【摘要】 深度学习已被大量应用于图像去雾领域,并取得优异表现。但现有的网络模型往往深度较深且包含大量参数,这会降低运算效率。为了实现轻量级网络和保持良好性能,提出了多尺度轻量级图像去雾网络(Multi-scale Lightweight Network,MLNet)。使用多支路设计和多尺度卷积提取不同层次的特征信息,在网络中进行特征融合后重建残差图像。最终使用残差图像与输入图像求和得到恢复图像。实验结果表明该方法在主观感知和客观指标上均优于对比方法。更多还原

【Abstract】 Deep learning has been used extensively in image dehazing and has achieved excellent performance. However,existing network models tend to be deep and contain numerous parameters,reducing operational efficiency. To implement a lightweight net and maintain high performance,the multi-scale lightweight image dehazing network is proposed. A multi-branched design and multi-scale convolution are used to extract feature information at different scales. multi-scale features are integrated to reconstruct the residual image. Final,the recovered image is obtained by summing the residual image with the input image. The experimental results show that the method outperforms the comparison method in terms of both subjective perception and objective metrics.更多还原