Hypergraph coarsening for image superpixelization

Image segmentation is a hard task and many methods have been developed to alleviate its difficulties. A common preprocessing step designed for this purpose is to compute an over-segmentation of the image, often referred to as superpixels. In this paper, we propose a new approach to superpixels compu...

Full description

Saved in:
Bibliographic Details
Main Authors: Ducournau, Aurélien, Rital, Soufiane, Bretto, Alain, Laget, Bernard
Format: Conference Proceeding
Language:English
Subjects:
Clustering algorithms
Computer vision
Hypergraph coarsening
Image color analysis
Image segmentation
Partitioning algorithms
Pixel
Superpixels algorithm
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Image segmentation is a hard task and many methods have been developed to alleviate its difficulties. A common preprocessing step designed for this purpose is to compute an over-segmentation of the image, often referred to as superpixels. In this paper, we propose a new approach to superpixels computation. In a first step, a hypergraph-based representation of the image is built. Then, a coarsening approach is operated on the resulting hypergraph to group pixels which belong to the same homogeneous region. This leads to a smaller hypergraph where each component represents a superpixel of the image. Our approach is very fast and can deal with great sized images. Its reliability have been tested on several real images from nature scenes with comparison to other methods. We show in particular that hypergraphs offer a more accurate image representation than graphs.
DOI:10.1109/ISVC.2010.5654894