
Saunder September 1, 2015 16:18 K23081˙C007
80 Video Cataloguing: Structure Parsing and Content Extraction
using the K -means algorithm and assign the label of the centroid to each cluster.
These labeled patches are named representative feature patches.
So far, a movie scene s is supposed to be composed of a shot set, that is, s =
{t
1
,,t
n
}. Meanwhile, a panoramic-based key frame p
i
is obtained using the method
depicted in Chapter 4 for each shot t
i
. Specifically, the process for representative
feature patch extraction is depicted as follows:
1. For the panoramic key frame set P ={p
1
, p
2
,,p
n
}, we extract all the SIFT
features or points in each key