Training a Neural Network to Find Image Features

Training a Neural Network to Find Image Features

Chad Abbott and Joel Cockrell





This is the initial data obtained for finding eyes. The user clicks on a desired
search pattern and a blue box is drawn around that pattern. The white boxes are drawn
whe
n the error from the search criteria and the test criteria is found within tolerable
limits.

The process
is capable of finding the target feature but also produces both Type I
(alleging the feature
—
“eye”
—
is

not

present where it is
) and Type II errors (
alle
ging the
feature
—
“eye”
—
is present where it is not
).