基于对称性区域过滤的显著性区域检测方法  

Salient Region Detection Method Based on Symmetric Region Filtering

作　　者：董本志[1] 陈迪 景维鹏[1] DONG Benzhi;CHEN Di;JING Weipeng(College of Information and Computer Engineering,Northeast Forestry University,Harbin 150040,China)

机构地区：[1]东北林业大学信息与计算机工程学院,哈尔滨150040

出　　处：《计算机工程》2019年第5期216-221,共6页Computer Engineering

基　　金：国家自然科学基金(31770768)

摘　　要：为获得精确、完整的目标区域分割图,提出一种基于对称性区域过滤的检测方法来进行图像分割。利用改进的简单线性迭代聚类算法将图像分割成若干超像素,并以超像素为节点建立吸收马尔科夫链。计算转移节点到吸收节点的被吸收时间,将其作为显著值来获取显著图。根据图像目标区域的对称性特征,对显著图进行对称性检测,获取对称轴,通过两侧像素点到对称轴的距离对图像显著值进行区域过滤,从而获得目标图像分割区域。实验结果表明,该方法提取的图像显著目标区域较阈值分割法、最小生成树法和LRR法提取结果更为完整。In order to obtain the accurate and complete target region segmentation image,a detection method based on symmetric region filtering is proposed.The improved Simple Linear Iterative Cluster(SLIC) algorithm is used to segment the image into several superpixels.The absorption Markov chain is established with superpixels as nodes,and the absorption time from the transfer node to the absorption one is calculated as the saliency value to obtain a saliency map of the image.According to the symmetric feature of the target region,the symmetry of the saliency map is detected and its symmetric axis is obtained.Based on the distance between the side pixel points to the symmetric axis,the saliency value of the image is filtered,and the segmentated region of the target image can be obtained.Experimental results show that compared with the threshold segmentation,minimum spunning tree and LRR methods,the method in this paper can extract salient target regions more completely.

关 键 词：显著性检测 超像素 吸收马尔科夫链 对称性检测 图像分割 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]