Font recognition was poor, due to forcing a 1st and 2nd choice at
a character level, when the total score for the correct font is often
correct at the word level, so allowed the propagation of a full set
of fonts and scores to the word recognizer, which can now decide word
level fonts using the scores instead of simple votes.
Change precipitated a cleanup of output data structures for classifier
results, eliminating ScoredClass and INT_RESULT_STRUCT, with a few
extra elements going in UnicharRating, and using that wherever possible.
That added the extra complexity of 1-rating due to a flip between 0 is
good and 0 is bad for the internal classifier scores before they are
converted to rating and certainty.
During our initial chopping for each word, pay attention to whether a
dangerous ambiguity (like rn <-> m) would lead us to a dictionary word.
If so, make sure that blob gets chopped so that we can evaluate said
dictionary word during the segmentation search.
Large accuracy improvement, especially on English printed books (~9%).
git-svn-id: https://tesseract-ocr.googlecode.com/svn/trunk@713 d0cd1f9f-072b-0410-8dd7-cf729c803f20
bad idea -- a VIABLE_CHOICE's Blob->NumChunks can be modified as we go
by a call from Dict::LogNewSplit(). Relying on the auxilury
segmentation_state makes alt choices sometimes reference the wrong
blobs.
git-svn-id: https://tesseract-ocr.googlecode.com/svn/trunk@711 d0cd1f9f-072b-0410-8dd7-cf729c803f20
