An improved image inpainting method of generative adversarial network
By employing an image inpainting method based on generative adversarial networks, combined with U-Net and spatial attention mechanisms, the problem of poor image inpainting results for large-area missing areas is solved, achieving high-quality image inpainting effects.
Patent Information
- Application Number
- CN202310391460.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-04-13
- Publication Date
- 2026-05-29
- Estimated Expiration
- 2043-04-13
AI Technical Summary
Existing image restoration methods are ineffective when large areas are missing. Traditional methods are computationally intensive and slow, while deep learning-based methods lack global feature extraction capabilities, resulting in disordered contours in the restoration results.
An image inpainting method based on generative adversarial networks is adopted, which combines U-Net network and spatial attention mechanism. It extracts large-scale multi-scale feature information through improved dilated convolution gated residual blocks, and uses a weighted loss function of minimum absolute error loss and bulldozer distance loss for training, combined with a penalty function for regularization.
It significantly improves the image restoration effect in cases of large-area loss, enhances the global consistency and quality of the generated image, and achieves better restoration results.