Trying to extract all cells from honeycomb pictures:
- simple_thresholding relying on Otsu
- local_thresholding local histogram generalization before Otsu
- simple_gradient thresholded labels from image gradient
- double_gradient consolidated segmentation based on 2 thresholded sets of labels from image gradient
- refine_segmentation initial segmentation gets iteratively refined by using the honeycomb pattern
Under test I'm applying the best method on new images.
The resulting segmentation was quite slow on large images. My efforts to optimize the algorithm are under optimization.
I noticed how different the extracted segments in new images look compared to broodmapper images. Therefore, I decided to manually label images for optimizing the cell classifier. Under test/segments/ I extracted and labeled honeycomb segments. They are in turn used by the classifier.