Description »

Computer vision and image processing techniques have proven essential in Neuroscience. The large volumes of data acquired in this domain evidence the need for automated algorithms to make the processing of these data feasible. These automated algorithms generally employ Machine Learning methods to discover patterns in labeled data. The annotation process to generate this labeled data is usually done by the domain experts and, due to the size of the data involved, ends up being a major bottleneck.

The goal of this project is to develop an Intelligent Interactive Segmentation and Labeling Tool, helping the expert annotate salient structures in 3D. The starting point would be to cluster neighboring voxels into supervoxel regions, since supervoxel's boundaries are likely to be aligned with salient structures. Moreover, the expert is also to be assisted by interactive re-clustering of pixels in regions of interest while keeping the rest of the super-voxels intact. Constrained-clustering approaches would be investigated for this task.

As a means of demonstrating the feasibility and usefulness of this approach, an existing cross-platform User Interface (UI) would be used and extended as needed. A successful project would not only allow for the evaluation of the interactive annotation algorithms, but would also yield a powerful and easy-to-use tool for Neuroscience and Medical Imaging tasks.