A Novel Dehazing Network Based on Point-by-Point Attention Enhancement
Hongyuan Jing, Hui Zhang, Yang Yi, Kaiyan Wang, Hong Chen, Aidong Chen · 2024
Under stormy weather conditions, the image quality captured by the imaging equipment is low, showing low image contrast and low scene brightness. The image processed by the non-local algorithm has phenomena such as edge blur in dense fog areas, object edge halo, and color distortion. In order to solve this problem, a new end-to-end defogging network based on attention feature fusion technology is proposed. The proposed network uses the RepVGG-SSE module in the encoding part to enhance features, and uses a point-wise pixel-level attention module to achieve high-dimensional feature enhancement. Extensive experimental results show that this method outperforms most existing defogging methods. This method has a remarkable image defogging effect, which improves image details while reducing the impact of image color distortion.