基于多尺度残差特征融合的图像去雾算法

Image defogging algorithm based on multi-scale residual feature fusion

针对现有去雾算法处理后图像色彩暗淡、视觉保真度差、细节特征丢失的问题,本文提出一种基于多尺度残差特征融合的图像去雾算法。首先,设计多尺度并行特征层,旨在从不同尺度下提取图像特征以提升网络的鲁棒性;然后,设计残差网络连接层,实现多个卷积层之间信息的传递和连接,提高特征的利用率,加快特征提取速度;接着,设计嵌入注意力机制的深度特征信息融合层重点关注图像关键信息,有效提高图像的清晰度,降低背景噪声干扰;最后,设计基于去雾理论及曝光融合的色彩矫正增强方法,用于解决网络去雾后图像色彩暗淡的问题。实验结果表明,所提的去雾增强算法在SOTS、OTS、RTTS公开数据集上的峰值信噪比(PSNR)、结构相似性(SSIM)、均方误差(MSE)分别达到了21.37 dB、82%、473.6,有效改善因雾霾天气造成的图像质量退化现象,性能更佳。

Aiming at the problems of dim color,poor visual fidelity and loss of detail features of the image after processing by existing defogging algorithms,this paper proposes an image defogging algorithm based on multi-scale residual feature fusion.Firstly,a multi-scale parallel feature layer is designed to extract image features from different scales to improve the robustness of the network.Then,the residual network connection layer is designed to realize the transmission and connection of information between multiple convolutional layers,improve the feature utilization rate and speed up feature extraction.The depth feature information fusion layer embedded in the attention mechanism is designed to focus on the key information of the image.It can effectively improve the clarity of the image and reduce the background noise interference.Finally,a color correction and enhancement method based on fog removal theory and exposure fusion is designed to solve the problem of dim image color after network defogging.The experimental results show that the proposed defogging enhancement algorithm achieves the peak signal-tonoise ratio(PSNR),structural similarity(SSIM)and mean square error(MSE)of 21.37 dB,82%and 473.6 on the public data sets SOTS,OTS and RTTS,respectively,which effectively improves the image quality degradation caused by foggy weather with better performance.