一种用于遥感图像云雾处理的小波系数加权算法

Application of Wavelet Coefficient Weighted Algorithm to Remote Sensing Image Processing

依据小波变换理论分析,得出图像经多层小波变换后,低层细节系数频率高于高层细节系数,近似系数的频率最低.遥感图像景物的频率较高,云雾频率较低.提出通过选择合理的分界层数,将景物信息尽量分配到低层细节系数,云雾噪声尽量分配到高层细节系数,对低层、高层细节系数、近似系数分别给予权重,增大低层细节系数,突出景物信息,减小高层细节系数,削弱云雾噪声,依据近似系数包含云雾状态,保持或减小近似系数.同时还提出以信息熵作为分界层数和权重选择的依据.实验证明,该算法优于同态滤波和Retinex算法.

On the basis of wavelet transform theory,it is concluded that the frequency of detail coefficients in low levels is higher than that in high levels,and the frequency of approximate coefficients is the lowest after the digital image is decomposed into wavelet coefficients.In remote sensing images the frequency of thin cloud and mist is lower than that of the sceneries.This paper proposes an algorithm of the weighted wavelet coefficient,i.e.,scenery information is distributed to detail coefficients in low levels and cloud and mist noises are distributed to those in high levels to the full by choosing reasonable level number.Detail coefficients in low levels and high levels,and approximate coefficients are weighted with different factors to increase detail coefficients in low levels and decrease those in high ones.Thus scenery information is enhanced,and cloud and mist noises are weakened.At the same time,taking information entropy as a criterion for choosing the number of the demarcation levels and the weigh ted factors is also proposed.The algorithm is proved to be better than homomorphic filtering and Retinex by experiments.