F20 preprocessing removing outliers from raw images

[DmytroKushnirSoftserveinc]

Moved here implementations of two filtering methods for filtering registration artifacts:

1. Take only the blob with the biggest border-length
2. Use the k-means to cluster background and noise to separatable clusters using color information.

Also incorporated filtering stage into the whole reconstruction pipeline.
As a result, we have stable reconstruction results as before outliers caused the failure of matching image parts.