基于SIFT特征的图像检索.pdfVIP

第25卷第 2期 计 算 机 应 用 研 究 V o.l 25 N o. 2 2008年2月 ApplicationR esearch of Computers Feb. 2008 * SIFT , , (厦门大学计算机系人工智能研究所, 福建 厦门 361005) : 提出一种多尺度图像检索算法, 该算法基于 IFT特征提取, 它将一幅图像转换成特征向量的集合, 图 像间的相似距离是通过计算两幅图像特征向量间的欧氏距离来实现的实验结果很好地说明了该算法具有尺 度平移旋转不变性, 一定的仿射光照不变性以及算法能很好地应用在特定形状特征目标的检索中 : 基于内容的图像检索; 尺度不变特征变换特征; K-L变换; 近似最近邻搜索; 欧氏距离 : TP391. 3 : A : 1001- 3695( 2008) 02- 0478- 04 Image retrieval based on IFT features WU Ru-i hang, LI hao-zi, ZOU Feng-m ei (Institute of A rtfi icial Intelligence, D ep t. of ComputerScience, X iamen University, X iamen Fuj ian 361005, China) Ab stract: Th is paper presented a new mi age retrieval algorithm wh ich used a new class of mi age features as the mi age de- scriptors, the IFT features. It transform ed the mi age into a set of IFT features. The smi ilarity w as computed using Euclidean d istance. Expermi ents show that the features are invariant to mi age scaling, translation, rotation, and change in viewpoint, and partially invariant to illum ination changes and affine. It is qu ite applicable to typ ical mi age retrieval. K ey word s: content-based mi age retrieval; IFT( scale invariant feature transform) features; K-L transform; ANN searching; Euclidean d istance , , b) IFT, , c) , IFTd) , IFT , , , IFT, , , K-L,