Retrospective on “Interpreting line drawings as three-dimensional surfaces”

✍ Scribed by Harry G. Barrow; J.M. Tenenbaum

• Publisher: Elsevier Science
• Year: 1993
• Tongue: English
• Weight: 568 KB
• Volume: 59
• Category: Article
• ISSN: 0004-3702
• DOI: 10.1016/0004-3702(93)90172-8

✦ Synopsis

In 1977 we proposed a computational model of visual perception which focused our own work and influenced that of a number of others (Barrow and Tenenbaum [4]). A key feature was the simultaneous recovery of retinotopic arrays of characteristics intrinsic to the scene, such as surface color, depth, and orientation. Such a representation appeared useful in its own right (e.g., for obstacle avoidance and grasping) and as a facilitator for later stages of vision, such as finding surfaces and their boundaries, or recognizing objects. We were encouraged that humans seem able to estimate intrinsic scene characteristics reliably under a wide range of conditions. Moreover, algorithms (based on very particular assumptions) had been developed for finding surface shape from shading (Horn [13] ), or stereo (Marr and Poggio [23] ), or for finding reflectance or color from brightness gradients (Horn [12]).

The central problem in recovery is the confounding of intrinsic scene characteristics in the image data. The brightness measured at each point in an image depends upon several independent characteristics of the corresponding scene point, primarily the intensity of incident illumination, the