Interactive Image Segmentation by Semi-supervised Learning Ensemble
Jiazhen Xu, Xinmeng Chen, Xuejuan Huang · 2008
Due to the instinct difficulties of fully automatic segmentation, interactive image segmentation becomes a hot research topic in the past several years, and many approaches have been proposed. Recently, an approach viewing this task as a semi-supervised learning problem shows great promising. However, this approach ignores the influence of potential noise mingled in the user-provided information. Considering this information is mostly given by hand with the help of broad brush or region selection tools, the noise is inevitable. In this paper, We propose a more robust solution against noise by adopting Laplacian SVM method. We also develop an ensemble method to increase the performance. Experiment results demonstrate the improvement over the former approaches.