Study for Cross-Media Retrieval Text Features Relationship

摘要

Construct cross-media feature relationship properly is significant to query by song (QBS). In this paper the cross-media features relationship include three parts: text-based music features, text-based lyrics features and text-based picture features which relate with music. We analysis what are the associated relationship among musk, text-based lyrks and pictures. Then we design a cross-media text-indexing structure and propose a new associate approach that named "word linking matching". At the same times we improve the existing construction methods for these feature library in order to satisfactory the need of cross-media information retrieval. Pilot experiment results show the feasibility of our study work for Our QBS system.