External Query Reformulation for Text-Based Image Retrieval

摘要: In text-based image retrieval, the Incomplete Annotation Problem (IAP) can greatly degrade retrieval effectiveness. A standard method used to address this problem is pseudo relevance feedback (PRF) which updates user queries by adding terms selected automatically from top ranked documents in a prior run. PRF assumes that target collection provides enough information select effective expansion terms. This often not case since images only have short metadata annotations leading IAP. Our work proposes use of an external knowledge resource (Wikipedia) process refining queries. our method, Wikipedia strongly related query ("definition documents") are first identified title matching between and titles articles. These definition as indicators re-weight initial search run on abstract using Jaccard coefficient. The new weights combined with scores rated different indicators. Query-expansion then based these for documents. evaluated ImageCLEF WikipediaMM task document fields. results show significant improvement compared methods.