基于波段权重的多尺度Retinex遥感图像渐晕校正方法

针对传统函数逼近法存在的校正质量不稳定、耗时长以及Retinex算法存在的光晕、泛灰和光谱数据失真的问题,该文提出了一种带光谱恢复的多尺度Retinex渐晕校正方法。通过对无人机遥感图像全局亮度的估计以及光谱恢复因子的引入,实现无人机遥感光谱图像的渐晕校正。利用该文方法对遥感图像进行处理,并与基于高斯曲面的函数逼近法和多尺度Retinex算法结果进行对比,依据灰度分布情况、标准差、平均梯度、清晰度、光谱相关系数以及光谱角指标进行评价分析。试验结果表明,该文提出的方法可以取得较好的渐晕校正效果,结果不存在光晕、泛灰现象,结果的平均梯度和清晰度均值分别为0.077 4和49.33,相较原始图像和函数逼近法以及多尺度Retinex算法处理结果,平均梯度分别提高了5.94%、5.56%和4.78%,清晰度分别提高了8.94%、6.79%和6.63%,该文方法校正图像的对比度和清晰度更优,方法具有较好的渐晕校正效果。

Vignetting correction for remote sensing image using multi-scale retinex based on band weight

Abstract: UAV low-altitude remote sensing is an important way for monitoring the growth and physiological conditions of crops. However, due to the limitations of drones and environments, the images acquired by multispectral cameras mounted on drones are always distorted. One kind of distortion is the vignetting effect of the image, which introduces errors into quantitative analysis of those remote sensing images. To solve the problems of unstable correction quality and being time-consuming in traditional function approximation method as well as being halo and gray and spectral data distortion in multi-scale Retinex (MSR) algorithm, a multi-scale Retinex algorithm with spectral restoration vignetting correction was proposed in this research. By estimating the global brightness component and introducing a spectral restoration factor, vignetting correction for spectral images in UAV remote sensing was achieved. Mean images of each band were first calculated from UAV remote sensing images of one flight, then the global brightness components of each band with smooth brightness change were estimated from the mean images by gauss kernel function. The second step was to calculate the reflectance components of each band by using the global brightness components. In this step, spectral distortion was introduced into the correction result because the reflectance components of each band were calculated independently. So the spectral restoration factor which was obtained from the original spectral image was proposed and applied in each reflection component. The quality of the corrected image would be affected by spectral restoration and there were 2 parameters which were used to balance the effect of image quality and spectral restoration. Finally, the corrected images were obtained after quantitative stretching. The proposed method was compared with the function approximation method based on gauss model and the multi-scale Retinex algorithm. On the one hand, the experimental results indicated that the proposed method could obtain a good vignetting correction effect visually and the result of the proposed method did not show gray and halo. On the other hand, the results were evaluated in terms of gray distribution, standard deviation, average gradient, clarity, spectral correlation coefficient and spectral angle index. The average gradient and clarity were 0.0774 and 49.33, respectively. Compared with the original image, function approximation method and multi-scale Retinex algorithm, the average gradient increases by 5.94%, 5.56% and 4.78%, and the clarity increases by 8.94%, 6.79% and 6.63%, respectively. The result showed that the contrast and clarity of the image corrected by the proposed method were better than those corrected by the other 2 methods. The standard deviation of MSR result was lower than that of the proposed method, which indicated that the proposed method reduced the gray effect of Retinex theory method. The average spectral correlation coefficient and spectral angle showed that the proposed method obtained a good effect of spectral restoration but the spectral quality of the proposed method was slightly worse than that of function approximation method. However, the relative deviation of the spectral correlation coefficient and spectral angle obtained by this method were smaller than those of the function approximation method, which showed that the spectral recovery effect of the proposed method was relatively stable. In addition, the proposed method effectively improved the image quality and spectral quality of the correction results based on Retinex theory. In conclusion, the image quality and spectral quality of the proposed method were better than those of the other 2 methods and the proposed method reduced the phenomena of gray and halo in the corrected images. However, there were many adjusting parameters in the proposed method. So the further research can be focused on parameter optimization to improve the efficiency of the method.