Evaluation of an Image and Music Indexing Prototype

Abstract

This paper describes a technical solution for automated semantic indexing of music and images for a media archive environment. The indexing is based on a multi-modal low-level feature extraction and semantic high-level feature classification such as mood, genre, daytime or visual scene types. The classification on both, the audio and the visual information is based on a generic machine learning core architecture. A combination and cleansing process validates for improving the classification results. This paper presents the technical realization of a prototype and its corresponding evaluation. Finally, the practical relevance of this technology results, based on the findings of the evaluation is discussed.