| 基于多尺度优化和动态特征融合的图像去模糊研究 |
| 投稿时间：2022-12-09 修订日期：2022-12-26 点此下载全文 |
| 引用本文： |
| 摘要点击次数: 33 |
| 全文下载次数: 0 |
|
|
| 中文摘要:目前采用U-Net结构的去模糊算法存在细节损失、图像质量欠佳等问题，因此本文对U-Net进行改进，提出一种基于多尺度优化和动态特征融合的图像去模糊方法。首先针对细节损失，提出一种精简且有效的多尺度残差注意力模块，通过增加特征尺度多样性，以提取更精细的图像特征。此外，为了将更有利的特征传递到解码部分，在跳跃连接处设计动态特征融合模块，采用注意力加权的方式选择性融合不同阶段的编码特征。该算法采用多尺度内容损失和多尺度高频信息损失进行约束训练。在GoPro数据集和RealBlur数据集上的实验结果表明，该方法能有效改善图像质量，复原更丰富的细节信息。与现有去模糊算法相比，本文算法在主观视觉和客观评价等方面均具有一定优势 |
| 中文关键词:图像去模糊 特征加权 多尺度特征 U-Net结构。 |
| |
| Image Deblurring Research Based on Multi-scale Optimization and Dynamic Feature Fusion |
|
|
| Abstract:Existing U-Net based image deblurring algorithms have some problems, such as detail loss and limited feature transfer capability. Therefore, this paper aims to improve the U-Net structure and proposes an image deblurring method based on multi-scale optimization and dynamic feature fusion. To solve the detail loss, a simplified and effective multi-scale residual attention module is proposed to extract finer image information by increasing the diversity of features. In addition, in order to transfer more favorable features to the decoding part, a dynamic feature fusion module is designed at the skip connection, which can selectively fuse different stage of encoding features by attention weighted. The algorithm adopts multi-scale content loss and multi-scale high-frequency loss to supervise the training process. Numerous experiments on the GoPro and RealBlur datasets indicate that our model achieves remarkable performance. Compared with the existing deblurring algorithms, our method shows superiority in terms of subjective visual and objective evaluation. |
| keywords:image deblurring feature weighted multi-scale feature U-Net structure |
| HTML 查看/发表评论 下载PDF阅读器 |