Early Cognitive Vision: Feedback Mechanisms for the Disambiguation of Earlyvisual Representation

In this monograph we present an Early CognitiveVision framework that aims at providing a rich andreliable scene representation from visualinformation. This framework preserves conflictinghypothesis in its early stages, and makes use offeedback mechanisms between different visualprocesses and layers of representation to achievedisambiguation. The first part presents how symbolic local image descriptors are extracted fromthe responses of early vision filters, and howperceptual grouping constraints can be applied to theresulting image representation. A second partdiscusses the use of stereopsis to reconstruct anequivalent 3D representation of visual information. Interactions between perceptual grouping, stereopsisand 3D reconstruction processes are discussed. Thethird part integrates visual information across timeto further disambiguate the representation. Thisframework has been used successfully in severalcontexts that are discussed in the conclusion of thisbook.