关键词: 图像去雾, 曲线估计, 曝光增强, 注意力机制

Abstract: A single-image defogging network based on curve estimation and attention feature fusion was proposed to address the problems of poor generation quality and low contrast of most single-image defogging algorithm models based on end-to-end learning. The proposed model mainly consisted of the curve feature estimation module and the defogging module. In the curve estimation module，a codec network was designed to extract the curve feature map in the foggy image，and a simple quadratic curve was designed to enhance the exposure of the foggy image. In the defogging module，we improved the model's performance and generalization ability by introducing spatial and channel attention mechanisms to adaptively adjust the importance of different spatial locations and channels， respectively，to improve the model's perception of different locations. The experimental results showed that the proposed method performed better than other defogging algorithms on both PSNR and SSIM with reference evaluation metrics and FADE and NIQE without reference evaluation metrics.

Key words: image defogging, curve estimation, exposure enhancement, attention mechanism