基于多模态信息融合的图像显著性检测算法研究  

Research on the Image Saliency Detection Algorithm Via Fusing Multimodality Information

作　　者：徐正梅[1] 王慧玲[1] 韦良芬[2] XU Zheng-mei;WANG Hui-ling;WEI Liang-fen(School of Computer and Information Engineering,Fuyang Normal University,Fuyang 236037;Department of Computer Science and Technology,Anhui Sanlian University,Hefei 230601,Anhui,China)

机构地区：[1]阜阳师范学院计算机与信息工程学院,安徽阜阳236037 [2]安徽三联学院计算机科学与技术系,安徽合肥230601

出　　处：《韶关学院学报》2018年第12期13-17,共5页Journal of Shaoguan University

基　　金：安徽省教育厅自然科学重点项目(KJ2016A252;KJ2018A0606);阜阳师范学院自然科学研究项目(2014FSKJ08)

摘　　要：图像显著性检测是以高亮的形式将图中最能引起用户兴趣、最能表现图像内容的区域标注出来.近年来,图像显著性检测得到快速发展,但在复杂场景中检测效果仍有待提高.采用两种方式提升显著性检测的泛化性能:其一,利用RGB图像和红外图像的互补性,融合两种模态信息进行检测;其二,考虑到图像跨边界问题,采用Kmeans算法对边界种子点进行筛选.在数据集RGBT-Saliency-Dataset上与当前流行的8种算法进行实验对比.实验结果表明,该算法能有效提升弱光照图像、图像包含噪声、显著目标跨边界以及前景背景相似等情况下的显著性检测效果.Image saliency detection marks the areas of the image that are most interesting to the user and can best represent the content of the image in highlight. In recent years, image saliency detection has been rapidly developed, but the detection effect in complex scenes still needs to be improved. We adopt two methods to improve the generalization performance of saliency detection. The first method is that the complementary of RGB images and thermal images are used to fuse two modes of information for detection and the second one is that the K-means algorithm is used to optimize the selection of boundary seeds on considering of the target cross-boundary problem. In RGBT-Saliency-Dataset,thealgorithm with other eight popular algorithms are compared. The experimental results show that the proposed algorithm can effectively improve the effect of saliency detection in the following cases:weak illumination images, image containing noise, saliency objects crossing boundary and saliency objects similar to background etc.

关 键 词：多模态 流行排序 显著性检测 K-MEANS算法 

分 类 号：TP309.2[自动化与计算机技术—计算机系统结构]