基于分形特征的云雾遥感图像分离方法

Study for Separation of Fog from Clouds Based on Fractal Dimension

雾是一种气象灾害,将雾从卫星云图上分离出来仍非易事。分数维给予图像纹理统计意义的描述,有效地体现了纹理的复杂度和粗糙度,揭示了纹理内在的自相似性,为云雾图像纹理分析提供了新的思路。简要阐述了纹理图像的差值盒维数计算方法,计算并分析了云雾纹理图像的差值盒维数特征。针对差值盒维数在表现云雾纹理特征和云雾分离方面存在的问题,提出基于样本图像灰度均值的加权盒维数算法,以改变出现灰值差异较大的不同云类具有相同盒维数的情况,并与云雾的光谱特征结合,实现雾与云的识别与分离,在实际应用中取得了较好的效果。

Fog is one of the important disasters and separating fog from clouds using satellite images is still a difficult task nowadays. The concept of fractal dimension offers a new way to image texture analysis, because, it accords image texture a statistical description and exhibits complexity and roughness of texture, In this paper we describe a method of difference boxcounting dimension（DBD） in texture image and then analyze the advantages and disadvantages of DBD. In order to avoid the case which different clouds have same DBDs and to improve abilities of fog separating using DBD, a modified DBD （MDBD） based on average gay weighting is proposed and a novel algorithm is developed for MDnDs and spectral features in combination to distinguish fog from Cloud with fairly good results achieved in our experiments.