Two-Stage Method For Effective User-Guided Video Segmentation

Video segmentation is a complicated task and an active area of research. Many applications need user interaction to obtain desirable segmentation outcome. In this work, we describe a two-stage method to obtain desirable video segmentation within practical execution time, such that it can be verified and modified seamlessly by a user. In the first stage, we use a block-based clustering technique to reduce the complexity of the task. In the second stage, graph-based processing is used to exploit the spatiotemporal dependencies within video frames. User interaction in the form of selection of a few pixels to be labeled as RoI or non-RoI is used to initialize the graph in this stage. Experiments show that our method is effective in producing desirable video segmentation and is user-friendly.