Biased discriminant analysis with feature line embedding for interactive image retrieval

Abstract

The problem of content based image retrieval is to narrow down the gap between low-level image features and high-level semantic concepts. In this paper, a biased discriminant analysis with feature line embedding (FLE-BDA) is proposed for performance enhancement in the relevance feedback scheme. We try to maximize the margin between relevant and irrelevant samples at local neighborhoods. In the reduced subspace, relevant images would be closed as possible; while irrelevant samples are far away from relevant samples. The evaluation results on dataset SIMPLIcity are given to show the performance of the proposed method.