Hello all,

I have implemented the post-processing step of the SLIc superpixels to enforce superpixel connectivity. The algorithm is essentially the same of the original paper: "R.Achanta et al. SLIC Superpixels Compared to State-of-the-art Superpixel Methods, IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 34, num. 11, p. 2274 - 2282, May 2012.":

1. Loop through all the image
2. Get the size of a connected component (adjacent pixels with same label)
3. If the size is less than a threshold, merge it to an adjacent cluster

At the end, the generated superpixels are 4-connected (or 6 in 3 dimensions), so each label corresponds to a single connected component. This differs from the actual implementation, where a label may have multiple disconnected components. 

To be said, the processing time does not suffer. See the corresponding pull request