摘要
为了解决基于二维图像的显著性检测方法中出现的光噪声、前景背景相似、多目标遮挡等问题,有效地突出显著区域并抑制背景区域,基于颜色、深度信息提出一种基于深度选择性差异及背景先验的显著性检测模型.首先,根据深度图质量进行颜色以及深度特征所占比例的调节;其次,依据深度图的内在特性,计算图像的基于深度选择性差异的显著性;然后,基于所获取的边界背景集合和基于深度先验的背景集合,计算图像的基于背景先验的显著性;最后,对前期得到的2个显著图进行融合及调整,并对显著图进行优化,得到最终的显著性检测结果.实验结果表明,该模型能较好地反映颜色以及深度信息对显著性检测的影响,计算结果更符合人类视觉.
Depth information plays an important role in the human visual system. Based on color and depth features,a saliency detection model was proposed in this paper based on depth selective difference and prior background prior. First,the proportion of color and depth features was adjusted according to the quality of depth map. Second,the saliency map based on depth selectivity difference was calculated according to the intrinsic characteristics of depth map. Then,based on the acquired boundary background set and depth prior background set,a saliency detection method was presented based on prior background prior. Finally,the two saliency maps obtained in the previous period were fused and adjusted,and the saliency maps were optimized to obtain the final saliency detection result. Experiments on two publicly available datasets show that the proposed method performs better than that of other state-of-the-art approaches.
作者
付利华
李灿灿
崔鑫鑫
王丹
彭硕
FU Lihua;LI Cancan;CUI Xinxin;WANG Dan;PENG Shuo(Faculty of Information Technology, Beijing University of Technology, Beijing 100124, China)
出处
《北京工业大学学报》
CAS
CSCD
北大核心
2019年第9期838-852,共15页
Journal of Beijing University of Technology
基金
北京市属高等学校高层次人才引进与培养计划资助项目(CIT&TCD201504025)
北京市自然科学基金资助项目(4173072)
北京工业大学基础研究基金资助项目(040000546317523)
关键词
三维视觉信息
显著性检测
深度选择性差异
背景先验
深度图
边界连通性
three-dimensional visual information
saliency detection
depth selective difference
background prior
depth map
boundary connectivity
