.png){kind=icon}
.png){kind=icon}
Return
Wavelet-based learning and optimized sampling for image deraining
.png){kind=icon}
PRE
AI
DOI：10.1016/j.patcog.2025.112782.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• We propose WLOS, a novel single-image deraining model based on convolutional neural networks. Extensive experiments on multiple benchmark datasets demonstrate its effectiveness and competitiveness, even surpassing some Transformer-based approaches. • We design the Wavelet Kernel Learning module (WKL), which employs learnable filters to partition the feature maps into four sub-bands of distinct properties, and then applies differently shaped convolutions to each sub-band. This design effectively tackles the complexity of rain streak degradation. • To address the loss of important high-frequency information incurred by conventional downsampling, we introduce the Optimized DownSampling module (ODS). By capturing intricate rain streak patterns through a fine-grained learning scheme in the downsampling phase, ODS further boosts overall model performance.