Ask Your Question

Revision history [back]

Are Hu Moments that sensitive?

Hello guys, as a part of my final year project I am working on a character recognition using a Raspberry Pi. Because the Pi and the camera will be installed on-board of a UAV (unmanned aerial vehicle) I have chosen to go with Hu Invariant Moments to identify the character in the image. I created a look-up table containing all 7 moments for each alphanumeric character (based on images i created containing the character ) in Arial font (no bold characters) and based on that im comparing the moments of a captured image containing a character with the moments in my look-up table.

What i realized is that bold characters have way different moments from normal characters. Therefore, i have included bold characters as well in my look-up table. By doing that i slightly increased the accuracy of my algorithm. My problem now is that i have 2 different images of the character "H" (normal, no bold characters) but the moments are different again hence it recognizes other character rather than "H".

image description <-original character stored in look-up table

image description <- character im trying to identify

Original character moments: 0.687866, 5.11891, 10.3687, 9.26865, -19.1775, -12.1325, -19.9885

Character i am trying to identify moments: 0.695746, 4.34985, 8.28186, 11.1912, -20.9336, 14.3113, -23.1491

Because the original 6th moment is -ve and the character to identify 6th moment is positive, it gives me a large error and identifies the character as another letter and not "H".

Are Hu moments that sensitive that they cannot correctly match a bold character from a normal one? Or even 2 normal characters?

Thank you.