Abstract
Humans can judge liquid viscosity using only visual information. For example, honey is ‘thick’ (i.e. high viscosity), while water is ‘thin’ (low viscosity). Even though this sort of liquid viscosity estimation is mundane, vision science has not addressed what image processing underlies the estimation. The purpose of this study was to examine visual factors enabling us to estimate liquid viscosity, focusing on dynamic visual information. We employed the Blender physics engine to simulate kinetic viscosity of liquids, and created 50 movies (ten scenes each with five kinematic viscosities). The observers were asked to watch each of the movies, and subjectively rate liquid viscosity We first confirmed that the rated viscosity increased nearly in proportion to the simulated kinematic viscosity. To examine how simulated viscosity was related to the speed of liquid flow, we computed optical flow of liquid flow. The histogram of the speed of motion signals indicated that a liquid flow with a lower kinetic viscosity tends to have a higher overall speed. To examine the effect of overall speed of liquid flow on viscosity rating, we manipulated frame duration of each movie, and found that the rated viscosity increased with the frame duration. To investigate the importance of the spatial pattern of motion speed, we divided a movie into an array of cells (each of the cells was windowed by a tapered cosine circular function), and randomized image orientation within each cell (direction scramble), or shuffled the positions of cells (position scramble). Viscosity rating was generally unchanged when the stimuli were scrambled in these ways. However, when the cell size was small, scrambling position decreased the correlation between observers’ viscosity rating and simulated viscosity. The results indicate that in estimating liquid viscosity the visual system makes use of motion speed distribution.
Meeting abstract presented at VSS 2012