Filipe Coelho and Cristina Ribeiro
9th International Workshop on Content-based Multimedia Indexing (CBMI), June 2011, Madrid, Spain
Journalists and bloggers need to find useful images to illustrate news stories and blog entries with high quality photos. The dpikt text illustration system uses multimedia information retrieval to assist this content enrichment task. Users query the system with text fragments and get collections of candidate photos. Images in the results can be visually sorted according to a selected photo, or be used as a seed for interactive searches over the entire collection. Dpikt incorporates a recent visual descriptor, the Joint Composite Descriptor, and an approximate indexing scheme designed for large-scale image collections, the Permutation-Prefix Index. We have used the SAPO-Labs large-scale news stories photo collection, containing almost two million high quality photos with short descriptions, as the resource for the illustration task.