From images to objects: Global and local completions of self-occluded parts.

Abstract

Listen

Translate article

When solid objects are viewed from only one direction, some parts are necessarily occluded. Expectations about possible completions of novel objects (cubes with indentations) were measured by the response times to determine whether two images could be two depth-rotated views from the same object. Completions are considered global if the completed indentations are identical to the visible indentations (i.e., they are globally simple because of the repetitions) and are considered local if they are locally simple (e.g., straight contours with a minimal number of additional surfaces). The results of three experiments show that local and global completion tendencies which were well-established for two-dimensional surfaces appear valid for three-dimensional objects as well. Additionally, it is shown that structural object aspects play an important role in object completion.