A Pseudo-Labeling Framework for Content-based Image Retrieval

Abstract

This paper presents a new pseudo-label fuzzy support vector machine (PLFSVM)-based active learning framework in interactive content-based image retrieval (CBIR) systems. One of the main issues associated with relevance feedback in CBIR systems is the small sample problem where only a limited number of labeled samples are available for learning. This is because image labeling is time consuming and users are often reluctant to label too many images for feedback. Learning from insufficient training samples often constrains the retrieval performance. To address this problem, we propose a new algorithm based on the concept of pseudo-labeling. It incorporates carefully selected unlabeled images to enlarge the training data set and assigns proper pseudo-labels to them. Further, some fuzzy rules are utilized to automatically estimate class membership of the pseudo-labeled images. Fuzzy support vector machine (FSVM) is designed to take into account the fuzzy nature of some training samples during its training. In order to exploit the advantages of pseudo-labeling, active learning and the structure of FSVM, we develop a unified framework to perform content-based image retrieval. Experimental results based on a database of 10,000 images demonstrate the effectiveness of the proposed method