Image Deraining Network Based on Dense Hybrid Attention and Global Compensation
To address the issue of rain streaks remaining due to the poor extraction of rain line features by existing image deraining algorithms,an image deraining network based on dense hybrid attention and global compensation is proposed.Shallow features of the input rainy image are extracted through multi-layer convolution operations.By integrating the advantages of dense connections and residual networks,a dense residual attention module is designed by introducing multiple attention mechanisms to achieve feature recycling and capture multi-scale features of the image.A global compensation module is added to ensure the comprehensiveness of feature extraction.Features are reconstructed through convolutional layers to obtain a clear and rain-free image.Experimental results show that the proposed algorithm outperforms existing classical and novel algorithms,effectively removing rain streaks and enhancing the overall visual quality of the image.
Deep learningImage derainingDense hybrid attentionGlobal compensation
NETL
NSTL
万方数据
