城市卫星遥感图像融合处理质量评价研究

Assessing quality of images fusion of different approaches

常用的遥感融合方法常导致较严重的光谱畸变,为减少融合图像光谱特征的扭曲,提出三种新融合方法即合成变量比值法(SVR)、平滑滤波亮度调制法(SFIM)和Gram_Schimdt变换法(GS)。采用定量分析方法,分别对中等分辨率Landsat ETM+数据和高分辨率Quickbird数据的融合效果进行了评价。结果表明,不同方法具有不同的光谱保真度和空间信息融入度。同一种方法对于不同分辨率的遥感数据具有不同的融合效果。对中等分辨率Landsat ETM+数据,SFIM能产生较高的空间信息融入度和光谱保真度。利用中等分辨率Landsat ETM+数据进行融合处理时,SFIM优于合成SVR和GS;在高分辨率Quickbird数据的融合中,SVR能产生较高的空间信息融入度和光谱保真度。利用高分辨率Quickbird数据进行融合处理时,SVR则优于SFIM和GS。在中等分辨率Landsat ETM+数据、高分辨率Quickbird数据融合处理中,基于SFIM、SVR融合方法能分别获得较好的视觉效果,又能改善目视解译和遥感分类精度。

The popular image fusion methods in remote sensing community usually distort the spectral characteristics. To reduce the spectral distortion, some image fusion techniques have been developed. This paper addresses the issue in quality assessment of fused images from three new fusion methods. These methods are synthetic variable ratio （ SVR）, smoothing filter - based intensity modulation （SFIM） and Gram_Schimdt transform （GS） that are recently developed. In this study we employed these methods in image fusion of Landsat 7 ETM ＋ panchromatic with multispectral images and Quickbird panchromatic with multispectral images. The quantitative methods such as standard deviation, information entropy, correlation coefficient, and spectral bias index were used＇to assess the quality of fused images. The results indicate that different approaches have their specific properties and adapt to different purposes based on spectral fidelity and high spatial frequency information gain. The quality of fused images based on SFIM and SVR methods is better than that of GS method, respectively, in medium - resolution images and high - reso- lution images in urban area. Therefore, the SFIM and SVR methods can meet the needs of mapping - oriented fusion, classifica- tion - oriented fusion, and visualization - oriented fusion purposes, respectively in medium - resolution images and high - resolu- tion images in urban area.