带水印图像复原方法及其装置、设备、介质、产品
By combining image segmentation and refining models, watermark images and transparency masks are extracted, and multi-level convolution operations and smooth synthesis are performed to solve the problem of difficult removal of light watermarks, achieving high-quality image restoration effects, which are suitable for large-scale image processing such as e-commerce platforms.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- GUANGZHOU HUANJU SHIDAI INFORMATION TECH CO LTD
- Filing Date
- 2022-12-01
- Publication Date
- 2026-07-17
AI Technical Summary
Existing technologies struggle to effectively remove visually perceptible light watermarks, resulting in unsatisfactory image restoration effects, especially when the watermark occupies a large space, leading to severe loss of background information.
An image segmentation model is used to extract watermark images, transparency masks, and redundant information feature maps. Multi-channel feature maps and image refinement models are used to perform multi-level convolution operations, and watermark masks are combined for smooth synthesis to achieve fine restoration of the depth-restored image.
It achieves high-quality image restoration, naturally and realistically restoring the content of the original image that overlaps with the watermark, and is suitable for large-scale image processing.
Smart Images
Figure CN116823630B_ABST