Laplacian pyramid adversarial network for face completion

作者：

Highlights：

• We propose an end-to-end optimization framework to address the face completion task via modeling a generative adversarial network with hybrid architecture.

• We propose to incorporate deep generative adversarial networks with a Laplacian pyramid mechanism into a unified framework that can recover the spatial information of missing face regions in a coarse-to-fine manner.

• We construct a new deep dilated convolutional residual learning architecture that can generate the high-frequency details progressively and eliminate color discrepancies to ensure visual consistency in the completed image.

摘要

•We propose an end-to-end optimization framework to address the face completion task via modeling a generative adversarial network with hybrid architecture.•We propose to incorporate deep generative adversarial networks with a Laplacian pyramid mechanism into a unified framework that can recover the spatial information of missing face regions in a coarse-to-fine manner.•We construct a new deep dilated convolutional residual learning architecture that can generate the high-frequency details progressively and eliminate color discrepancies to ensure visual consistency in the completed image.

论文关键词：Face completion,Generative adversarial network,Laplacian pyramid

论文评审过程：Received 26 April 2018, Revised 25 October 2018, Accepted 17 November 2018, Available online 5 December 2018, Version of Record 13 December 2018.

论文官网地址：https://doi.org/10.1016/j.patcog.2018.11.020