融合背景感知和颜色对比的显著性检测方法

摘要

Saliency detection is one of the hot topics in the field of image and video processing, at present,most of the methods need to be combined with some prior knowledge. For the problem of how to effectively set a pri-ori condition for the image with complex scenes, a new method of saliency detection based on background per-ception and foreground color contrast was proposed in this paper. Firstly, this paper introduced a method of boundary connectivity to detect the image background. It characterized the spatial layout of image regions with respect to image boundaries and was much more robust. Secondly, the foreground object was extracted by calcu-lating the contrast of color features. Thirdly, this paper proposed a principled optimization framework to integrate multiple low level cues to obtain the saliency map. Our experimental results on MSRA and CSSD benchmark datasets are efficient and the evaluation indicators perform better than 5 classical and 3 state-of-the-art similar methods.%显著性检测是图像和视频处理领域的研究热点内容之一，目前大多数显著性检测方法需要配合一定的先验知识。针对场景复杂的图像如何有效的设定先验条件的问题，提出一种基于背景感知和前景颜色对比的显著性检测方法。首先利用边界联通性的方法感知图像背景，由于该方法考虑了图像中物体的空间分布，因此具有较好的鲁棒性；然后利用颜色特征计算对比度提取前景目标；最后提出融合多个低级特征的优化框架来获得显著图。在 MSRA和 CSSD数据集上将文中方法与5种经典方法以及3种目前较为流行的相似方法进行对比实验，并通过多种指标进行评价，结果表明，该方法在复杂场景中适用性更强。