降噪卷积自编码装置及降噪卷积自编码方法
By employing a noise-reducing convolutional autoencoder device and method, image features are processed using skip connections and multi-step convolutional layers. Combined with loss function training, it adapts to various distortion types, achieving efficient image denoising and blur removal, improving image recognition accuracy, and solving the problems of high computational cost and insufficient adaptability in existing technologies.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ACER INC
- Filing Date
- 2021-11-29
- Publication Date
- 2026-07-17
AI Technical Summary
Existing deep learning methods for image denoising require deep network training, which is computationally intensive and difficult to adapt to various types of distortion. Furthermore, the trained models can only process images with a single type of distortion.
A noise-reducing convolutional autoencoder device and method are adopted, which utilizes skip connections to pass image features, combines multi-step encoding and same-dimensional convolutional layers to perform image dimensionality reduction and enhancement, and designs a loss function based on mean square error and structural similarity index for training, which can adapt to various distortion types.
It achieves effective denoising and blur removal of images with multiple distortion types in a single training run, improves image recognition accuracy, reduces training computation and prevents blockiness, and achieves denoising performance that is the same as or better than existing technologies.
Smart Images
Figure CN116206116B_ABST