Fourier Transform for object recognition with noise

I am using Fourier Transform for object recognition. I got > 90% recognition accuracy. when i add noise say salt & pepper noise to my test images, my recognition accuracy increases. However, It should decrease. Can someone please help me out on this?

[Geetika]

Any Help Please?

Answer 1

[Image Analyst]

No, but that's really fantastic! I suggest you next start to work on a perpetual motion machine.

[Geetika]

Perpetual motion machine? How is that linked with this? Any idea why is this happening?

[Image Analyst]

No I don't. How could I? You've kept your code secret, as you should , until you file your patent. An algorithm that is more accurate the more noise is added would be very very valuable . One can imagine all sorts of valuable commercial, medical, and industrial uses so I encourage you to file a patent on it as soon as possible.

[Walter Roberson]

Oddly, there are some systems that improve accuracy when noise is added.

http://perso-laris.univ-angers.fr/~chapeau/papers/perpign2.pdf

http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.104.7905&rep=rep1&type=pdf

http://www.syr.edu/news/facultyvoices/fv-varshney-01-10.html

[Geetika]

so that means the behavior is fine? and this cannot be due to some error? But for what reason this could happen?

[Walter Roberson]

You are likely not using the kind of non-linear resonance that the theory is developed about. You probably have a bug in your code on line 17 or 28.

[Geetika]

But I haven't uploaded any code??