遥感图像中基于视觉显著性分层目标检测_张国敏.pdfVIP

37 3 吉林 大学 学 报 ( 工学 版) Vol.37 No.3 2007 5 Journal of Jilin University (Engineering and Technology Edition) ay 2007 张国敏,殷建平,祝 恩, 强永刚 ( 国防科学技术大学 计算机学院, 长沙410073) :针对遥感图像自动判读的应用场景, 提出了 种基于视觉显著性的分层目标检测方 法该方法对于每 个特征定义 个偏好函数, 并定义 个将各偏好函数计算的多 个视觉显 著性图像合成为 个视觉显著性图像的合成偏好函数首先对低分辨率目标图像分别使用各 偏好函数计算对应特征的视觉显著性图像;然后使用合成偏好函数计算合成的视觉显著性图 像; 最后定位出合成偏好函数取得最大值的区域, 并将其作为最有可能包含目标的候选区域 这三 个步骤在定位出的候选区域对应更高分辨率的图像上反复进行, 直到待检测目标被发现 或者被确定不存在为止实验结果证明,该方法能够尽早地略过那些不包含目标的区域, 与传 统的目标检测方法相比, 具有更好的检测效果和更少的计算量 :信息处理技术;遥感图像; 视觉显著性;特征;目标检测 :TP391 :A : 1671-5497(2007) 03-0625-05 Hierarchical object detection in remote sensing image based on vision saliency Zhang Guo-min, Yin Jian-ping, Zhu En, Qiang Yong-gang ( School of Comp uter , National Univ ersity of Def ense Technology , Changsha 410073, China) Abstract:A hierarchical object detection method based on vision saliency was proposed for automatic remote sensing image interpretation. A bias function for each single feature was defined, and a compound bias function for synthesis of vision saliency image computed by each function was also defined. For the low resolution original image, each bias function was used to generate a vision saliency image corresponding to a certain feature. Then synthesized vision saliency image was computed using the compound bias function from these vision saliency images. Furthermore, thearea where the compound bias function was maximum was located as the most promising object area. These three stepswere repeated over the located areas in higher resolutionlevel image until the objects were found or assuredly absent. Experiments show that this method can skip over the region where there is no object as early as possible, get better detection result compared with