基于上下文信息聚合的两阶段去雾算法

Two-stage Dehazing Algorithm Based on Context Information Aggregation

针对传统去雾算法在去雾过程中存在细节丢失问题,提出了一种基于上下文信息聚合的两阶段图像去雾算法。算法包括去雾和细节恢复两个部分,整体网络由雾残差子网和细节恢复子网组成。雾残差子网结合挤压激励操作和残差网络,融入通道信息更高效地提取特征;细节恢复子网添加上下文细节聚合块,具有更大的接收域,能充分利用空间上下文细节信息更好地恢复图像细节。实验结果表明,算法保留了清晰的纹理细节信息,在主观评价和客观评价方面均表现良好。

Aiming at the problem of detail loss in traditional dehazing algorithm,a two-stage image dehazing algorithm based on context information aggregation was proposed.The algorithm is consisted by dehazing and detail recovery.The whole network is composed of haze residual subnet and detail recovery subnet.The haze residual subnet is applied by the squeezing excitation operation and residual network,and integrates the channel information to extract features more efficiently.The context detail aggregation block is added in the detail recovery subnet,which has a larger receiving domain and can make full use of spatial context details to better recover details.Experimental results show that the algorithm retains clear texture details and performs well in both subjective and objective evaluation.