Iterative Search with Local Visual Features for Computer Assisted Plant Identification

Abstract

To support computer assisted plant species identification in a realistic, uncontrolled picture-taking condition, we put forward an approach relying on local image features. It combines query by example and relevance feedback to support both the localization of potentially interesting image regions and the classification of these regions as representing or not the target species. We show that this approach is successful, and makes prior segmentation unnecessary.