I wanted to start on the implementation of the MSER[1] blob detection algorithm. I am leaning towards implementing the approach of the original paper[2], which has a linear time

complexity for all practical images. The alternative is a guaranteed linear time implementation[2], but the paper does not have as many citations. 

Any thoughts are welcome.

[1] : https://en.wikipedia.org/wiki/Maximally_stable_extremal_regions

[2] : https://www.robots.ox.ac.uk/~vgg/research/affine/det_eval_files/matas_bmvc2002.pdf

[3] : http://link.springer.com/chapter/10.1007%2F978-3-540-88688-4_14

Thanks

Vighnesh