基于相似性融合的显著区域检测方法研究  

Research on Significant Region Detection Based on Similarity Fusion

作　　者：蒋益锋 胡琳娜[2] 刘冉冉[3] JIANG Yifeng;HU Linna;LIU Ranran(Information Center,Jiangsu University of Technology,Changzhou 213001,China;Zijin College,Nanjing University of Science&Technology,Nanjing 210046,China;School of Automotive and Transportation Engineering,Jiangsu University of Technology,Changzhou 213001,China)

机构地区：[1]江苏理工学院信息中心,江苏常州213001 [2]南京理工大学紫金学院,江苏南京210046 [3]江苏理工学院汽车与交通工程学院,江苏常州213001

出　　处：《江苏理工学院学报》2021年第6期50-58,共9页Journal of Jiangsu University of Technology

基　　金：国家自然科学基金项目“基于分解-协调机制的非均匀采样非线性系统辨识方法研究”(62003150)。

摘　　要：提出一种简单有效的基于相似性融合的显著区域检测方法:首先,通过最小方差量化降低颜色数,在将量化图像分割为超像素后,融合环绕性和边界连通性线索计算超像素的加权融合显著值,加权系数基于融合方法的显著相似性获得;同时,引入融合修正以增强前景区域;最后,通过显著性平滑和增强抑制图像背景,得到高亮且均匀的显著区域检测结果。为了验证所提出方法的有效性,在ASD、ECSSD、ImgSal三个公开数据集上与8个现有方法进行了性能比较,实验结果表明:所提出的融合机制可有效提升MaxF和S-measure等性能指标,并得到更为理想的检测结果。In this paper,a simple and effective salient region detection method based on the similarity fusion is presented.Firstly,the color number is reduced by minimum variance quantization.After the quantized image is segmented into superpixels,the weighted fusion saliency value of superpixels is calculated by fusing the surrounding and boundary connectivity clues,and the weighted coefficients are obtained based on the significant similarity of the fusion method.In addition,a fusion correction is introduced to enhance the foreground area.Finally,high-bright and uniform detection results of salient regions are obtained by smoothing and enhancing the image background.In order to verify the effectiveness of the proposed method,eight existing methods are compared on ASD,ECSSD and ImgSal public datasets.The experimental results show that the proposed fusion mechanism can effectively improve MaxF and S-measure performance indicators,and obtain more ideal detection results.

关 键 词：显著性 显著区域检测 相似性 融合 

分 类 号：TN911[电子电信—通信与信息系统]