Image Retrieval
Image retrieval is a classic problem in computer vision that is still an active area of research. In this research project, we designed a system capable of utilizing multimodal data. Specifically, we treated images as nodes in a graph, and constructed graph edges based on the dissimilarity between pairs of images. The dissimilarity score was computed as a combination of the dissimilarity in the image content, as well as textual metadata associated with the pair of images. This resulted in a retreival system that fared much better than utilizing either modality in isolation. In particular, the crux of our contribution was in learning how to combine textual and image based dissimilarity in a manner that optimizes retrieval accuracy.
Relevant Publications
Multi-label Learning with Fused Multimodal Bi-relational Graph
Jiejun Xu, Vignesh Jagadeesh, B.S. Manjunath
IEEE Transactions on Multimedia, TMI 2014
Jiejun Xu, Vignesh Jagadeesh, B.S. Manjunath
IEEE Transactions on Multimedia, TMI 2014