Batch Mode Active Learning for Interactive Image Retrieval

In content-based image retrieval, relevance feedback is an effective approach to narrow the semantic gap between low-level feature and high-level semantic interpretation by using user's feedback to judge the relevance images in the retrieval process. One important issue of RF-algorithms is how to efficiently and effectively select the helpful unlabelled samples for labelling so that the retrieval performance can be improved most efficiently. In this paper, we propose a batch mode active learning scheme for informative sample selection in interactive image retrieval. Firstly, a decision boundary is learned via Support Vector Machine (SVM) to filter the images within database. Then, a new selection criterion is defined by considering both the scores of SVM function and similarity measures between the query and the images in the database. By using this new selection criterion, the proposed scheme could obtain the most informative and representative samples for improving the efficiency of SVM active learning, thus the retrieval performance is improved significantly. The experimental results on publicly available datasets show that the proposed scheme is more effective than the traditional approaches.