Redes convolucionales siamesas y tripletas para la recuperación de imágenes similares en contenido

The objective of this paper was the development of a content-based image retrieval system, using siamese and triplet convolutional neural networks. These networks were used to generate visual descriptors, extracting semantic information from two images (siamese) or three images (triplet) at the same time. Then, a similarity learning was done, encoding these two or three visual descriptors. In the proposed scheme the storage of descriptors is not required. The experimental results show that the schemes based on convolutional neural networks extract more semantic information. The siamese and triplet architectures, apart from extracting semantic information, improved the image retrieval rate. It is concluded that the proposed scheme solved three of the main challenges in these systems, such as, semantic gap, similarity learning and storage space, which have not been solved in the previous works.