Efficient block-wise temporally consistent contour extraction in image sequences

In this paper we investigate the problem of temporally consistent consecutive contour extraction in image sequences, including both single and multiple boundaries. By formulating this problem in the form of an optimal surface detection in 3D volume, we are able to resort to a graph-theoretic approach for exact solution. In order to cope with the high computational complexity caused by the potential unboundedness in time (i.e., an image sequence can be arbitrarily long) and heavy noise, we propose three approximate block-wise variants to accelerate the solution process. The effectiveness and efficiency of our approach is exemplarily demonstrated on simulated data and real ultrasound data for arterial wall detection. It is shown that the approximate variants dramatically reduce the computation time without loss of solution quality.