EPITA 2020 MLRF practice_01-90_twinit-bonus v2021-05-17_160644 by Joseph CHAZALON
This is a bonus session, you are free!
From the original full image (shown below)…
… generate segmented bubbles, in separated files, with the background masked in black, like this:
Use connected component labelling twice, after thresholding the image, first to select the biggest component in the background (and remove the bubbles and their content), then to remove the small elements within the background.
You can find useful documentation here:
You can also check the files twin_it/bubbles_labels_stats.npz
and twin_it/mask_bubbles.png
that we produced.
We did not do them, so good luck.
Plot the distribution of the distances/similarities for a given distance/similarity matrix.
Is it possible to find an appropriate normalization (per bubble?) so we can have a better global threshold? (I do not think so).
Extract the largest possible rectangle inscribed in the bubble so we have a nice patch without background.
An interesting approach: http://d3plus.org/blog/behind-the-scenes/2014/07/08/largest-rect/
#TODO