基于形态学与梯度域导向滤波的图像去雾算法

目前的多数图像去雾方法不适用于浓雾场景，存在去雾后图像亮度偏暗及光晕伪影等问题，提出一种利用图像形态学和梯度域导向滤波的去雾算法。通过暗通道先验算法得到初始透射率，并根据图像形态学闭、开运算细化和平滑初始透射率。运用梯度域导向滤波优化透射率图，以平滑透射率图的边缘和消除矩形块状效应。为更好地估计出大气光值，对雾图的最小强度图进行形态学灰度腐蚀，并经过导向滤波处理，以此结果作为暗通道图，选取其最亮的前0.1%像素点对应到原图中，最高的像素值作为大气光值，得到大气光值后利用大气散射模型求出去雾后的图像。将除雾后的RGB图像转换到HSI颜色空间，利用多曝光融合框架对I通道进行无雾图像整体亮度提高，最终转到RGB颜色空间。实验结果表明，该算法能够恢复更多的细节信息，保证图像具有合适亮度，且颜色自然，无光晕伪影，优于暗通道先验和颜色衰减先验等去雾算法。

Image Dehazing Algorithm Based on Morphological and Gradient Domain Guided Filtering

Most current image dehazing methods are unsuitable for dense fog scenes, the image brightness is dark after dehazing, and there are halo artifacts.To solve these problems, this article proposes a dehazing method using image morphological and gradient domain guided filtering.First, the initial transmittance is obtained by using the dark channel prior.Second, this study uses image morphological closing and opening operations to refine and smooth the initial transmittance.Then, to smooth the edges of the transmittance map further and eliminate the rectangular blocky effect, gradient domain guided filtering is used to optimize the transmittance map.To estimate the atmospheric light better, this study first performs morphological grayscale erosion on the minimum intensity map of the fog map.This is followed by the guided filtering process.The result is used as a dark channel map.The first 0.1% of the brightest pixels correspond to the original image, and the highest pixel value is used as the atmospheric light value.After the atmospheric light value is obtained, this study uses the atmospheric scattering model to calculate the haze-free image.Finally, the Red-Green-Blue (RGB) image after dehazing is converted to the Hue, Saturation, and Intensity (HSI) color space.The study uses a multi-exposure fusion framework to improve the overall brightness of the fog-free image on the I channel and then transfers it to the RGB color space.Experiments show that the algorithm can restore more-detailed information and ensure that the image has appropriate brightness, natural colors, and no halo artifacts.It is superior to the dark pass prior, color attenuation prior, and other dehazing algorithms.