Object Recognition and Video Indexing for Media Analysis

OEGAI Journal

4-10

This short contribution provides an overview of multimedia information retrieval. In particular, we focus on object recognition and visual video indexing. We briefly review the current state of the art based on interest point extraction (e.g. Harris points, Maximally Stable Extremal Regions) and calculation of invariant descriptors around these points (e.g. SIFT, differential invariants). The main challenge for recognition from videos is the efficiency due to the huge size of the database.

We will illustrate these concepts on the recently started project, TheoMedia (joint work between TU Graz, Joanneum Research and Institute for Fundamental Theology at KFU Graz) where we use recognition and indexing technology for analyzing films in order to find religious symbols. The goal of the project is to use all the available information retrieval technology to substantially speed up the analysis process.

PDF