How to select and customize object recognition approaches for an application?

Abstract

Recently, object recognition has been successfully implemented in a couple of multimedia content annotation and retrieval applications. The employed recognition approaches are carefully selected and adapted to the specific needs of their tasks. In this work, we propose a framework to automate the simultaneous selection and customization of the entire recognition process. This framework only requires an annotated set of sample images or videos and precisely specified task requirements to select an appropriate setup among thousands of possibilities. We use an efficient recognition infrastructure and iterative analysis strategies to make this approach practicable for real-world applications. A case study for face recognition from a single image per person demonstrates the capabilities of this holistic approach. © 2012 Springer-Verlag.