Boundary connectedness based video cut for moving object segmentation

A new algorithm for automatic segmentation of moving objects in video based on spatio-temporal saliency and Neutro-Connectedness is presented in this paper. First, we propose a simple model to compute video saliency by combining initial saliency maps computed in spatial and temporal domains. Then, based on the detected spatiotemporal saliency map and temporal superpixels, initial background and foreground regions can be detected and taken as input of our proposed boundary connectedness based video cut (BC-video cut) to achieve moving object segmentation. Our model predicts jointly appearance models, Neutro-Connectedness, and pixel labels via an iterative energy minimization framework. Experiments show a good performance of our algorithm to segment moving objects on benchmark datasets.