Método de Aprendizado Não Supervisionado Baseado no Produto Cartesiano de Rankings para Busca de Imagens
Resumo
Despite the consistent development of visual features, effectively measuring the similarity among images remains a challenging problem in Contentbased Image Retrieval (CBIR) systems. In this scenario, similarity learning approaches capable of improving the effectiveness of retrieval in an unsupervised way are indispensable. This undergraduate research work presents a method called Cartesian Product of Ranking References (CPRR) which was developed with this objective. An extensive experimental evaluation was conducted considering various aspects, four public image datasets, and several features. Besides effectiveness, experiments were also conducted to assess the efficiency, considering parallel and heterogeneous computing on CPU and GPU devices. The method has achieved significant effectiveness gains, including competitive state-of-the-art results on popular benchmarks.
