摘要
针对自然场景图像,提出了一种快速的显著对象自动分割方法.首先,将图像从RGB颜色空间变换至HSV颜色空间,利用色调和亮度等特征获得显著度图,得到待分割对象所在的区域；然后利用改进的分水岭算法将原始图像预分割为若干子区域,将这些区域描述为超像素,使用混合高斯分布描述其特征,用于构建图切分方法的网络图,经过迭代分割之后,获得最终的显著对象.实验结果表明,该算法与人工交互的图切分算法相比,能自动获得前景和背景的先验知识,在不影响分割精度的情况下,加快了分割速度.
Abstract
We propose an automatic salient object extraction method for nature scene in this paper. The method first changes images from RGB color space to HSV color space to obtain visual saliency map using chrominance and intensity information. Then pre-segments the input images to obtain super-pixels or over-segmentation regions by the improved watershed algorithm, which are described with mixture Gauss distribution. At last we take these super-pixels as nodes of a weighted graph to get the salient object iteratively. Our experimental results show that the proposed algorithm can automatically get a prior knowledge of the foreground and background without human interaction, at the same time speed up the segmentation without decreasing the accuracy of segmentation.
NETL
NSTL
维普
万方数据