昆明理工大学信息工程与自动化学院
TP391
School of Information Engineering and Automation,Kunming University of Science and Technology,Kunming
针对卷积神经网络进行图像去雾时存在模型复杂度高、参数量大的问题,提出一种轻量型卷积神经网络LDNet(Lightweight image Dehazing Network,LDNet)。首先,改进了大气散射模型的表达公式,直接剔除雾噪声,减小估计中间变量的累计误差。其次,设计了融入注意力机制的倒残差模块RNAM(Reverse residual Network module with Attention Mechanism,RNAM),该模块能够多尺度提取图像特征,关注图像中重要的语义信息,同时解决网络参数量大、复杂度高的问题。最后,使用L1平滑损失函数和MS-SSIM损失函数作为联合损失函数,使恢复的无雾图像与真实无雾图像之间的距离尽可能最小化。实验结果表明,所提出的算法在RESIDE合成数据集上结构相似性(SSIM)和峰值性噪比(PSNR)均优于其它对比算法,无论在合成图像还是真实场景都能取得良好的去雾效果,且该方法具有参数少、运算快的特点。
In order to solve the problems of high model complexity and large parameters in image dehazing using convolutional neural network, a light image dehazing network (LDNet) is proposed. Firstly, the expression formula of atmospheric scattering model is improved to eliminate fog noise directly and reduce the cumulative error of intermediate variable estimation. Secondly, a reverse residual network module with attention mechanism (RNAM) is designed, which can extract image features at multiple scales, pay attention to the important semantic information in the image, and solve the problems of large network parameters and high complexity. Finally, L1 smoothing loss function and MS-SSIM loss function are used as joint loss functions to minimize the distance between the restored image and the real image. The experimental results show that the proposed algorithm is superior to other contrast algorithms in terms of SSIM and PSNR on the composite data set of reside, and can achieve good dehazing effect in both synthetic images and real scenes, and the method has the characteristics of less parameters and fast operation.
