【摘要】 由于自然图像抠图具有高度的不确定性,目前的抠图方法中对于前背景颜色较为复杂的图片处理效果并不理想。本文首先通过采集大量的样本对来估计初始的掩膜值,但是通过采样得到的样本对并不能很好地估计掩膜值,为此定义未知像素与前背景样本点的空间距离和颜色距离函数来对其进行优化。通过最小化该代价函数来求得最优的样本对,使得最终得到的结果实现了准确性和鲁棒性的良好均衡。自然图像抠图实验结果表明,与其他几种抠图算法相比,本算法在视觉效果上和均差误差上都优于其他算法。更多还原

【Abstract】 Because matting is a ill-posed problem,many matting approaches perform poorly with the complex image. In this paper, we combine the propagation with sampling to collect foreground and background sample points. We define the space distance and color distance between the unknown pixels and the sample pairs,by minimize the function to choice the best sample pair, make the final results achieve the balance of accuracy and robustness. Experiments show that compared with other algorithms, the proposed method performs better in the quantitative and quantitative.更多还原