I'm writing an algorithm that uses template matching function provided by OpenCV. I'm interested in normalized cross-correlation (NCC) method.
I noticed that the performance of the OpenCV algorithm is not so good as I expected.
I read "OpenCV performance on template matching" on stackoverflow (http://stackoverflow.com/questions/7139606/opencv-performance-on-template-matching), in particular the Chris's answer that is very deep and interesting.
I also read a lot of paper about new techniques that perform NCC template matching faster than FFT/DFT.
Is in your opinion the machTemplate function implemented in OpenCV library the best implementation in terms of performance?
What is in your opinion the best paper that I have to use to re-implement a the template mathing function in a faster way? (or do you know if it is available a faster implementation?)
I tested also pyramidal approaches, but I'm interested in template matching without resize.