Abstract
We address the issue of measuring the perceived spatial layout of pictorial space. We focus on a modified version of a recently developed relative size task in which observers had to equalize the relative sizes of an object pair in pictorial space. In our version of the task, one object was used as a fixed size standard probe throughout each experimental session. Two photographs were used, both of the interior of a campus building entrance hall. 53 and 61 pictorial sample points were selected, respectively. We collected ground truth data of the actual scene by measuring for each point the distance to the camera with a laser distance meter, ranging from 9 to 46 meter.
Five observers each completed 12 experimental sessions: two photographs, two standard probe positions and three repetitions. The repeated trials consistency, quantified by the coefficient of determination, was generally high: on average 0.86 (lowest 0.72 and highest 0.94). The mean results of the three repetitions were used for further analysis. For pictorial relief (3D shape) it has been previously found that observers use different mental viewpoints, as quantified by an affine transformation resolving differences between observers. In our spatial layout experiment, we found very few of these cases. This could indicate a categorical difference in perceiving surface relief and spatial layout. Furthermore, we compared the observer settings with the ground truth data. Perhaps surprisingly, we found that observers were generally very close to the ground truth. Coefficients of determination between observers' settings (averaged over three repetitions) and ground truth data were on average 0.93 (lowest 0.88 and highest 0.97). Overall perceived depth was not consistently more compressed or stretched than the actual depth. Also, no significant affine correlations were found. The results seem to suggest that humans perceive pictorial spatial layout differently from pictorial relief.
Meeting abstract presented at VSS 2012