中文摘要：

通过对图像分割与图像抠图的比较和分析，从图划分的角度考虑抠图问题，提出一种具有纠偏性的图像抠图的全局优化方法．该方法在最小化前景对象与背景相互分离的软分割开销的同时，最大化前景对象的内部关联度．理论分析和实验结果表明，与其他形式的抠图优化目标函数相比，文中方法能够更有效地提取出全局最优的抠图结果，有利于实现自动或半自动的抠图处理．

英文摘要：

A global algorithm of optimization of deviation rectification for image matting is presented, which treats the process of image matting as that of graph partition after carefully comparing image segmentation with image matting. While minimizing the cost of soft segmentation between foreground object and background, the approach can maximize the degree of association among the foreground object and background, the approach can maximize the degree ot association among the Ioreground objects. Both theoretical analysis and experimental process are studied to demonstrate that the proposed method can be more efficient to extract global optimal matte compared with other function expressions of matting optimization. This research will have a contribution to the implementation of automatic or semi-automatic matting.