引导滤波的单幅图像前景精确提取

该文提出了一种利用引导滤波优化二值马尔科夫随机场图像分割结果的前景提取新方法。首先通过二值马尔科夫随机场图像分割模型将原图映射为加权图模型，然后利用最大流／最小割算法对该加权图模型进行最优分割，并通过二值化得到前景区域的初始分割结果；之后通过图像引导滤波对该初始结果进行优化处理，保证所提取的前景区域具有精确的边缘信息。实验结果表明该方法能从图像背景中精确的提取出前景区域，并且对光照、噪声均具有较强的鲁棒性。

Single Image Foreground Precise Extraction Based on Guided Filtering

This paper proposes a new method that using guided image filter in binary Markov random field （MRF） image segmentation result for image foreground extraction. First, we build image graph model of origi- nal image using MRF model, and then use max-flow/min-cut algorithm to separate the foreground and background area to get a binary segmentation result which only have 0 and 1, that get the initial foreground area; and then use original image to estimate precise alpha image to guarantee accurate edge of object region based on guided image filter. Experimental results show that the method can extract a clear foreground of background image, and with strong robustness of light, noise and a slight camera shake.