通道注意网络和模糊划分熵图割的图像去雾

针对雾图成像时变化的场景光及去雾过程中不同雾相关信息在处理上的差异性,提出了通道注意网络和模糊划分熵图割的单幅图像去雾算法。以考虑变化场景光的大气散射物理成像模型为基础,首先使用通道注意的编码解码网络来估计透射率,并在编码器最后及解码器起始处添加通道注意模块,以便为编码器提取的不同雾相关特征图分配不同的权重,准确地计算透射率;然后利用所提出的模糊划分熵图割算法将透射率划分为不同场景光覆盖下的近景、中景、远景,此分割策略将考虑空间相关性的图割算法与模糊划分熵的阈值分割算法相结合,解决了单一阈值分割算法产生的区域误分问题;最后估计场景光和大气光,得到去雾图像。实验结果表明,算法在合成雾图及真实雾图上均有较好的去雾效果。与已有的去雾算法相比,本文算法在峰值信噪比及结构相似性上均有提升,单张图像的平均处理时间为3.9 s。

Image Dehazing Method Using Channel Attention Network and Fuzzy Partition Entropy with Graph Cut

Aiming at the varying scene light in imaging process and difference between the haze relevant features in image dehazing,a channel attention network and fuzzy partition entropy with graph cut for single image dehazing based on the improved atmospheric scattering model with varying scene light is proposed.Firstly,the encoder-decoder network with channel attention mechanism is utilized to estimate transmission map.Then the proposed channel attention module is applied in the end of encoder and the beginning of the decoder for assigning different weights to different haze relevant feature maps and obtaining accurate transmission map.Then,the fuzzy partition entropy combined with graph cut is used to segment the transmission map into distant scene,middle scene and close scene covered with varying scene light.This scheme combines spatial correlation and fuzzy partition entropy,solving misclassified problem introduced by the threshold-based segmentation.Finally,a clear image is obtained with the predicted transmission map,estimated scene light and atmospheric light.Extensive experiments demonstrate that this method achieves promising effective on synthetic images and real images.Comparing with exiting methods,our method improves dehazing results in both peak signal to noise ratio and structural similarity.The average running time for handling single hazy image is 3.9 s.