一种光学遥感图像的快速降噪去霾算法

An image noise reduction and haze removal algorithm based on multi-frame merge

针对光学遥感成像结果受噪声和雾霾影响而劣化的问题,提出了一种结合多帧融合降噪和暗通道先验法去霾的快速降噪去霾算法.针对星载计算机主频较低,算力有限的特点,本文在降噪阶段,将传统多帧融合算法的逐像素配准改为两级配准,在全局和局部分别使用绝对误差和(SAD)与相位相关配准;提出了一种基于配准结果的配准质量评价新方法,减少了评价时间;在去霾阶段,用高斯核与透射率模版卷积,替代了传统的精细化方法.实验结果表明,采用本文提出的算法可以使图像平均峰值信噪比(PSNR)提升8.97 dB,通过计算量优化,去霾耗时减少了73.41%.

Aiming at the problem that the results of optical remote sensing imaging are degraded by noise and haze,a fast noise reduction and haze reduction algorithm combining multi-frame fusion noise reduction and dark channel prior method for haze reduction is proposed.In view of the low main frequency and limited computing power of space-borne computers,this paper changes the pixel-by-pixel registration of the traditional multi-frame fusion algorithm to two-level registration in the noise reduction stage,and the absolute error and phase-related registration are respectively used globally and locally.A new registration quality evaluation method based on the registration result is proposed,which reduces the evaluation time;in the haze removal stage,a Gaussian kernel are used to convolve with the transmittance template,instead of the traditional refinement method.The experimental results show that the algorithm proposed in this paper can increase the average peak signal-to-noise ratio(PSNR)of the image by 8.97 dB.Through the optimization of the amount of calculation,the time consumption of haze removal is reduced by 73.41%.