【摘要】 为了缩小图像低级视觉特征与高级语义概念之间的差距,在物体层次上进行图像检索,论文提出了一种新的基于感兴趣区域(ROI)的图像细匹配法。与已有的基于感兴趣区域的检索方法相比,该方法很好地利用了感兴趣区域的空间信息,采用多分辨率检索思想,能够有效地对不同大小任意形状的物体进行检索。基于Corel图像库的实验结果表明,该方法能够有效地对各种物体进行检索,对物体的平移、缩放具有良好的鲁棒性。更多还原

【Abstract】 In order to shorten the gap of images between visual features and semantic concepts,this paper presents a novel image matching method based on regions of interest(ROI) to image retrieve at the level of objects.Compared with existing ROI-based image retrieval approaches,it makes use of space information of ROI and multi-resolution retrieval concept,is able to query shape-free and any size objects.Experiments based on COREL database have been proved that this method is effective to retrieve various objects,and robust to shifting and scaling of objects.更多还原