文献类型：期刊文献

英文题名：An Improved Computational Approach for Salient Region Detection

作者：Zhang, Qiaorong[1,2];Liu, Haibo[1];Shen, Jing[1];Gu, Guochang[1];Xiao, Huimin[3]

第一作者：Zhang, Qiaorong;张巧荣

通讯作者：Zhang, QR[1]

机构：[1]Harbin Engn Univ, Coll Comp Sci & Technol, Harbin, Heilongjiang, Peoples R China;[2]Henan Univ Finance & Econ, Coll Comp & Informat Engn, Zhengzhou, Henan, Peoples R China;[3]Henan Univ Finance & Econ, Coll Informat, Zhengzhou, Henan, Peoples R China

第一机构：Harbin Engn Univ, Coll Comp Sci & Technol, Harbin, Heilongjiang, Peoples R China

通讯机构：[1]corresponding author), Harbin Engn Univ, Coll Comp Sci & Technol, Harbin, Heilongjiang, Peoples R China.

年份：2010

卷号：5

期号：7

起止页码：1011-1018

外文期刊名：JOURNAL OF COMPUTERS

收录：EI(收录号：20102713057136);Scopus(收录号：2-s2.0-78651585633);WOS:【ESCI(收录号:WOS:000218005300004)】；

基金：This work was supported in part by a grant from National Natural Science Foundation of China (No. 60774041).

语种：英文

外文关键词：salient region; saliency map; visual attention; image processing; feature integration

摘要：Salient region detection in images is very useful for image processing applications like image compressing, image segmentation, object detection and recognition. In this paper, an improved approach to detect salient region is presented. The proposed method can generate a robust saliency map and extract salient regions with precise boundaries. In the proposed method, local saliency, global saliency and rarity saliency of three kinds of low-level feature contrast of intensity, color and orientation are used to compute the visual saliency. A new feature integration strategy is proposed in this paper. This method can select features and compute the weights of the features dynamically by analyzing the effect of different features on the saliency. Then a more robust saliency map is obtained. It has been tested on many images to evaluate the validity and effectiveness of the proposed method. We also compare our method with other salient region detection methods and our method outperforms other methods in detection results.