Automatic Training Image Set Creation for Object Recognition

Training sets for object recognition are of fundamental importance for image classifiers. However, this fact seems to be generally neglected in the computer vision community, which focuses primarily on the construction of descriptive features and the design of fast and effective learning mechanisms. Furthermore, collecting training sets is a very expensive step, which needs a considerable amount of manpower for selecting the most representative samples for an object class. In this thesis, we face this problem, following the very recent trend of automatizing the collection of training samples for image classification: in particular, we utilize lexical and ontology analysis to drive the search on different image search engines, considering diverse image metadata, such as textual tags. The thesis will discuss on different automatic search strategies for building training sets, discussing the first and encouraging results that we have achieved.