A Probabilistic Associative Model for Segmenting Weakly Supervised Images

Zhang, Luming, Yang, Yi, Gao, Yue, Yu, Yi, Wang, Changbo and Li, Xuelong (2014) A Probabilistic Associative Model for Segmenting Weakly Supervised Images. Ieee Transactions On Image Processing, 23 9: 4150-4159. doi:10.1109/TIP.2014.2344433


Author Zhang, Luming
Yang, Yi
Gao, Yue
Yu, Yi
Wang, Changbo
Li, Xuelong
Title A Probabilistic Associative Model for Segmenting Weakly Supervised Images
Journal name Ieee Transactions On Image Processing   Check publisher's open access policy
ISSN 1057-7149
1941-0042
Publication date 2014-09-01
Year available 2014
Sub-type Article (original research)
DOI 10.1109/TIP.2014.2344433
Volume 23
Issue 9
Start page 4150
End page 4159
Total pages 10
Place of publication Piscataway, NJ United States
Publisher Institute of Electrical and Electronics Engineers
Collection year 2015
Language eng
Formatted abstract
Weakly supervised image segmentation is an important yet challenging task in image processing and pattern recognition fields. It is defined as: in the training stage, semantic labels are only at the image-level, without regard to their specific
object/scene location within the image. Given a test image, the goal is to predict the semantics of every pixel/superpixel. In this paper, we propose a new weakly supervised image segmentation model, focusing on learning the semantic associations between superpixel sets (graphlets in this paper). In particular, we first extract graphlets from each image, where a graphlet is a small-sized graph measures the potential of multiple spatially neighboring superpixels (i.e., the probability of these superpixels sharing a common semantic label, such as the sky or the sea).  To compare different-sized graphlets and to incorporate imagelevel labels, a manifold embedding algorithm is designed to transform all graphlets into equal-length feature vectors. Finally,  we present a hierarchical Bayesian network to capture the semantic associations between postembedding graphlets, based on which the semantics of each superpixel is inferred accordingly.   Experimental results demonstrate that: 1) our approach performs competitively compared with the state-of-the-art approaches on three public data sets and 2) considerable performance enhancement is achieved when using our approach on segmentation-based photo cropping and image categorization.
Keyword Probabilistic model
Weakly supervised
Segmentation
Associations
Q-Index Code C1
Q-Index Status Confirmed Code
Institutional Status UQ

Document type: Journal Article
Sub-type: Article (original research)
Collections: Official 2015 Collection
School of Information Technology and Electrical Engineering Publications
 
Versions
Version Filter Type
Citation counts: TR Web of Science Citation Count  Cited 47 times in Thomson Reuters Web of Science Article | Citations
Scopus Citation Count Cited 61 times in Scopus Article | Citations
Google Scholar Search Google Scholar
Created: Sun, 22 Feb 2015, 10:47:59 EST by System User on behalf of School of Information Technol and Elec Engineering