Region Segmentation, Background and Salient Object Extraction by VSP and Global Features

Fumihiko Mori, Naotoshi Sugano

A new region segmentation and salient object extraction method is constructed based on virtual super pixel and global features. The salient object region was obtained through the following process: (1) extraction of the background regions, (2) calculation of the saliency of the segmented region, (3) unification of the low-saliency regions to the high ones. One of the important elements of the region saliency is the contrast between adjacent regions. However, it is hard to find a conventional method treating the adjacent situation between regions. In this paper the adjacent situation between regions was easily extracted using the VSP and it was used to calculate the contrast with adjacent regions. The salient objects extracted through the proposed method matched those extracted by humans and the scores of F-measure applied to the database CSSD composed of 200 images and the ECSSD of 1000 images were about the same as those presented in Yan et al. (2013) and Shi et al. (2015).

Journal of The Institute of Electrical Engineers of Japan

The Institute of Electrical Engineers of Japan

136

2

180

188

0385-4221

0385-4221

https://doi.org/10.1541/ieejeiss.136.180