Salient Object Segmentation Based on Automatic Labeling

摘要

This paper proposes an automatic salient object extraction framework. Firstly, the saliency model are developed by applying the low level color features and the boundary prior. The initial salient regions are extracted by adaptive thresholding. Multiple classifiers are trained with extracted initial region, which reflect color information of images or adopt label propagation. Then, the labels for segmentation are generated automatically via classifier composition. Finally, the conditional random field (CRF) model based on multi-feature fusion is applied for salient object segmentation. Empirical study reveals that the proposed algorithm achieves satisfying performance.