基于多特征图像显著目标的自动识别_片兆宇.pdfVIP

28 4 Vo.l 28, No. 4 2009 8 J. InfraredM illmi . W aves August, 2009 : 1001 9014( 2009) 04 0293 05 片兆宇, 孟祥萍, 舒英利 ( , 130012) : 现有的许多显著目标检测算法, 大都依赖像素间的相互 系, 而缺乏对焦点对象特征的理解. 本文提出了 一 种无监督的显著目标自动识别算法. 首先, 应用具有仿生学特质的多尺度 Gabor模型检测图像中的初级特征, 构成 显著目标的基本元素特征. 在此基础上, 结合显著目标的自身特性定义图像的局部特征, 进 一步确定感兴趣区域的 位置. 最后, 在模糊阈值算法的基础上提出 一种新的提取策略, 并将其应用到由初级特征和局部特征构成的显著图 上, 从而准确地确定显著目标的位置. 将此方法应用于具有不同特点的图像进行仿真实验, 得到了较好的结果, 证 明该算法是切实可行的. : 显著目标; 伽柏滤波器; 差异率; 模糊阈值 : TP 391 : A AUTOMATIC IDENTIFICATION OF SALIENT OBJECT BASED ONMULTI-FEATURES IMAGES PIAN Zhao Yu, MENG X iang Ping, SHU Y ing L i ( School of Electrical Inform ation Technology, Changchun Insititu te of Technology, Changchun 130012, China) Abstract: M any existing detecting algorithm s of salient objectm ostly depend on the relationsh ip between the p ixels and lack the com prehension of salient object features. An unsupervised algorithm for autom atically extracting the interesting object w as presented. Firstly, the jun ior featurew as extracted as basic elem ents of salient object by using am ultiscaleGaborm od elw ith b ion ics characteristics. Then by comb in ingw ith the understanding know ledge of the salient object, the local feature of mi age w as defined and the salient regionsw as further located. Finally, a novel extracting scheme based on fuzzy thresh old approach was proposed, and was used in the consp icuitym ap integrated by the jun ior and local features to locate the sa lient object. By applying the algorithm to various mi ages of smi u lation expermi ents, the satisfying resultsw ere obtained. The results show that the algorithm is feasible. K ey ords: salient objec