【摘要】 在逐块纹理合成中,Graph Cut方法被广泛用于优化块间重叠区域的像素取值。传统Graph Cut方法采用的累积距离度量,使得切割路径趋于走捷径而穿过高误差区域。针对此问题,提出了一种基于非标量距离度量的GraphCut方法。提出了一种基于该度量的高效最小割算法,并证明了其最优性;讨论了改进Graph Cut方法的规则性问题,并给出了解决方法。实验结果表明,由改进的Graph Cut方法所得到的切割路径更加曲折和平滑,块边界隐蔽性更强。更多还原

【Abstract】 Graph cut technique is widely used in patch-based texture synthesis algorithms to optimize patch boundaries. The traditional graph cut method is based on the cumulative distance metric which sometimes leads the path to taking short cuts through high cost areas.To overcome this problem,a graph cut method based on the non-scalar distance metric was proposed.A minimum cut algorithm based on this metric was presented,and its optimality was proved.The regularity problem of the improved graph cut method was discussed and a solution was provided.Experimental results show that the cutting paths by the improved graph cut method are smoother and more circuitous,and the patch seams are less obvious.更多还原