基于前景优化和概率估计的显著性检测  

Saliency detection based on foreground optimization and probability estimation

作　　者：翟继友[1] 周静波[1] 霍瑛 ZHAI Ji-you;ZHOU Jing-bo;HUO Ying(School of Computer Engineering,Nanjing Institute of Technology,Nanjing 211167,China)

机构地区：[1]南京工程学院计算机工程学院,南京211167

出　　处：《控制与决策》2021年第4期917-924,共8页Control and Decision

基　　金：国家自然科学青年基金项目(61802174);江苏省高教学会评估项目(2018-C09);南京工程学院高等教育研究项目(2017YB15)。

摘　　要：为了对图像中的显著目标进行更精确的识别,提出一种基于前景优化和概率估计的区域显著性检测算法.所提出算法主要包括前景与背景线索选择、前景线索优化及基于概率估计的显著性区域检测3部分.首先,采用简单线性迭代聚类算法对图像进行初始分割;然后,分别检测图像的背景线索和前景线索,并利用背景线索对前景线索进行优化;最后,采用概率估计算法分别对基于背景线索和优化后前景线索进行显著性区域检测,并对两者结果进行融合.对比实验表明,所提出算法相比其他算法取得了较高的查准率,具有较好的检测性能.In order to recognize the salient objects in images more accurately, a saliency detection algorithm based on foreground optimization and probability estimation is proposed. The proposed algorithm mainly includes three parts, i. e.,foreground and background cues selection, foreground cues optimization and salient region detection based on probability estimation. Firstly, the simple linear iterative clustering algorithm is used for initial segmentation of images. Then, the background cues and foreground cues are detected from the given image respectively. In addition, the foreground cues are optimized by background cues. Finally, saliency maps are obtained using the probability estimation algorithm on the basis of the background cues and optimized foreground cues respectively. The results are fused to generate final saliency map. Experimental results show that the proposed algorithm achieves higher precision in comparison with other algorithms, which has better detection performance.

关 键 词：显著目标 前景优化 概率质量函数 显著性检测 线索 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]