In this section, we introduce semi-MSL that finds a
unified low-dimensional and sufficiently smooth subspace
over all views simultaneously. To clearly explain the
Fig.3 shows the framework of this paper for cartoon
character retrieval and clip synthesis. First, cartoon
characters are represented by the four visual features.
According to PAF [2], semi MSL builds the local patches.
There are two types of local patches: local patches for
labeled characters and local patches for unlabeled
characters. Based on these patches, the part optimization
can be performed to get the low-dimensional subspace for
each patch. Afterward, all low-dimensional subspaces
from different patches are unified by whole alignment to