Automatic image co-segmentation: a survey

Image co-segmentation is important for its advantage of alleviating the ill-pose nature image segmentation through exploring correlation between related images. Many automatic algorithms have been developed in last decade, which are investigated comprehensively this paper. We firstly analyze visual/semantic cues guiding co-segmentation, including object and cues. Then, we describe traditional methods three categories elements based, regions/contours common model based. In next part, deep learning-based reviewed. Furthermore, widely used test datasets evaluation criteria introduced reported performances surveyed compared with each other. Finally, discuss current challenges possible future directions conclude Hopefully, comprehensive investigation will be helpful development technique.