Restauração de imagens coloridas usando uma rede neural convolucional residual

Guilherme Lizarzaburu, Renato Candido, Magno T. M. Silva

DOI: 10.14209/SBRT.2020.1570649730

Evento: XXXVIII Simpósio Brasileiro de Telecomunicações e Processamento de Sinais (SBrT2020)

Keywords: Rede neural convolucional residual Restauração de imagens Função de espalhamento de ponto gaussiana Índice de similaridade estrutural

Abstract

Image restoration aims to attenuate distortions caused in the acquisition process. In this paper, we use a residual convolutional neural network to restore colored images which were degraded by a Gaussian point spread function. As cost function, we consider the mean square error, and a function based on the structural similarity (SSIM) index. By means of simulations, we verify that the best restoration results were achieved with the function based on SSIM.

Download