欢迎您！东篱公司
退出
-
申报数据库
-
立项数据库
-
成果数据库
-
项目获奖数据库
中文摘要：
参数最小割是一种常用的似物性推荐方法,主要用于在图像中快速定位物体区域。针对该类方法中容易生成大量无效前景种子影响处理效率,提出一种基于层次化融合逐级筛选的前景种子生成算法。基于由颜色、纹理复杂度控制的层次化融合方法得到候选区域集,从候选区域集中结合尺度变化率选出具有稳定外观的候选前景种子,最后基于似物性分数排序,确定有效的前景种子。实验结果表明,提出的前景种子生成算法具有较高的物体发现率,将其应用于参数最小割方法中,在使用更少的种子、生成较少区域时,可达到与前沿算法相近的区域级物体定位能力。
英文摘要：
Parametric min-cut is a kind of common objectness proposal method, which is mainly used to quickly locate objects in the images. For such method easily generates a lot of ineffective foreground seeds will affect speed, this paper presented a foreground seed generation algorithm, which was based on the hierarchical grouping to filter ineffective seeds. Firstly, based on the color, texture complexity to control hierarchical grouping for generating a set of candidate regions, and then, according to the scale changing rate to select candidate foreground seeds that had stable appearance. Finally, according to the objectness scores rank to determine the effective seeds. Experimental results show that the proposed algorithm has a higher discovery rate. Used in parametric rain-cut, it will generate fewer regions, while keeping the similar regional-level objects positioning capability with the state-of-the-art methods.
同期刊论文项目
同项目期刊论文
期刊信息
位置：