A sparse-to-dense method for 3D optical flow estimation in 3D light-microscopy image sequences

We present a two-stage 3D optical flow estimation method for light microscopy image volumes. The method takes a pair of light microscopy image volumes as input, segments the 2D slices of the source volume in superpixels and sparsely estimates the 3D displacement vectors in the volume pair. A weighted interpolation is then introduced to get a dense 3D flow field. Edges and motion boundaries are considered during the interpolation. Our experimental results show good gain in execution speed, and accuracy evaluated in computer generated 3D data. Promising results on real 3D image sequences are reported.