Abstract: Existing salient object detection algorithm based graph-based manifold ranking is less effective in detecting images with complex background due to its idealistic prior background assumption. This paper proposes an improved algorithm based upon affinity propagation clustering and graph-based manifold ranking. First, the background superpixels on the boundary is extracted according to their color contrast. And then the affinity propagation clustering algorithm is utilized to adaptively obtain the color clusters which are used to compute the saliency of object and background as queries in the manifold ranking. Finally, the salient map is determined by integrating multiscale saliency. This proposed algorithm is compared with nine state-of-the-art methods in terms of precision, recall, F-measure, PR curves, AUC values and visual effect on four popular and public datasets of ASD, ECSSD, DUTOMRON and SED2, and the experimental results show the improvements over the state-of-the-art methods.

Key words: salient object detection, saliency, background prior, manifold ranking, affinity propagation clustering

摘要： 针对现有基于图的流行排序的显著目标检测研究算法对于背景先验假设过于理想导致其在复杂背景图像检测中效果较不佳的问题，提出一种基于仿射传播聚类和流行排序的改进算法。首先根据位于边界的超像素集的颜色对比度进行背景提取；然后在背景估计和前景估计的显著性计算中利用仿射传播算法将提取的背景按颜色自适应聚类，根据各聚类簇分别采用经典的流行排序算法计算显著性，最后合并排序结果并融合多尺度显著值得到最终的显著图。在常用的公开的ASD、ECSSD、DUTOMRON和SED2数据集上与九种流行算法就准确率、召回率、F-measure、PR曲线和AUC值等指标和直观的视觉检测效果进行了比较，证明了所提算法的有效性。

关键词: 显著目标检测, 显著性, 背景先验, 流行排序, 仿射传播聚类