摘要
夜间图像光照不均匀,存在色偏,去雾难度较大。目前图像去雾算法主要针对白天场景,有关夜间图像去雾算法的研究较少。基于结构-纹理分层模型提出新的夜间图像去雾算法,将夜间有雾图像分解为结构层和纹理层。在结构层采用中值滤波器估计环境光,利用加权范数L1正则化模型对其进行优化,并进行去雾和颜色校正处理;在纹理层利用离散余弦变换系数估计透射率。最终融合纹理层与去雾后的结构层得到去雾图像。实验结果表明,采用该算法对夜间图像去雾后图像细节清晰,颜色自然,去雾效果显著。
The non-uniform illumination and color deviation lead to the difficulty in haze removal for nighttime image. The current image dehazing methods are mostly designed for daytime images. There are few studies on nighttime image dehazing. Therefore, we propose a new nighttime image dehazing method based on the structure- texture image decomposition model. Firstly, the haze image is divided into a structure layer and a texture layer. Secondly, to estimate and then optimize the initial atmospheric light, the median filter and the weighted norm L1 regularization model are introduced in the structure layer. After that, dehazing and color correction are performed. Thirdly, the transmittance is estimated with discrete cosine transform coefficients in the texture layer. Finally, the ultimate haze-free image is recomposed with the texture layer and the haze-free structure layer. The experimental results show that the proposed algorithm is effective in the nighttime haze image processing, generating haze-free images with clear details and natural colors.
作者
杨爱萍
王南
Yang Aiping;Wang Nan(School of Electrical and Information Engineering, Tianjin University, Tianjin 300072, Chin)
出处
《激光与光电子学进展》
CSCD
北大核心
2018年第6期95-102,共8页
Laser & Optoelectronics Progress
基金
国家自然科学基金(61472274)
关键词
图像处理
夜间图像去雾
结构-纹理分层
加权范数L1正则化模型
离散余弦变换系数
image processing
nighttime image dehazing
structure-texture image decomposition
weighted norm L1 regularization model
discrete cosine transform coefficients