Applying relevance feedback with a linear programming classifier
Bin Li · Caai Transactions on Intelligent Systems · 2007
This paper presents a method of applying relevance feedback to an automated sketch retrieval system by means of linear programming (LP) classification. A linear programming classifier was designed by combining feature selection with classification learning. The proposed classifier not only achieved real-time learning based on small sets of user-annotated samples, but also identified sensitive features from user’s interactive selections according to their contribution to the classification of candidate sketches, effectively capturing the intent of user feedback with only a small set of training samples giving relevance feedback. Experiments in sketch retrieval prove that the proposed method is both effective and efficient for relevance feedback.