The currently popular image inpainting methods based on deep neural networks typically employ large receptive field feature extractors. However, when restoring local patterns and textures, they often generate artifacts or distorted textures, thus failing to recover the overall semantic and visual structure of the image. To address this issue, a novel image inpainting method, known as Optimized Receptive Field (ORFNet) , is introduced, which combines coarse and fine inpainting by employing an optimized receptive field strategy. Initially, a coarse inpainting result is obtained using a generative adversarial network with a large receptive field. Subsequently, a model with a small receptive field is used to refine local texture details. Finally, a global refinement inpainting is performed using an encoder-decoder network based on attention mechanisms. Validation on the CelebA, Paris StreetView, and Places2 datasets demonstrates that ORFNet outperforms existing representative inpainting methods. It leads to a 1.98 dB increase in PSNR and a 2.49% improvement in SSIM, along with an average 2.4% reduction in LPIPS. Experimental results confirm the effectiveness of the proposed image inpainting method, showcasing superior performance across various receptive field settings and achieving a more realistic and natural visual outcome.

[2023-11-15] Accepted Paper /doc/2023.09.0406/v1