Abstract: RGB-D saliency detection aims to identify the most visually significant objects in a pair of RGB images and depth images. At present, the academic community has proposed various effective methods for single RGB-D image saliency detection, and these detection methods have complementary advantages. Therefore, the studies that the saliency maps generated by various methods are fused to improve the accuracy of saliency detection are indispensable. After studying the related work about saliency maps fusion, two levels of fusion process are proposed.Firstly, the initial saliency maps are generated by using a variety of off-the-shelf RGB-D saliency map detection methods. Then, the saliency map fusion at the image level and the pixel level is studied respectively, and the saliency maps obtained at these two levels are fused in proportion to obtain the final saliency map. The experimental results show that the effect of the saliency map fusion method is not only better than the single image saliency detection method, but also has certain advantages compared with other fusion methods.

Key words: RGB-D image, visual saliency map, saliency map fusion

摘要： RGB-D图像显著目标检测旨在从一对RGB图像和深度图像中识别视觉上最显著的目标。目前，学术界已经提出了各种有效的单幅RGB-D图像显著性检测方法，而这些检测方法之间存在着优势互补。因此，对各种方法生成的显著图进行融合以提高显著性检测精度的研究工作同样不可或缺。在对相关显著图像融合工作研究后，提出两个级别上的融合过程。首先，运用多种现成的RGB-D显著图检测方法生成初始显著图；其次，分别对图像级和像素级两个层面上的显著图融合工作进行了研究，再将这两个层面上得到的显著图按比例进行融合，得到最终的图像显著图。实验结果表明，该显著图融合方法的效果不仅优于单个图像显著性检测方法，而且和其他的融合方法相比，也具有一定优势。

关键词: RGB-D图像, 视觉显著图, 显著图像融合