论文部分内容阅读
针对Haze optimized transformation(HOT)方法存在的地物敏感性、过度矫正、红绿蓝波段RGB合成影像色彩失真等问题,提出了相应的改进方法。首先,采用归一化差分植被指数(NDVI)结合地物红蓝光谱差(RBSD)制作通用掩膜,并利用掩膜提取原始影像植被覆盖区对应的原始HOT图部分作为HOT值评估雾霾强度的有效像素集;然后从有效像素集出发推断非植被区的HOT值,得到有效HOT图;最后以有效HOT图为参考,实施暗目标减法。在暗目标减法过程中,首先利用直方图取百分位数的方法确定起始波段的改正值,然后根据散射模型计算其他波段的改正值。在红蓝光谱空间中,去雾后的影像表现出与原始无云区相似的特征,同时保持了不同地物间的差异。实验表明:改进的HOT方法能有效去除雾霾及薄云;有效解决了HOT对水体、裸地、人造地物等地表覆被类型的敏感性问题,避免了RGB合成影像的色彩失真;并且统一了不同波段的纠正尺度,解决了某一(或几个)波段的过度矫正问题,防止了块斑和光晕的产生。
Aiming at the problems of the ground object sensitivity, overcorrection and the color distortion of RGB synthetic image of Ha-green-blue band in Haze optimized transformation (HOT) method, the corresponding improvement methods are put forward. First, a generic mask was generated using the Normalized Difference Vegetation Index (NDVI) combined with the Landsat red-blue spectral difference (RBSD) and the raw HOT map corresponding to the original vegetation cover was extracted as a HOT value to estimate the haze intensity And then from the set of valid pixels to infer the non-vegetation area HOT value, get an effective HOT map; Finally, the effective HOT map as a reference, the implementation of dark target subtraction. In the dark target subtraction process, the histogram is first used to determine the percentile of the initial band correction, and then calculated according to the scattering model of the other band correction. In the red-blue spectral space, the de-fogged image shows characteristics similar to those of the original cloud-free area while maintaining the differences between the different features. Experiments show that the improved HOT method can effectively remove the haze and the thin cloud, effectively solve the sensitivity of HOT to the surface cover types such as water bodies, bare land and man-made objects, and avoid the color distortion of RGB composite images. Correct the scale of different bands, to solve a (or several) over-correction band, to prevent the formation of blobs and flare.