针对高光谱图像解译需求, 提出了一种基于目标检测理论的 NSCT域高光谱图像与全色图像融合方法。首先对高光谱图像进行 RX异常目标检测, 得到目标背景分离图像; 然后对参与融合的波段图像进行无下采样轮廓波分解, 得到不同分辨率的低频子带和多方向的带通子带; 对于背景区域的低频子带系数和高频子带系数, 分别采用加权平均和平均梯度自适应加权的融合策略, 对于目标区域, 则根据不同特征采用区域方差自适应加权的低频系数融合方法和区域方差取大的高频系数融合方法; 最后进行 NSCT逆变换得到融合图像。实验结果表明, 本文提出的融合方法能够有效提高高光谱图像的目视效果, 突出目标与背景区域的差异, 有利于目视解译工作的进行。

To easily interpret hyperspectral imagery, a new fusion method based on the target detection theory in the Nonsubsampled Contourlet Transform(NSCT) domain is proposed in this paper. First, target detection is performed on hyperspectral imagery to separate the targets of interest from the background; then, the bands of images chosen for fusion are decomposed into a low frequency subband and several bandpass directional subbands by NSCT. For the low frequency subband and directional subbands of the background region, the coefficients are fused using the weight average method and the adaptive weighted method based on the regional average gradient. Next, for the target part, the coefficients are fused using the adaptive weighted method based on regional average energy and regional average variance considering the different positions of the pixels. Finally, the fusion image is reconstructed using the fusion coefficients by inverse NSCT. The experimental results indicate that the fusion imagery obtained by the proposed method has better performance with respect to either visual effect or objective evaluation indexes including standard deviation, average gradient, information entropy, spatial frequency, and figure definition for both self-fusion of hyperspectral imagery and fusion of hyperspectral imagery and panchromatic images with higher spatial resolution.