基于多尺度连接的区域自适应去雾算法  

Region Adaptive Dehazing Algorithm Based on Multi-Scale Connection

作　　者：韦照川[1] 王皓坤 纪元法[1] WEI Zhao-chuan;WANGH Hao-kun;JI Yuan-fa(Guangxi Key Laboratory of precision Navigation Technology and Application,Guilin University of Electronic Technology,Guilin Guangxi 541004,China)

机构地区：[1]桂林电子科技大学广西精密导航技术与应用重点实验室,广西桂林541004

出　　处：《计算机仿真》2023年第2期510-516,共7页Computer Simulation

基　　金：国家自然科学基金(61561016,61861008);广西科技重大专项基金(AC16380014,AA17202048,AA17202033);“认知无线电与信息处理”教育部重点实验室(CRK2200109);桂林电子科技大学研究生教育创新计划资助项目(2020YCXS028)。

摘　　要：针对传统单幅图像去雾算法中细节信息丢失导致去雾后物体轮廓不清晰,以及雾浓度分布不均匀使得去雾不完全等问题,提出一种基于多尺度连接的区域自适应去雾算法,设计了一个由编/解码部分以及区域非局部块构成的自适应去雾网络;首先在不同编码层中使用密集连接残差块和非局部操作提取不同尺度的特征信息。然后将来自不同编码层的特征信息聚合后,送入各级解码器中进行解码操作。最后将解码输出的特征图与原始雾图逐元素相加后恢复出清晰图像。在RESIDE数据集上进行仿真,将实验结果与较新算法进行定性和定量比较。提出算法的峰值信噪比提高了1.77dB、结构相似度提高了0.02,去雾效果得到明显提升。For the traditional single image dehazing algorithm due to the loss of detail information resulting in unclear object profile and uneven distribution of hazy density resulting in incomplete dehazing problems,a regional adaptive dehazing algorithm based on multi-scale connection is proposed,and an adaptive dehazing network composed of coding/decoding parts and regional non-local blocks is designed.Firstly,densely connected residual blocks and non-local operations were used in different coding layers to extract feature information of different scales,and then the features from different coding layers were aggregated and sent to the decoders of each layer for decoding operations.Finally,the decoded feature map and the original fog map were added element by element to restore a clear image.Experimental simulations were carried out on the RESIDE dataset,and the dehazing results were compared qualitatively and quantitatively with other state-of-the-art algorithms.The peak signal-to-noise ratio of the algorithm has increased by 1.77dB and the structure similarity is increased by 0.02 and the dehazing effect is significantly improved.

关 键 词：图像去雾 背景细节 多路连接 特征聚合 区域非局部块 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]