Face Image Inpainting with Parallel Multi-scale Feature Fusion Network

Abstract

The image inpainting method based on deep learning shows good performance and has a wide range of applications in many fields. In this paper, we propose a face image inpainting method based on Parallel Multi-scale Feature Fusion Network (PMFFN) for the problems of structural distortion and unreasonable semantic information in face image inpainting. Firstly, encoder downsampling extracts image feature to capture the semantic part of the missing area of the image. Then the context information is obtained through the Parallel Multi-scale Aggregation (PMA) module to improve the consistency of the generated area and the known area. Finally, image reconstruction is completed by decoder upsampling, and the Feature Attention Fusion (FAF) module is added to the network to fuse the encoded and decoded feature information. We evaluate the model on public datasets using regular masks and irregular masks. The experimental results show that our model is able to generate vivid textures and achieve satisfactory results.