Learning an image manifold for retrieval
Proceedings of the 12th annual ACM international conference on Multimedia
Content-based multimedia information retrieval: State of the art and challenges
ACM Transactions on Multimedia Computing, Communications, and Applications (TOMCCAP)
PCM'10 Proceedings of the 11th Pacific Rim conference on Advances in multimedia information processing: Part I
Measuring multi-modality similarities via subspace learning for cross-media retrieval
PCM'06 Proceedings of the 7th Pacific Rim conference on Advances in Multimedia Information Processing
A Multimedia Retrieval Framework Based on Semi-Supervised Ranking and Relevance Feedback
IEEE Transactions on Pattern Analysis and Machine Intelligence
Mining Semantic Correlation of Heterogeneous Multimedia Data for Cross-Media Retrieval
IEEE Transactions on Multimedia
IEEE Transactions on Multimedia
Effective Image Retrieval Based on Hidden Concept Discovery in Image Database
IEEE Transactions on Image Processing
Content-Based Multimedia Retrieval Using Feature Correlation Clustering and Fusion
International Journal of Multimedia Data Engineering & Management
Hi-index | 0.00 |
Cross-media learning is a new hot topic in multimedia content analysis and retrieval. Because multimedia data of different modalities are heterogeneous in feature space and there exists the well-know semantic gap, one of the most challenging issues for cross-media learning is to mine underlying semantics and estimate cross-media correlation. In this paper we propose a cross-media semantics mining approach based on Sparse Canonical Correlation Analysis and relevance feedback. First, we analyze sparse canonical correlation between low-level feature matrices of different modalities in training stage, and construct a Multimodal Sparse Subspace where both canonical correlation and most meaningful features are preserved; then based on geometric distance in the subspace we estimate cross-media correlation and enable cross-media retrieval; also we provide long-term relevance feedback strategy for performance optimization. Our approach is tested with general multimedia data, including image, audio and text. Experiment and comparison results are encouraging and show that the performance of our approach is effective.