基于L0平滑的超像素块最短测地距离的显著区域提取方法

Salient region detection based on super-pixels and shortest geodesic distance a fter L0 smoothing

摘要:

为了有效地检测出图像中的显著区域,提出了一种基于L₀平滑 的超像素块最短测地距离的图像显著区域提取方法。首先对图像进行L₀平滑得到边缘锐化内部细节平滑 的图像,进而用简单线性迭代聚类(SLIC)算法得到图像的超像素块；然后提取超像素块的 颜色和空间特征,计 算超像素块之间的最短测地距离,根据图像边界的连通性得到每个超像素块的初步显著值； 最后通过显著 区域聚焦等后处理优化操作,滤除图中的背景噪声,得到最终显著图。利用公开数据库与之 前方法进行 比较的结果显示,本文方法能够更好地加亮整个显著目标区域,去除背景干扰区域。通过 ROC面积的比较,证明了本文提出方法的有效性。

Abstract:

To detect salient region in the image more effectively,an optimized algorithm b ased on the shortest geodesic distance of super-pixel blocks after L₀ smoothing i s proposed in this paper.Firstly,an edge-sharpening and interior-smoothing image is obtained by L_0smoot hing,and the simple linear iterative cluster (SLIC) algorithm is used for super-pixel blocks segmentation.Secondly,the shortest geodesic distances of super-pixel blocks are calculated based on the color distance of adjacent super-pixel bloc ks,and the initial salient value of each super-pixel block is obtained according to the boundary connectivity of t he image.Finally,the background noises in the initial salient map were filtered out by using the key region focu sing and other post-processing optimization,and the final saliency map is obtained.Compared with the results of previous methods in a public database,it is proved that our proposed method can better highlight the whole salient region and remove the background noise regions.The effectiveness of the proposed method is also prov ed by comparing the receiver operating characteristic (ROC) area of different methods.