A novel active learning method for content based remote sensing image retrieval
Begüm Demir, Lorenzo Bruzzone · 2015
This paper presents a novel active learning (AL) method for retrieving remote sensing images from large archives. The proposed AL method defines an effective set of relevant and irrelevant images with regard to a query image by jointly evaluating three criteria: i) uncertainty, ii) diversity and iii) density of images in the archive. The proposed AL method assesses jointly the three criteria based on two consecutive steps. In the first step the most uncertain images are selected from the archive based on margin sampling strategy. In the second step the images that are both diverse to each other and associated to high density regions of the image feature space in the archive are chosen from the most uncertain images. This step is achieved by a novel clustering based strategy. Experimental results show the effectiveness of the proposed AL method.