感知多重特征的双分支网络实现针对真实世界文本图像的超分辨率的方法
By constructing a dual-branch network that perceives multiple features and integrates visual and textual features, the problem of insufficient feature extraction in real-world text image super-resolution is solved, achieving more efficient text recognition results.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- EAST CHINA UNIV OF SCI & TECH
- Filing Date
- 2023-06-01
- Publication Date
- 2026-07-17
AI Technical Summary
Existing text image super-resolution methods struggle to fully extract important features when processing low-resolution text images in the real world, resulting in low recognition rates. Furthermore, conventional methods fail to effectively simulate complex degradation scenarios in the real world.
A dual-branch network for perceiving multiple features is constructed, including a super-resolution branch and a text recognition branch. Visual and text features are fused through an image-image fusion module, a frequency-spatial perception module, and a text-image fusion module. Features are learned in the frequency and time domains, and training is performed using image reconstruction loss and text recognition loss.
It improves the super-resolution reconstruction effect of real-world text images, enhances the perception ability of visual features and text sequence knowledge, and improves the accuracy of text recognition.
Smart Images
Figure CN116703725B_ABST