Add feature to dissociate image pairs based off of total control point count

[Neolexious]

I had a project where automatic control point creation managed to connect virtually all overlapping images together, but stitching was rendered impossible due to each image also having a high number of low quality connections with completely unrelated images.

Example 1 - low quality control points: http://i.imgur.com/weP0mRG.png
Example 2 - note that none of the images with less than six control points actually overlap: http://i.imgur.com/5igA2VZ.png
Example 3 - the best stitch I could get using built-in cp cleaning and optimization tools: http://i.imgur.com/dSm7LFW.png

By writing a python script to delete all control points between any image pairs with a lower-than-typical control point count, I managed to get it to stitch properly.

http://i.imgur.com/ycUqr30.png
http://i.imgur.com/qsHhv6r.png

[tmodes (tmodes)]

I'm not convinced that the number of control points is a good measure for the "quality" of the control points. In other projects there can also be image pairs with only 2 control points, but these can be essential to keep all images together.
So simply use the number is in general a bad idea and I will not implement it.

I see that it may be helpful in your case. But you are using very difficult images, where all building are more or less the same gray level.
So in this case it could be a solution to use the "--linearmatch" option in cpfind to prevent generating "wrong" control points in the first place instead of trying to remove the wrong one later.