Computer Vision for Connectomics
The field of Connectomics (mapping neural circuits) emerged in recent years with the development of high throughput imaging methods for the recording of brain tissue. Electron Microscopy (EM) allows resolving neural circuits at synaptic level and thus has great potential to push forward our understanding of nervous systems.
In the last decade, automatic neuron segmentation has become an active part of computer vision research, spanning several subfields of applied computer science. State-of-the-art techniques combine low-level image processing, deep-learning methods, structured output prediction, and discrete-optimization based approaches for image segmentation.
Although significant progress has been made already, the goal of getting automated segmentations accurate enough to answer biological questions has not been reached yet. This thesis project aims at improving the state of the art by designing and learning specialized features for this interesting and challenging goal. The project will include working on following problems:
- investigation of supervoxel extraction methods suitable for EM images
- design of features to predict potential merges of adjacent supervoxels
- training of classifiers to improve merge prediction
- incorporation of these results in pre-build reconstruction pipelines and quantitative analysis on annotated EM volumes
Candidates with a background in mathematics, computer vision and good programming skills (Matlab/C++) are particularly encouraged to apply.