Hybrid ROI extraction with bottom-up and top-down visual attention

作者：Chen Zailiang; Zou Beiji^*; Li Jianfeng; Shen Hailan; Mao Yun

来源：Journal of Information and Computational Science, 2011, 8(15): 3481-3488.

• Low-level features
• Regions of interest
• Topdown
• Visual Attention
• Weight

This paper proposes a hybrid extraction method of region of interest (ROI) with bottom-up visual attention (VA) and top-down VA. The proposed method takes advantages of two bottom-up models, i.e. the Itti-Koch model and Stentiford model to obtain saliency regions of images effectively. Additionally, to reduce the semantic gap between the observer';s query intension and the low-level features of images, the method adopts a way of transforming top-down information into the weight of low-level features in saliency map. Experimental results show that the proposed method can exactly extract ROI, and the ROI is in accord with the observer';s query intentions.

• 出版日期2011
• 单位中南大学