Unsupervised Selective Rank Fusion on Content-Based Image Retrieval

Mainly due to the evolution of technologies to store and share images, the growth of image collections have been remarkable for years. Therefore, developing effective methods to index and retrieve such extensive available visual information is indispensable. The CBIR (Content-Based Image Retrieval) systems are one of the main solutions for image retrieval tasks. These systems are mainly supported by the use of different visual descriptors and machine learning methods. Despite the relevant advances in the area, mainly driven by deep learning technologies, accurately computing the similarity between images remains a complex task in various scenarios due to the well known semantic gap problem. As distinct features produce complementary ranking results with different effectiveness performance, a promising solution consists in combining them. However, how to decide which visual features to combine is a very challenging task. This work proposes three novel methods for selecting and combining ranked lists by estimating their effectiveness in an unsupervised way. The approaches were evaluated in five different image collections and several descriptors, achieving results comparable or superior to the state-of-the-art in most of the evaluated scenarios.