Objectness to assist salient object detection

• DOI: 10.1049/iet-ipr.2015.0487
• Publication Date: 2016-02-08T22:06:35Z
• Authors: Xiaoliang Sun; Ang Su; Shengyi Chen; Qifeng Yu; Xiaolin Liu
• Keywords: 13. Climate action; 0202 electrical engineering, electronic engineering, information engineering; 02 engineering and technology; 7. Clean energy

When dealing with salient object that contains several regions different appearances, detection can be a difficult task as often only parts of the are highlighted and consistency between is poor. This study tackles this problem by introducing objectness to assist detection. Rather than treating in same manner other low‐level cues (e.g. uniqueness, location etc.) for determination regional saliency values, authors emphasise should also play significant role tuning regions. The integrate objectness, uniqueness centre bias find potential then enforce these using full‐connected Gaussian Markov random field weights determined score. Experimental results on public benchmark datasets indicate authors’ method performs well many images which cannot detected traditionally.