Perhaps you should try a different approach. It looks like the original image has a black grid over the entire image. That is probably what is generating the high score, and the text is the minor portion that doesn't match.
I would suggest filtering the image to remove the grid. You could use either an averaging filter or a median/min/max filter to remove the grid. The best results would depend on the original image.
Once you remove the grid, you should have a solid background with text on it. This will work much better with pattern matching.
If you post an original image, I could suggest ideal processing for it.
Bruce