Robust retrieval of three-dimensional structures from image stacks

Robust high-breakdown-point location estimators are employed to analyze image stacks under the piecewise constant image structure model. To reduce the effect of bias along the Z-axis, the class parameters are extracted using three consecutive slices. The segmentation algorithm first determines the most reliable seed regions, which are then used in a region-growing procedure supported by local evidence. The robustness and stability of the proposed technique is shown with both synthetic and real data, the latter consisting of one MRI and one confocal microscopy set. The performance of the algorithm is consistent with the ground truth obtained with manual segmentation by physicians.