Calculating image text orientation in scanned documents

I’m trying to calculate the text orientation in scanned documents in order to de skew them. I’m aware of some of the common methods which include using hugh transform to do the same. However I recently read an article which I can’t find.
If I recollect correctly, it basically said that find the nearest neighbours for the connected components in the image. This should connect the adjacent alphabets in the words, connect the centroid of those components and build a histogram of those vectors. The histogram should give you the skew angle of the vectors.
Does anybody know such an approach for identifying document skew?
Any other approaches are appreciated as well.
Thank you.