Synthesizing, Segmenting and Tracing Filamentary Structured Objects in 2D & 3D Images

Brief Summary

The aim of this project is to synthesize, segment and trace filamentary structured objects in 2D & 3D images.

For the segmentation, we have specifically proposed a value-added segmentation module, a feature learning method, and learning to segment un-annotated fundus images by simulations.

It is also important to seperate individual filamentary structured objects, which is described by our digraph-based tracing approach here.

Besides, we investigate the problem of generating realistic looking filamentary structured objects from a small training set.