.png){kind=icon}
.png){kind=icon}
Return
A Mixed Distortion Image Correction Method Based on Improved U-Net
.png){kind=icon}
PRE
AI
DOI：10.1007/s11760-025-04702-7.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Due to issues in manufacturing processes and lens assembly accuracy, digital cameras often experience nonlinear distortion when imaging. Most of the existing deep learning-based correction methods are targeted at the same type of distortion and are limited by datasets. Some correction methods have poor generalization performance. To address the above issues, we constructed a hybrid distortion dataset containing both the original image and the distorted image based on the hybrid distortion model of the camera. A new distortion correction model based on U-Net network is proposed for the hybrid distortion model, converting the distortion correction task into the problem of pixel-by-pixel displacement prediction of the image. Meanwhile, multiple weighted loss functions are designed for the two outputs. Experiments show that this model outperforms traditional algorithms and has good generalization and robustness.
Keywords:
Mixed distorted image data set
Deep Learning
Adaptive U-Net network
Multiple weighted loss function
Journal
IF:
2.1
Papers:
877
Citations:
4.6K
Organization
No organization information available