全局参数自动估计的彩色图像偏振去雾方法

雾天环境下由于大气粒子对光线的散射作用导致成像质量下降, 针对雾霾等天气下图像退化问题, 提出了一种全局参数自动估计的彩色图像偏振去雾方法。利用不同角度的3幅偏振图像, 自动估算无穷远处的大气光和大气光的偏振度, 根据大气散射模型得到去雾后的图像。从RGB 3个色彩通道分别计算相应的参数, 使得算法适用于彩色领域。首先使用暗通道方法估计无穷远处的大气光和传输图, 并通过导向滤波对传输图优化; 然后基于大气光和目标光的不相关性, 采用全局搜索的方法估计大气光的偏振度; 最后根据大气散射模型恢复出清晰目标图像, 并利用对数变换进行增强。本文方法在雾霾天气下能够得到清晰的去雾图像, 且在浓雾天气下, 去雾图像的信息熵提升了约21%, 平均梯度提升了约2倍多, 标准差提升了约12%。实验结果表明, 本文方法较好地解决了人工取景估计参数不佳的问题, 提高了复原目标图像的清晰度和对比度, 可以用于彩色图像的目标探测与识别。

Polarization defogging method for color image based on automatic estimation of global parameters

In the foggy environment, the image quality is degraded due to scattering of light by atmospheric particles. For the image degradation under haze and other weather conditions, a global image polarization defogging method based on automatic parameter estimation was proposed.Using three polarization images of different angles, the degree of polarization of atmospheric light and atmospheric light at infinity was automatically estimated, and the image after defogging was obtained based on the atmospheric scattering model. The corresponding parameters were calculated from the three RGB color channels, making the algorithm suitable for the color field. Firstly, the dark channel method was used to estimate atmospheric light and transmission at infinity, and the transmission map was optimized by guided filtering. Then the global search method was used to estimate the degree of polarization of atmospheric light based on the non-correlation between atmospheric light and target light.Finally, clear target images were recovered from the atmospheric scattering model and enhanced using logarithmic transformation.This method can get clear defogging images under hazy weather, and in thick fog weather, the information entropy of defogged images is increased by about 21%, the average gradient is increased by about 2 times, and the standard deviation is increased by about 12%. Experimental results show that the proposed method can solve the problem of poor estimation parameters of artificial framing, improve the sharpness and contrast of the restored target image, and can be used for target detection and recognition of color images.