Abstract

Aiming at automatically discovering the common objectscontained in a set of relevant images and segmenting themas foreground simultaneously, object co-segmentation hasbecome an active research topic in recent years. Although anumber of approaches have been proposed to address thisproblem, many of them are designed with the misleading assumption, unscalable prior, or low flexibility and thus still suffer from certain limitations, which reduces their capability in the real-world scenarios. To alleviate theselimitations, we propose a novel two-stage co-segmentationframework, which introduces the weak background prior toestablish a globally close-loop graph to represent thecommon object and union background separately. Then anovel graph optimized-flexible manifold ranking algorithmis proposed to flexibly optimize the graph connection andnode labels to co-segment the common objects.Experiments on three image datasets demonstrate that ourmethod outperforms other state-of-the-art methods.