基于对比度与空间位置特征的显著性区域检测

Image salient region detection algorithm based on contrast and spatial location

依据图像区域的对比度以及空间位置等先验视觉显著性知识,进行了自下而上、数据驱动的图像显著性区域检测。首先,提取图像中的前景区域,构造区域的对比度、空间位置特征函数,然后融合这些特征计算显著图。该算法将图像的空间关系与区域关系联系起来,得到了较精确的显著图。通过对国际上现有的公开数据集MSRA-1000的测试结果表明:本文算法可以抑制非显著区域干扰,显著图的一致性较高。同时,将本文算法的显著图应用于分割显著性区域,能够得到较好的分割效果。

A bottom-up data driven algorithm is proposed to extract image salient region by utilizing visual saliency prior knowledge such as regional contrast and spatial location etc.In this algorithm,first,the foreground region in the image is extracted.Then,the regional contrast and spatial position feature functions are constructed.Finally,the saliency map is acquired by fusing these features.The model well establishes the relationship between the spatial space and regional contrast,and generates saliency maps with precise details.The results demonstrate that the proposed method consistently outperforms existing saliency detection algorithms,suppresses non-significant regional disturbance,and yields higher consistent saliency map,when evaluated using one of the large publicly available data sets MSRA-1000.Meanwhile,the extracted saliency map is applied to automatically segment saliency region and better segmentation results are obtained.