The similar cartoon characters has to be retrieved
from cartoon database, which is prepared by extracting the
characters from cartoon videos as shown in fig 1using
LOG (Laplacian of Gaussian) filter , then color histogram,
hausdorf edge feature, skeleton features and curvature
scale space are extracted from cartoon characters. Fusion
of these features is done by using semi-MSL algorithm.
Alternating optimization are utilized by semi-MSL to
obtained the low dimensional subspace. Based on each
low dimensional subspace, a standard retrieval procedure
is conducted for all characters in the database. For
example, one cartoon character is selected as a query, and
then all the other characters in the database are ranked
according to the Euclidean distance to the query computed
in the low-dimensional subspace. For each query cartoon
character, the top 10 characters are obtained. The detailed
work of the project is explained in the following sections
(cartoon image feature extraction and semi-MSL).