find max with sub accuracy

edit

I hate to dig up this old question, however he might still have email notification turned on.

check out the new topic. http://answers.opencv.org/question/29665/getting-subpixel-with-matchtemplate/

I have a link to my source code.

If I understand you well, you are trying to fit some arbitrary basis function to your data and want to find a (possibly local) maximum of that function.

A blunt and computationally expensive way of doing that is to select an environment of the maximum you are interested in, compute the 2D fourier transform, possibly remove some high frequencies from that to your taste, and transform back at higher resolution (scale your transform appropriately). Then, you can find the maximum of the regridded smoothed function, which may or may not have anything to do with your original data.

There are many choices of basis functions possible. Use a priori knowledge of the sampled data as much as you can to choose your basis.

In the first,Smooth your matrix by gaussian then find maximum in new matrix.

Note :This technique is used to find the local maximum in accumulator Because of the discrete nature in Which maximum votes will win by the votes of its close neighbors.