摘要
为了解决雾天可见光图像降质问题,提出一种简单、高效的去雾算法。为充分利用可见光图像的色彩信息和近红外图像的细节信息,首先,根据暗通道估算出可见光图像中雾的浓度,根据雾浓度对可见光图像进行分区;然后,分别对可见光和近红外图像进行平稳小波分解,结合雾浓度分区和脉冲耦合神经网络(pulse coupled neural network,PCNN)分别融合可见光与近红外图像的高频分量和低频分量,复原得到一幅清晰而不失真的图像;最后,引入引导图像滤波对融合图像做滤波处理,平滑分区边缘的同时保留源图像的边缘信息。为验证算法的有效性,与当前主流去雾算法进行对比实验,对比指标包括去雾图像的信息熵、均值、标准差,以及算法运行时间。实验结果表明,在相同图像分辨率条件下,本文算法去雾后图像视觉效果更加理想,同时,无雾区域能够很好地保持色彩信息,反映图像细节和清晰化的各项指标优于其他算法,而且算法处理时间显著降低。
In order to address the problem of visible image degradation caused by the hazy weather conditions,a dehazing algorithm was proposed,in which the color information of the visible image and details information of the near-infrared image were fully taken advantage of.Firstly,the haze density of visible image was estimated according to the information of dark channel,based on which the visible image was par- titioned.Then the visible image and near-infrared image were decomposed by stationary wavelet transform.By using haze density partitioning and pulse coupled neural network,the high-frequency component and low-frequency component in visible and near-infrared images werefused,and a clear and high-fidelity image was obtained.Afterwards,the composited image was filtered by a guidance filter to smooth the boundaries of parti- tioned areas and preserve the edge information of source image.To validate the effectiveness of the proposed algorithm,groups of experiments were conducted to compare it and other state-of-the-art dehazing algorithms.The comparison indexes include information entropy,mean value and standard deviation of dehazed image as well as computation time of algorithms.The results showed that the proposed algorithm achieved a better visual effect,and the color information in haze-free areas was retained.Besides,all the comparison indexes related to image detail and image clar- ity were superior to that of other algorithms.Meanwhile,the computation time cost of the proposed algorithmwas significantly decreased.
出处
《工程科学与技术》
EI
CAS
CSCD
北大核心
2018年第2期99-104,共6页
Advanced Engineering Sciences
基金
国家自然科学基金资助项目(61403065)
四川大学引进人才科研启动基金资助项目(2082204194074)
关键词
图像去雾
近红外图像
图像融合
暗通道
haze removal
near-infrared image
image fusion
dark channel
