The overall objective of this paper is to present a methodology for reducing the human workload through adapting an automatic scheme for Content-Based Image Retrieval (CBIR) engines. The proposed system utilizes an unsupervised hierarchical clustering algorithm, known as the Directed Self-Organizing Tree Map (DSOTM) that aims to closely mimic the process of information classification thought to be at work in the human brain [1, 2]. To further refine the search process and increase retrieval accuracy, a Semi-automatic relevance feedback approach is presented in this work. The Semi-automatic scheme refers to a relevance feedback CBIR engine, structured around the DSOTM algorithm. This system aims to learn from and adapt to different users' subjectivity under the guidance of an additional objective verdict provided by the DSOTM. Comprehensive comparisons with the Rank-based, relevance feedback, and automatic CBIR engines, demonstrate feasibility of adapting the Semi-automatic approach.
Human-Controlled Vs. Semi-automatic Content-Based Image Retrieval
01.04.2007
5208462 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Efficient Content-Based Image Retrieval Using Automatic Feature Selection
British Library Conference Proceedings | 1995
|Localized Content-Based Image Retrieval Using Semi-Supervised Multiple Instance Learning
Springer Verlag | 2007
|Here's Waldo: Content Based Image Retrieval
British Library Online Contents | 1998
|A content-based image retrieval system
British Library Online Contents | 1998
|