24 Apr 10:15 Huan Xu: A Novel Graph Cut Model using Random Forest for Image Segmentation and Data Classification

HIIT seminars in spring 2009 will be held in hall **C222** of Exactum, on Fridays starting at 10:15 a.m. Coffee available from 10.

Fri 24 April
Huan Xu
The University of Science and Technology of China

A Novel Graph Cut Model using Random Forest for Image Segmentation and Data Classification

Abstract:
In this talk, we describe a segmentation algorithm that operates simultaneously in feature space and in image space. We define an energy function over both a set of clusters and a labeling of pixels with clusters. In our method, a pixel is labeled with a single cluster (rather than, for example, a distribution over clusters).
Combined with the random forest classification algorithm, our energy function penalizes clusters that are a poor fit to the data in feature space, and also penalizes clusters whose pixels lack spatial coherence. Furthermore, we give theoretical justification that our energy function can be efficiently minimized using graph cuts which could provide a global minimizer. Preliminary results are presented on segmenting real and synthetic datasets and images with different color space. Then we empirically show that our algorithm is able to outperform current state-of-the-art segmentation algorithm and classification algorithm.

Bio:
Huan Xu received the double bachelor of Computer Science & Mathematics and Applied Mathematics from Central China Normal University in 2006.
She is currently finishing her master degree (expected June 2009) in the University of Science and Technology of China, on the subject of pattern recognition and image segmentation.

From February, 09 to April, 09, she does her internship in the Heidelberg University of Germany. Her current research interests include image analysis and machine learning.