Auto-marking Image Segmentation Based Manifold Ranking

作者：Zeng X H^*; Yi R H; Zhu S W; He S S

来源：International Conference on Artificial Intelligence and Industrial Engineering (AIIE), 2015-07-26 to 2015-07-27.

• image segmentation
• manifold ranking
• saliency
• background prior
• auto-marking

Interactive image segmentation requires adjusting label information manually which will lead to tedious marking process. We propose an auto-marking image segmentation method. It can obtain the object prior and background prior automatically. We segment an image into superpixels (regions) and get the object saliency map via the manifold ranking in the guide of background prior information, then we choose part of the superpixels with higher saliency values as object marked seeds, and select the background marked seeds with the combination of background prior and the result of manifold ranking, thus obtaining the final image segmentation through maximal similarity based region merging. Experimental results on images with single object and similar adjacent objects show that the proposed algorithm can automatically add the correct label information, and can obtain segmentation accuracy that is better than saliency-seeded region merging (SSRMf) algorithm, while is more convenient than interactive segmentation by avoiding manual operations.

• 出版日期2015
• 单位重庆大学