We are developing a website/smartphone application
where users can deposit pictures of plants. Volunteers can then
answer questions relating to these pictures: text questions, or
drawing comparisons, or comparisons with other pictures. When
sufficient data are available (50 or 100 of such answers) we can
then propose an identification. The volunteers give answers based
on “naive” observations of plant details. Example on a flower: do
you see 3 petals, 4 petals, 5-8 petals, many petals ? For leaves,
we display a small set of drawings, the set changes when the firsts
questions provide sufficient information. To finish the identification,
we allow for direct comparison with other pictures of similar detail.
The questions are inside Iframe, and so are easily included on a
network of participating web sites. Depending on the total number of
visits to these sites, we shall have identifications within a few minutes
or a few hours. We have now a set of 300 plants which are well-know
in France. An important point about the method is that the volunteers
help to classify the pictures we have as well as new pictures, so that
the system is bootstrapping and scales easily. Moreover, we can
easily experiment with new questions or drawings, to find those that
reduce entropy most rapidly.
