There is 3 method to decompose a matrix : Polar Decomp, QR Decomp and SVD.
We can avoid to speak about SVD because it's not a good method for affine transform.
We end with Polar Decomp and QR Decomp, the difference is the coord independent on Polar Decomp.
Coord Independent mean S * R or R * S can be extracted, QR Decomp only allow S * R.
Polar Decomp has a Rotation Scaling, QR Decomp doesn't has from what I saw.
I would avoid Rotation Scaling and QR Decomp looks a good candidate.
Is it the only difference ? Is QR Decomp safe like Polar Decomp ?
Edited by Alundra, 25 May 2014 - 03:08 PM.