结合残差学习和跳跃连接的图像去雾

Image Defogging Algorithm Based on Residual Learning and Jump Connection

针对雾天场景图像恢复过程中图像清晰度下降的问题,提出了一种结合残差学习和跳跃连接的图像去雾算法。使用清晰图像与对应的合成雾天图像构建残差网络,残差学习能够避免由于网络层数的不断加深而带来的梯度弥散、特征丢失等问题;跳跃连接结构极大地丰富了图像在重建去雾图像时的特征维度,并且弥补了纹理信息恢复的不足。实验结果表明,与目前经典的去雾算法相比较,文中去雾算法在合成雾天图像数据集和在自然雾天图像上,恢复的图像都具有较高的清晰度和对比度。

Aiming at the problem of image sharpness decreasing during image restoration of foggy scene,proposed is an image defogging algorithm combining residual learning and jump connection.Clear images and corresponding composite foggy images are used to construct residual network,residual learning can avoid gradient dispersion,feature loss and other problems caused by the deepening of network layers.The skip connection structure greatly enriches the feature dimension of the image in the reconstruction of defogging image and makes up for the deficiency of texture information restoration.The experimental results show that compared with the current classic fog removal algorithm,the proposed algorithm can restore high clarity and contrast in the data set of the composite fog image and the natural fog image.