Results
Statistics
As we expected, due to the fact that our standard set did not include handwriting
samples, our system's performance on handwritten data was not as accurate as with computer
fonts. Below is a graph of our percent error for each number, showing the results for
both hand written numbers as well as typwritten ones.
From this graph it is evident that our two main problems were the hand written 2's
and the type written 6's. To explore this further a table was prepared for both sets, which
shows the system's interpretation of each number and is much more helpful in terms of
understanding our problems.
Results for 33 Computer Fonts
Results for 13 Handwriting Samples
The problem with a six interpreting as an eight is a result of the thinning algorithm
creating an extra hole, thus our algorithm assumes that the digit must be an 8. The most
glaring problem with 2's is simply a result of the great variations in handwriting. In
the end we achieved a success rate of %90 with our sample set of computer fonts and %78
with our handwriting samples. Again we would like to point out that the post office
occasionally loses our mail, right?!
Previous