基于区域上下文感知的前景提取

Foreground extraction based on perception of regional context

为降低图像颜色分布和弱边缘对前景提取的负面影响,提出一种基于区域上下文感知的前景提取模型。结合图像亮度梯度和颜色信息将图像分成互不相交的区域,区域内颜色具有同质性,近邻区域间存在显著差异;依据颜色分布设计区域相似性测度,构建区域上下文关系和感知空间;在感知空间中将前景提取转化为二分类问题。实验结果表明,相对于传统模型,该算法提高了弱边缘、颜色非均匀分布图像的前景提取效果。

To reduce the negative impact of inhomogeneous regions and weak edges,a foreground extraction model based on perception of regional context was proposed.An image was divided into on-intersecting regions in terms of the color similarity and the intensity gradient properties.The color in a region was homogeneous,and there were significant differences between adjacent regions.The color distribution was analyzed,the regions similarity matrix and perception space were constructed.The foreground extraction was transformed to the binary classification problem in the perception space.Numerical examples indicate that,comparing to the traditional model,this model has better segmentation results for natural image with weak edges and inhomogeneous regions.