Blind single image super-resolution with a mixture of deep networks

作者：

Highlights：

• A new approach for blind image super-resolution via mixture of deep networks.

• An encoder network to model the blur kernel with a discrete latent variable.

• A pre-training method to properly initialize the encoder network.

• A lower bound of the likelihood function for joint training the whole model.

摘要

•A new approach for blind image super-resolution via mixture of deep networks.•An encoder network to model the blur kernel with a discrete latent variable.•A pre-training method to properly initialize the encoder network.•A lower bound of the likelihood function for joint training the whole model.

论文关键词：Blind super-resolution,Mixture of networks,Blur kernels,Lower bound,Latent variables

论文评审过程：Received 13 February 2019, Revised 28 November 2019, Accepted 15 December 2019, Available online 4 February 2020, Version of Record 8 February 2020.

论文官网地址：https://doi.org/10.1016/j.patcog.2019.107169