Multi-Level Wavelet-Based Network Embedded with Edge Enhancement Information for Underwater Image Enhancement

As an image processing method, underwater enhancement (UIE) plays important role in the field of resource detection and engineering research. Currently, convolutional neural network (CNN)- Transformer-based methods are mainstream for UIE. However, CNNs usually use pooling to expand receptive field, which may lead information loss that is not conducive feature extraction analysis. At same time, edge blurring can easily occur enhanced images obtained by existing methods. To address this issue, paper proposes a framework combines CNN Transformer, employs wavelet transform inverse encoding decoding, progressively embeds on raw process. Specifically, first, features its extracted step using convolution module residual dense attention module, respectively, obtain mixed maps different resolutions. Next, structure Swin Transformer group used extract global features. Then, resulting map encoder’s hybrid high-resolution reconstruction decoder. The experimental results show proposed method achieve excellent effect protection visual images. In addition, effectiveness each component model verified ablation experiments.