基于大气物理模型的快速视觉优化去雾算法

Fast visual optimization defogging algorithm based on atmospheric physical model

针对雾霾天气条件下单幅图像降质以及现有去雾方法时间复杂度高的问题,以环境光物理模型为基础,引出快速视觉优化去雾算法。首先对单幅图像阈值分割找到天空区域,并结合二叉树模型定位精确的天空光矢量,进而采用改进的约束最小二乘法滤波细化粗略透射比率,保证其边缘细节较完整且受噪声影响小,最后利用环境光物理模型实现无雾图像的还原,并采用平均梯度、信息熵和视觉保真度等指标对图像进行评价。实验结果表明,所提算法与基于多尺度Retinex的自适应图像增强方法、基于独立分量的复原方法、快速可视化复原方法和暗原色先验复原方法对比,指标值较好且实时性强。

Aiming at the problem of single image degradation and high time complexity of exiting defogging methods under foggy weather, a fast visual optimization defogging algorithm based on atmospheric physical model was proposed. The proposed method firstly used threshold segmentation to find the sky region, and combined with binary tree algorithm to locate global atmospheric light precisely, and then adopted improved constrained least squares filter which can keep the edge detail and reduce noise to optimize original transmittance map. Finally, the fog image could be restored by atmospheric physical model, and the average gradient, information entropy and the visual information fidelity index were adopted to evaluate the image. The experimental results show that compared with the adaptive image enhancement method based on multi-scale Retinex algorithm, the image restoration based on independent component analysis, a quick visual image restoration method and the dark-channel prior de-hazing algorithm, the proposed method has good visual evaluation indexes and strong real-time processing capability.