应用奇异值分解的海上场景显著性检测  

Saliency detection for sea visual scene using SVD

作　　者：任蕾[1] 施朝健[2] 冉鑫[2] 

机构地区：[1]上海海事大学电子工程系,上海201306 [2]上海海事大学航海系,上海201306

出　　处：《计算机工程与应用》2012年第23期161-164,172,共5页Computer Engineering and Applications

基　　金：国家自然科学基金(No.51109128);上海市重点学科(No.S30602);上海海事大学科研基金项目(No.20110050)

摘　　要：提出一种应用奇异值分解的海上场景显著性检测方法。提取海上场景图像中颜色和亮度各通道特征,并对各其分别进行奇异值分解,根据设定的阈值,选择各特征的典型分量。各特征的粗显著图定义为各特征和其典型分量的差。为进一步去除海杂波等干扰,在粗显著图中,计算其空间域全局显著性,以此形成显著性图。得到的颜色通道和亮度通道显著图通过线性合并为总显著图。利用海上场景图像进行了实验,结果表明提出方法的有效性。A saliency detection method for sea visual scene using Singular Value Composition（SVD）is proposed.Color and intensity features of sea visual scenes are extracted.SVD is conducted on each feature and the typical components are obtained based on certain threshold.The rough saliency map for each feature is defined as the difference between each feature and the typical components.To remove the remaining sea clutters,global saliency in spatial domain for rough saliency map is computed to provide the precise saliency map.The color and intensity saliency maps are linearly combined to achieve the master map.Experiments are made with images to show the validity and effectiveness.

关 键 词：奇异值分解 显著性检测 海上场景 空间域全局显著性 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]