# Combine Noisy predictions from an optical character recognition program

I am trying to perform optical character recognition on a field of text from different angles as the camera passes over it and beyond it. Due to the 3D skewing of the image, thee readings of the OCR are as of below:

GIS59ppnnnnu
Gsd5ldjjjkk
lISX99a\bnlsd
GISX1saa
**GISX1450**
GISX1450
GISX145na
Grtuoolbd


My primary question is if there is a method for reducing the noise and converging on a definite text from the data above (GISX1450 is the correct reading). An counter of the most common text could work, but I am looking for something that uses probability and makes a better guess.

An option here might be candidate pruning through fuzzy logic, but I don't know how to approach this either.