E. Dawit and T. Sortrakul (Thailand)
: OCR, Likelihood Ratio, Feature and Architecture
Forward integrated feature and architecture selections (FIFAS) is
crucial in the design of classification systems in particular using
neural networks algorithm for any kind of training datasets.
FIFAS is easy and simple way to obtain the suitable model for a
given number of features with acceptable classification accuracy
rate without having trial and error methodology. It succumbs to
faster, reliable and less resource usage. Furthermore, it conquers
the burden of computational cost and exhaustive searching for
ideal model although it may not suitable for the proponent of
middle stage between accuracy and speed. The other peculiar
feature of FIFAS is to permit practitioner to choice which pillar
comes first for the integrated approach. This new algorithm is
tested on new benchmark (Geez characters) and common
available handwritten English numerals.