Biased subspace learning for SVM Relevance Feedback in Content-Based Image Retrieval

Support Vector Machine (SVM) based Relevance Feedback (RF) is one of the most popular techniques for Content-Based Image Retrieval (CBIR). However, it is not appropriate to directly use the SVM as a RF scheme since it treats the positive and negative feedbacks equally. Additionally, it does not take into account unlabelled samples although unlabelled samples are very helpful in constructing a good classifier. To explore solutions to these two problems, we propose a Biased Maximum Margin Analysis (BMMA) and a Semi-Supervised Biased Maximum Margin Analysis (SemiBMMA) combined with SVM RF in this paper. Extensive experiments on a large real world image database demonstrate that the proposed scheme can significantly improve the performance of the traditional SVM-based RF for CBIR.