Two new bag generators with multi-instance learning for image retrieval
Wei Liu, Weidong Xu, Lihua Li, Guoliang Li · 2008
Multi-instance learning(MIL) is a new framework for learning from ambiguity, which is feasible for query-by-example(QBE) paradigm in content-based image retrieval(CBIR), since the query image posed by the user is often ambiguous and difficult to be perceived. Image bag generator, which can transform images into image bags, plays an important role in applying MIL for CBIR according to some researchers' works. In this paper, two new image bag generators named JSEG-bag and Attention-bag were proposed, respectively. JSEG-bag is based on the JSEG image segmentation algorithm and the Attention-bag is based on a saliency-based bottom-up visual attention computational model motivated by visual physiological experimental results. Preliminary experiments showed that the proposed image bag generators can achieve comparable results to some existing bag generators but are more efficient in indexing images.