Recognition and analysis of objects in medieval images

Abstract

Rapid and cost effective digitization techniques have lead to the creation of large volumes of visual data in recent times. For providing convenient access to such databases, it is crucial to develop approaches and systems which search the database based on the representational content of images rather than the textual annotations associated with the images. The success of such systems depends on one key component: category level object detection in images. In this contribution, we study the problem of object detection in the application context of digitized versions of ancient manuscripts. To this end, we present a benchmark image dataset of medieval images with groundtruth information for objects such as 'crowns' in the image dataset. Such a benchmark dataset allows for a quantitative comparison of object detection algorithms in the domain of cultural heritage, as illustrated by our experiments. We describe a detection system that accurately localizes objects in the database. We utilize shape information of the objects to analyze the type-variability of the category and to manually identify various sub-categories. Finally, we report a quantitative evaluation of the automatic classification of object into various sub-categories. © 2011 Springer-Verlag Berlin Heidelberg.