A new multiple instance learning algorithm based on instance-consistency

Multiple-instance learning (MIL) has been successfully utilized in image retrieval. Existing approaches cannot select positive instances correctly from positive bags, which may result in low accuracy. Inspired by the characteristic that consistencies are always among instances and instances, and bags and instances, we propose a new algorithm called multiple instance learning based on instance-consistency (MILIC) to mitigate this issue. First, we select potential positive instances effectively in every positive bag through the minimum cost of instance-consistency. Second, we use the L1-LR to select irrelevant instances from potential positive instances to further improve the retrieval efficiency. Then, we design a novel feature representation scheme based on the irrelevant potential positive instances to convert a bag into a single instance. Band on the feature representations, we finally conduct object-based image retrieval and image categorization by adopting the standard single-instance learning (SIL) strategy, such as the support vector machine (SVM), to verify the effectiveness of our proposal. © 2017 Totem Publisher, Inc. All rights reserved.