Information Technology Reference
In-Depth Information
Table 2.4
Five nearest tags in the learned tag subspace for each of the four selected tags
Selected tag Five nearest tags
Cat Grass, animal, pet, dog, vacation
Flower Blooms, butterfly, nature, spring, blossoms
Airplane Aircraft, travel, planes, photographer, airport
Buddhist Buddha, religion, buddhism, thailand, ancient
©[2012] IEEE. Reprinted, with permission, from Ref. [
34
]
T
able 2.5
Five nearest images in the learned image subspace for each of the four selected images
Image
Five Nearest Images
©[2012] IEEE. Reprinted, with permission, from Ref. [
34
]
We have employed smoothness constraints into the optimization function to pre-
serve the affinity structure in the low dimensional factor subspace. To show the
effectiveness of smoothness constraints, we show in Tables
2.4
and
2.5
the five near-
est tags and images for the selected tag and image, respectively. It is shown that
RMTF succeeds to mine the semantic correlations among tags and images from the
observed tagging data. Context- and semantic-relevant tags are close in the derived
tag subspace, while in the image subspace, visual and sematic similar images are
clustered together.
