基于方向特征融合的感兴趣区域提取算法  

Extraction of region of interest using orientation feature fusion

作　　者：陈再良[1] 邹北骥[1] 李海冰[1] 史操[1] 

机构地区：[1]中南大学信息科学与工程学院,湖南长沙410083

出　　处：《华中科技大学学报（自然科学版）》2011年第12期102-106,共5页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：国家自然科学基金资助项目(60970098;60803024);国家自然科学基金重大研究计划资助项目(90715043);湖南省自然科学基金资助项目(11JJ3067);中央高校基本科研业务费专项资金资助项目(201021200062)

摘　　要：针对图像底层特征对感兴趣区域(ROI)提取的影响程度问题,提出了一种方向特征融合的ROI提取算法.利用眼动实验数据得到实际的ROI图和方向叠加图计算相似度,融合方向叠加图和Stentiford视觉注意图,经反馈和权重调整后,得到不同图像类别的最佳权重和ROI图.根据该算法对图像ROI提取的结果,分析了平均反馈距离和图像位置等方向特征对提取ROI图的不同影响.实验结果表明:该算法的总体效果较好,特别是对于方向特征显著且物体较少的图像,能准确地提取图中ROI.Low-level features in image would influence the points of extracted region of interest（ROI）.Thus,an extraction ROI algorithm was proposed based on orientation feature and visual attention map.The algorithm got the real ROI in image by the data of eyelink,and calculated similarity between superposition map and the real ROI,then integrated superposition map and the Stentiford visual attention map.The algorithm got suitable weight and ROI of image after feedback and adjusted weight.According to experimental data,the different effects of extraction ROI by orientation feature were analyzed.Experimental result shows that the algorithm can exactly extract ROI in image when direction feature is obvious and there are a few objects in image.

关 键 词：特征提取 反馈 方向特征 融合 感兴趣区域 注视焦点 

分 类 号：TP317.4[自动化与计算机技术—计算机软件与理论]