Fusion of semantic and visual information for automatic indexing and annotation of key frame images

Modern database structures consisting of audio visual and low-level semantic data are in line with the needs and requirements of end user customers. It is difficult to expect the end user to provide any high level input rather than a very general key word annotation of the recorded data. A typical user scenario appearing in modern multimedia deals with having just audio-visual data with no additional knowledge. How to effectively annotate and enable efficient indexing and categorisation that would subsequently provide easy browsing and retrieval of desired content is the problem addressed in this paper.