Reconstrução de imagens em super-resolução usando redes neurais convolucionais
Eduardo P. L. Jaqueira, Felippe Durán V. G. Santos, Renato Candido, Magno T. M. Silva. Reconstrução de imagens em super-resolução usando redes neurais convolucionais. Simpósio Brasileiro de Telecomunicações – SBrT’2023, 2023, São José dos Campos, Anais do SBrT 2023, 2023, pp. 1–2. (in portuguese)
Two residual architectures based on convolutional neural networks are proposed to increase the resolution of grayscale images. We consider a function based on the structural similarity index as a cost function. Through simulations, we observe that the proposed solutions lead to results superior to those obtained with the bicubic interpolation.osed solution obtained an accuracy improvement up to 14% compared to the solution using energy-based arbitration.
Super-resolution, convolutional neural network, bicubic interpolation, structural similarity index.