- Shopping Bag ( 0 items )
Posted July 17, 2010
many unsolved problems
The book is a lengthy compilation of current research efforts in the adding of more intelligence to computing. Not surprisingly, many of the chapters refer to the Web 2.0 or the Semantic Web. Many challenges remain as the authors make clear.Was this review helpful? Yes NoThank you for your feedback. Report this reviewThank you, this review has been flagged.
One of these is described in Chapter 21, "Semantic Analysis for Multimedia Security Applications". The problem is how to programmatically extract meaning from videos, where these might be amassed via security cameras, or more generally from other cameras. A video consists of a sequence of frames. There is a hierarchy of structure, starting with a frame at the lowest level. A shot is a set of frames from 1 camera showing 1 event. While a scene is a series of shots taken at a single location. The entire video is then a collection of scenes. While a human observer can readily discern a shot and a scene, how is this done by computer? Various methods of using a histogram distance metric and a spatial distance metric are given. The chapter goes well beyond work confined to analysing single images.
Subsequent difficulties include how to find shot boundaries and getting a typical frame from a shot to define the shot; ie. how to get a keyframe? Another serious practical issue is the tracking of an object. Imagine following a person as she moves through a region. A higher level related task is to use motion to describe object activity in a video. So perhaps a parametric representation of the object in a 3 dimensional space can be tried. Where a bounding box could be found for an object in each frame [to the extent that this is indeed possible] and the centroid for that box is taken as the object center in establishing a trajectory.
Another common thread in several chapters is the use of ontologies. There have been and are massive efforts to bootstrap by building ontologies like WordNet. Various usages have been tried like extracting and using glosses (the textual description of a term in an ontology) for downstream analysis.
On the subject of ontologies, none of the chapters discussing this mention Ted Nelson's Xanadu, which was an early 1960s attempt at combining this with hypertext. Perhaps a sad reflection that that effort was too early and a failure.
Of the book's editors, Zadeh is clearly the best known, being the founder of fuzzy logic. Be aware that the text only has brief mentions of this topic.