Aprendizado ativo para recuperação e classificação de imagens

Abstract

Currently, image databases have been growing, resulting in the need for optimization and acceleration of the image retrieval and classification processes, together with the improvement of the quality of the returned results. In this context, this work proposes the use of active learning strategies for image classification and retrieval, in order to select more informative samples and to minimize the interaction of the specialist during the learning process. In addition, new active learning strategies are proposed for classification and content-based image retrieval tasks. To validate the proposals, experiments were performed using datasets from different application domains. From the obtained results, it is possible to observe significant gains presented by the proposals in relation to the strategies widely used in the literature.