基于图割的目标提取实时修正算法

Real-time correcting algorithm of extracted contour based on graph cuts

对目标提取结果进行后期修正是进一步提高提取精度的重要举措。针对在前期目标提取算法框架内实现修正困难、局限较多、难于推广等不足,提出了一种基于图割理论的独立修正算法。首先通过人机交互选定修正区域,然后映射成s-t网络,最后运用最大流/最小割算法对s-t网络进行切割得到修正后的目标轮廓。实验表明,该方法不仅操作简便,实时响应,不破坏已有的正确提取结果;而且参数少,抗噪能力强,适用于一般的基于轮廓的目标提取。

Post correcting of the extracted contour is an important measure to further improve extraction accuracy. Aiming at the difficulty and limitation of correcting implementation and generalization within the corresponding extracting algorithm framework, an independent correcting algorithm based on graph cuts was proposed. At first, the region to be corrected was obtained by human-computer interaction. Then the region was mapped into an s-t network. Finally a corrected object contour was obtained using the max-flow/min-cut algorithm. The experimental results show that the simplification, real-time responding, and strong anti-noise ability of the proposed approach are suitable for general contour-based object extraction.