I am totally puzzled with how the confidence reported at Word level relates to the confidences assigned to the characters of the same word.
I used the attached TIFF image to recognize a simple MICR line of a check.
The recognized text had two words:
495096 700000b01b205xX0eL00007010717
The confidence percentiles for the words were 59% and 38% respectively.
The confidence percentiles for the characters of the first word were (rounded):
4 97%
9 99%
5 100%
0 100%
9 99%
6 96%
I would like to know how with such high confidence scores for individual characters, one can compute the word level confidence at 59%.
I ran this test using fast training data for English with no training of my own. I am not worried about the accuracy, just curious about how to interpret confidence scores.
Thanks!