otbHarrisImageFilter applies final smoothing only over the 0-th direction; + wrong final scaling applied
Mantis Issue 273, reported by peterl, assigned to abricier, created: 2011-04-14
After computing the Hessian of an image, smoothing with m_SigmaI is applied. However the smoothing is applied only in one direction.
Moreover, I believe the final scaling should be sigmaD^4, since during the determinant computation the scale factor gets multiplied.
1303221214 - aurelienThe smoothing is now performed on both directions ( http://hg.orfeo-toolbox.org/OTB/rev/51aa5a812c8d ). Regarding the scaling, can you add a reference? SigmaD is used as described in the filter documentation.
1303226692 - peterLI checked the changes. And sorry I didn't describe well the scaling problem before...
line 1.15 is correct - sigmaD^2 (as required by the scale-space normalization) the changed line 1.16 is thus wrong
the scaling problem is only for the final number, i.e. the Harris cornerness; because of the scaling is applied after the cornerness computation, it should be sigmaD^4 since during the cornerness calculation a determinant and square of the trace is computed, it only squares the proper normalization factor => (sigmaD^2)^2 should be applied in the m_MultiplyScalarFilter
1303228698 - aurelienThank you for the feedback! The line 1.16 is wrong and was reverted just after the designated commit (http://hg.orfeo-toolbox.org/OTB/rev/f915acec9ceb). The