联合非降采样金字塔与经验模态分解的遥感图像融合算法

Fusion algorithm of remote sensing images based on nonsubsampled pyramid and empirical mode demoposition

为了更好地进行遥感图像融合,联合非降采样拉普拉斯金字塔变换(NLP)和二维经验模态分解(BEMD),提出了一种利用分解系数绝对值和瞬时频率作为融合特征的遥感图像融合新算法.首先利用非降采样金字塔对高分辨率全色图像(PAN)进行分解,使其低频部分和低分辨率全色图像(MS)具有相同的尺度特性;再对低频部分和MS图像进行BEMD分解,得到二维内蕴模态函数(bimf)和趋势图像,并计算各层bimf的4方向瞬时频率.为了尽可能提高空间细节质量,利用瞬时频率和分解系数绝对值作为融合特征,并考虑bimf部分对应位置系数的正负关系,采用加权算法对高频细节部分进行融合;最后进行相应的BEMD和NLP逆变换,得到融合图像.实验表明,该方法对融合影像的光谱质量和空间细节质量都有较好的改善.

In order to improve the quality of the fused image, this study proposes utilizing a new method of remote sensing image fusion based on the decomposition coefficient absolute and instantaneous frequency as the fusion fea- ture, and the combination of the nonsubsampled laplacian pyramid （NLP） and Bidimensional empirical mode de- composition （BEMD）. First, the high resolution panchromatic image （PAN） was decomposed by using the NLP, which makes the low frequencies have the same sealing features with the low resolution multispectral image （MS）. Then the low frequencies and the MS were decomposed by the BEMD to obtain the bidimensional intrinsic mode function （bimf） and trend image respectively, and the instantaneous frequencies in four directions and in every lay- er of the bimfs were calculated. In order to maximize the quality of the spatial detail, the decomposition coefficient absolute value and instantaneous frequency were taken as the fusion feature, and considerations were taken regard- ing the positive and negative relationship of the corresponding position coefficients of bimf. A weighted algorithm was designed for fusing the high frequency details. Next, the fusion image was obtained through inverse transform of BEMD and NLP. The experimental results showed that this method could improve both spectral quality and spatial detail quality of a fused image.