A major limitation of graph cuts for the segmentation of large 2D image sequences is its interactive nature. The user has to provide seeds for almost every image frame to get an accurate segmentation. Straightforward approaches like direct copying of seeds provided in the first frame to other frames of the sequence fail in cases of great contrast or topological changes that occur when there is a large temporal distance between frames. In this work, we propose a dynamic seed propagation technique which can automatically adjust to any contrast or topological changes in a sequence. To this end, distance transform and skeleton extraction methods are employed to initialize the segmentation of the current frame using the result of the previous one. The proposed methods were used for the segmentation of functional cine-MRI sequences of colon which is especially challenging due to the large temporal distance between its frames. Both quantitative and qualitative results show good performance of the proposed method.     «

A major limitation of graph cuts for the segmentation of large 2D image sequences is its interactive nature. The user has to provide seeds for almost every image frame to get an accurate segmentation. Straightforward approaches like direct copying of seeds provided in the first frame to other frames of the sequence fail in cases of great contrast or topological changes that occur when there is a large temporal distance between frames. In this work, we propose a dynamic seed propagation technique...     »