PRRNet: Pixel-Region relation network for face forgery detection

作者：

Highlights：

• A novel Pixel-Region Relation Network is proposed to exploit pixelwise and region-wise relations for face forgery detection.

• A pixel-wise relation module is proposed to represent the relation between every two pixels in feature map to enhance the discriminant ability of local features.

• A region-wise relation module is proposed to detect the inconsistency between regions by fusing multiple metrics.

• We achieve new state-of-the-art results on three face forgery detection datasets.

摘要

•A novel Pixel-Region Relation Network is proposed to exploit pixelwise and region-wise relations for face forgery detection.•A pixel-wise relation module is proposed to represent the relation between every two pixels in feature map to enhance the discriminant ability of local features.•A region-wise relation module is proposed to detect the inconsistency between regions by fusing multiple metrics.•We achieve new state-of-the-art results on three face forgery detection datasets.

论文关键词：Face forgery detection,Forgery localization,Inconsistency detection,Relation learning

论文评审过程：Received 1 July 2020, Revised 8 January 2021, Accepted 1 March 2021, Available online 24 March 2021, Version of Record 1 April 2021.

论文官网地址：https://doi.org/10.1016/j.patcog.2021.107950