Hypergraph Cuts & Unsupervised Representation for Image Segmentation

  • Authors:
  • Soufiane Rital

  • Affiliations:
  • Institut TELECOM, TELECOM ParisTech, LTCI CNRS, France. E-mail: soufiane.rital@telecom-paristech.fr

  • Venue:
  • Fundamenta Informaticae
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

This paper presents a novel approach to image segmentation based on hypergraph cut techniques. Natural images contain more components: Edge, homogeneous region, noise. So, to facilitate the natural image analysis, we introduce an Image Neighborhood Hypergraph representation (INH). This representation extracts all features and their consistencies in the image data and its mode of use is close to the perceptual grouping. Then, we formulate an image segmentation problem as a hypergraph partitioning problem and we use the recent k-way hypergraph techniques to find the partitions of the image into regions of coherent brightness/color. Experimental results of image segmentation on a wide range of images from Berkeley Database show that the proposed method provides a significant performance improvement compared with the stat-of-the-art graph partitioning strategy based on Normalized Cut (Ncut) criteria.