Integrating Descriptions to Characterize Multimedia Collections

In this paper, an innovative approach is proposed for integration of descriptions regarding multimedia content collections. The proposed method has been applied to audio– visual content clusterings, which were extracted using different algorithms. With the support of low level features, obtained with further algorithms, all input clusterings are characterized and then merged in a new integrated clustering. The proposed method performs the integration taking into account the relative cluster size (granularity) and the element relationships among clusters. The experimental results show that the initial information of input clusterings is preserved in the resulting integrated clustering, both in terms of distribution and semantics. They also show that the amount of information available in the resulting clustering increases in terms of low level features.