摘要
提出了一种基于似物性判定理论的单图像视觉目标检测算法。在组合几何学的引导下遴选候选图像窗口;应用创新提出的基于图像分割的结构化特征结合支持向量机对候选窗口的似物性进行评分;根据评分对候选窗口进行排序遴选。在PASCAL VOC2007数据集上进行了定量验证,结果表明:当候选集容量为1 000时,算法可达到96.1%的召回率。检测性能优于目标识别领域的4种经典算法。
An object detection method for single images based on objectness estimation theory is proposed.Original proposals are generated based on combinational geometry. The proposals are scored by segmentation-based structural feature and support vector machine. Proposals are sorted according to their score. Quantitative validation on PASCAL VOC 2007 dataset,when the number of the proposals is 1 000,the algorithm can achieve recall rate at96. 1 %. Its detection performnce outperforms four classic algorithms.
出处
《传感器与微系统》
CSCD
2017年第11期147-150,共4页
Transducer and Microsystem Technologies
基金
国家"863"高技术研究发展计划资助项目(GFJG-128205-E31401)
关键词
目标检测
似物性判定
基于图像分割的结构化特征
object detection
objectness estimation
image segmentation-based structural feature