Adaptive Methods for Structuring Multimedia Information
Otto-von-Guericke Universität, Magdeburg, Germany, (November 2005)Beschreibung
Current multimedia retrieval systems help users to manage big collections of multimedia documents. Specialized systems for text, image, audio, and video content are widely used. This work proposes a clustering approach for structuring and visualizing video documents. The system assists the user in browsing video content based on similarity of shots. It is based on a hexagonal self-organizing map which has already been applied to image documents. The map automatically adapts its structure to the feedback given by the user. At first, the video document is temporally segmented into shots by means of a recent shot boundary detection method. Next, keyframes are selected. Significant features are extracted from each keyframe, represented in a numerical format and used to train the self-organizing map. The visualization is based on a multi-level keyframe representation of shots. Diverse browsing functionalities have been implemented which include both content-based and temporal browsing.