Evaluation of Salient Object Extraction Method by Virtual Super Pixel

Fumihiko Mori, Naotoshi Sugano

A method of salient object region extraction by virtual super pixel(VSP) is applied to ECSSD(extended complex saliency dataset) composed of 1000 images and compared with state of the art using F-measure．The F-measure at zero threshold of the presented method was about the same as those at the best threshold of the-state-of-art methods. ECSSD is harder than MSRA1000 as the highest frequency of the latter is at around 0.9 of the chi-square distance between GT and the background and it of ECSSD at around 0.25.

Proceedings of the Annual Conference of Biomedical Fuzzy Systems Association

Biomedical Fuzzy Systems Association

28

265

268

https://www.jstage.jst.go.jp/article/fss/31/0/31_715/_pdf