雾霾天气下可见光图像场景再现

Haze Degraded Image Scene Rendition

为了再现雾霾天气下可见光图像的清晰场景,有效抑制雾霾退化造成的对比度、清晰度下降,提出了单色大气散射模型新的求解方法.首先,将单色大气散射模型类比Retinex模型,重新解释了大气传递图;依据大气传递图的先验知识和几点假设,建立目标函数的变分模型,将大气传递图的估计问题转化为二次规划问题.通过带约束的归一化最速下降法获取最优解,并采用多分辨率技术加速计算;在HSI空间的亮度分量上反解单色大气散射模型,得到反射图像,并依据大气传递图自适应校正饱和度分量.实验结果表明,新算法可有效去除雾霾,再现真实场景的对比度和清晰度,同现有去雾算法相比,本文算法取得了相似甚至更好的去雾效果.

In order to get a clear view of the image taken under bad weather, a novel monochrome atmospheric scattering model （MASM） is proposed, which is to restrain the decline of contrast and definition caused by haze. Firstly, the MASM is reconstructed to compare with the retinex model, and this leads to our new explanation about atmospheric transmission map. Then a variational model for atmospheric transmission map is proposed, thus the transmission map estimation problem can be formulated as a quadratic programming optimization problem. The projected normalized steepest descent method is adopted to get the optimized solution, and the multi-resolution algorithm is adopted to achieve a high computational efficiency. Finally, the reflectance image on lightness component in the HSI space in computed through solving MASM and the saturation component is adjusted according to transmission map adaptively. Experimental results yield that the proposed method could remove haze effectively and playback the realistic scene＇s contrast and definition, and has some advantages over some existing algorithms.