Abstract
This paper proposes a novel unsupervised modeling of object categories by analyzing the statistics of the link structure (i.e., the relationships between visual features). Given a set of totally unlabelled images with a single number K, the number of object classes to be classified, our approach solves the following two subtasks: (1) Category discovery: clusters the input images into K groups according to object categories. (2) Localization: detects the most probable region of the object in each classified image.
The proposed approach represents a set of visual information in the
form of a large-scale network (Fig.1.(a)) and formulates the unsupervised
classification and localization in the visual modeling as the problem
of finding hubs and communities. The hubs behave like important class-specific
visual information and the communities map to object categories. Following
two link analysis techniques are applied to infer meaningful information
for unsupervised modeling: (1) PageRank for ranked importance of visual
features with respect to an image or a category, (2) Blondel et al's
vertex similarity algorithm for structural similarity (Fig.1.(b)).
 |
 |
|
(a) A tiny part of the proposed network representation (b)
Intution of structural similarity |
|
For category discovery, we observed that our approach achieved competitive performance against most of previous work. For localization, our approach fairly well discovered real class representative visual features as hubs and suppressed the trivial information such as background as shown in Fig.2.
 |
|
|
Figure 2. Examples of localization.
|
Publication
1. Gunhee Kim, Christos Faloutsos, and Martial Hebert, "Unsupervised Modeling of Object Categories Using Link Analysis Techniques", IEEE Conference on Computer Vision and Pattern Recognition (CVPR 2008), Alaska, USA, June 24-26, 2008. (Oral) (Oral acceptance = 62/1593 ~ 3.9%) [pdf][ppt]
Funding
- Intelligent Robotics Development Program, a 21st Century Frontier R&D Programs by the Ministry of Commerce, Industry, and Energy of Korea.
Copyright notice
The documents contained in these directories are included by the
contributing authors as a means to ensure timely dissemination of scholarly
and technical work on a non-commercial basis. Copyright and all rights
therein are maintained by the authors or by other copyright holders,
notwithstanding that they have offered their works here electronically.
It is understood that all persons copying this information will adhere
to the terms and constraints invoked by each author's copyright.