一种协同显著目标自动分割方法  

Co-saliency Object Automatic Segmentation Method

作　　者：王超[1] 李静[1] 李东民[1] WANG Chao, LI Jing,LI Dong-min(College of Computer Science and Technology, Nanjing University of Aeronautics and Astronautics, Nanjing 211106, Chin)

机构地区：[1]南京航空航天大学计算机科学与技术学院,南京211106

出　　处：《小型微型计算机系统》2018年第4期830-835,共6页Journal of Chinese Computer Systems

基　　金：中央高校基本科研业务费专项基金项目(NS2015092)资助

摘　　要：图像分割是把图像中感兴趣的目标提取出来,广泛应用于图像识别、图像检索及目标追踪等领域,已成为国内外计算机视觉领域研究的一个热点.针对现有交互式分割方法需要用户进行有限步骤的交互,不能实现图像的自动分割这一问题,提出一种基于协同显著检测的多阶段显著目标自动分割方法.首先,利用基于聚类的协同显著目标检测方法获取协同显著图.然后,利用星形先验的图割方法和混合高斯模型拟合前景与背景,结合Grab Cut算法实现细分割.最后,利用显著目标的主动轮廓分割方法优化细分割的结果.在标准数据集上进行仿真实验,验证了提出方法的有效性.Image segmentation is to extract the image of interest in the target,which widely used in the fields of image recognition,image retrieval and target tracking,which has become a hot topic in the field of computer vision research both at home and abroad.Due to the existing segmentation methods requires the user to carry out steps of interaction,cannot achieve automatic segmentation of the image.The multi-stage segmentation method based on co-saliency detection is proposed.Firstly,co-saliency map is obtained using the cluster based co-saliency detection method.The low-level features are extracted from Lab and RGB color spaces in order to principle components method.Secondly,the star shape prior and the mixed Gaussian model are utilized to fit the foreground and background.Meanwhile,the GrabCut algorithm is used to achieve the fine segmentation map.Finally,the map is optimized by active contour method of the significant region.The effectiveness of the proposed method is verified by simulation experiments on standard datasets,comparing with other saliency segment method.

关 键 词：协同显著目标检测 自动分割 高斯模型 GRABCUT 主动轮廓模型 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]