Multi-instance Learning Using Information Entropy Theory for Image Retrieval

As a new learning framework, Multi-Instance learning is used successfully in vision classification and labeled recently. In this paper, a novel Multi-instance bag generating method is put forward on the basis of a Gaussian Mixed Model. The generated GMM model composes not only color but also the locally stable unchangeable components. It is called MI bag by researchers. Besides this, another method which is called Agglomerative Information Bottleneck clustering is ad opted here to replace the MIL problem with the help of single-instance learning ones. Meanwhile, single-instance classifiers are employed here for classification. Finally, ensemble learning is employed to strengthen classifiers´ generalization ability of RBM (Restricted Boltzmann Machine) as the base classifier. On the basis of large-scale datasets, this method is tested and the result of it shows that our method provides higher accuracy and performance for image annotation, feature matching and example-based object-classification.