基于小波阈值理论的光学图像去云处理新算法

分析了部分云覆盖的单幅光学遥感图像经过多层小波变换后,景物信息和云噪音在小波分解系数中的分布关系,并提出了云区阈值法来有效去除云噪音.通过选择适当的分界层数,将景物信息、云噪音尽可能分别分配到低层和高层细节系数中.高层细节系数中除主要包含云噪音外,也有部分有用景物信息.由于云噪音亮度大于景物,根据亮度特征合理选择高层细节系数的亮度阈值,去除云噪音,而保留其中的景物信息.通过对低层、高层细节系数和近似系数分别设置权重,增加景物对比度,减小残留云影响,从而重构得到恢复图像.提出了以信息熵作为分界层数、权重、阈值等参量选择的定量标准.实验证明,按信息熵标准能正确地选择参量,依据本文算法得到的去云效果远优越于同态滤波和Retinex算法,且能充分保留云区以外景物信息.

A New Algorithm of Cloud Removing for Optical Images Based on Wavelet Threshold Theory

Frequency relationships of wavelet coefficients describing sceneries and cloud in a single optical remote sensing image are analyzed after it is decomposed by wavelet transform.An algorithm of cloud threshold is proposed for removing cloud.Scenery information and cloud noises are distributed to coefficients of low levels and high levels respectively by choosing appropriate numbers of demarcation levels.Most of cloud noises and some scenery information are included in high level coefficients where cloud is brighter than scenery.Cloud can be removed from these coefficients and scenery information can be kept by setting a threshold with brightness.Weight factors are assigned to detail coefficients of low level,high level,and approximation coefficients respectively for enhancing the contrast of sceneries and decrease remaining cloud.The three parts of coefficients are reconstructed and fused to get processed result.Entropy is proposed to be a criterion to choose best demarcation level,weighted factor and threshold.It is proved that according to the entropy as a criterion these parameters can be chosen correctly by experiments.And the proposed algorithm is better than homomorphism filtering and Retinex algorithm.