Goal of the project is to implement neuron segmentation on NeuroFinder dataset. Dataset includes 19 training sets(includes images + neuron regions) and 9 testing sets(includes only images).
We have tried three different approaches.
- Starting with a simple Naive Segmentation, where an effort to find relation on how the neuron pixels and non-neuron pixels is made.
- Non-negative Matrix Factorization method where NMF is applied on image data and then connected components which satisfy a condition are extracted as neuron segments.
- Neural Networks in which a binary classification is applied using dense neural networks. Blob detection is performed on the output of network to get desired regions of interest.
For nmf implementation, can be run using command python3 main.py. It accepts two command line inputs:
"-d", "--dataset" <path to directory that includes neurofinder.**.**.test folders>"-o", "--output" <path to directory to which output json files of each dataset are written>
For nn implmentation, can be run using python src/nn.py.
See the contributors for details.
This project is licensed under MIT License.
- Thanks to Shannon Quinn for helping in clearing doubts in neural networks.
- Thanks to Nihal Soans and Dharamendra for suggestions on neural networks and nmf.