019;19(10):251a. doi: https://doi.org/10.1167/19.10.251a.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
The ability to estimate material properties enables us to visually differentiate between fresh and rotting food and warns us that red-glowing metal might be hot, for example. From our experiences interacting with materials, we have semantic knowledge which tells us for example, that steel is usually harder than silk, which implies the existence of a specific distribution of material properties per material. In the first experiment in the study by Fleming et al. (JoV 13(8) 2013) participants rated 10 photographed materials for 9 material properties. In the second experiment, they collected ratings for 42 material properties for 6 material names - i.e. only semantic information. They performed a principal component analysis (PCA) for both experiments, and found that photographic and semantic materials have a very similar representation in material feature spaces. In our study we had 316 participants estimate 10 material properties for 1350 stimuli (15 classes containing 90 exemplars) depicted in paintings. As expected, these materials show striking similarities with Flemings’ photographic and semantic materials on the distribution of materials upon the first two PCA dimensions. These first two primary components in a PCA visualize the majority of variability between the different materials, in other words these two components allow for differentiating between materials. This raises the question if the majority of variability within different materials is similar – i.e. do we use the same material properties to differentiate between and within materials? We applied a PCA on each individual material to examine the purely visual contributions on the material feature space and noted striking differences in the structure of these material feature spaces. This implies that we use both a general, robust material feature space to distinguish between materials, but defer to a specific material feature space for the differentiation within a material.
This PDF is available to Subscribers Only