Image Dataset
Candidate
Annotations
Final
Annotations
Image-to-Image
Relation (S*)
Image-to-Word
Relation (Y)
Word-to-Word
Relation (K*)
Image-based Graph Learning
Word-based Graph Learning
Visual Content
Features
Annotated
Images
Annotation Set
*
α α
*
β β
Fig. 2. Overview of image annotation framework.
mental comparisons among several related work and our scheme. Conclusions
and future work are given in section 6.
2 Image Annotation Framework Based on Graph Learning
The proposed framework consists of two learning processes denoted as ”basic
image annotation” and ”annotation refinement”, and three kinds of relations
as mentioned above. In the basic image a nnota t io n process, image-to-image
relation and image-to-word relation are integrated to obtain the candidate
annotations. In the annotation refinement process, the word-to-word relation
is explored to refine those candidate annotations from the prior process. An
overview of the framework is shown in Fig. 2.
2.1 Basic Image Annotation
The basic image annotation can be deemed as a semi-supervised learning
process on an image-based graph, i.e., propagating labels (annotations) from
annotated images to un-a nnotated images according to their similarities. The
learning process includes two key issues: one is how to measure the relations
among images (S
II
), especially the train- to-test image relation, and the other
is how t o model the distribution of words in annotated images (S
IW
). Then
4