中国激光, 2015, 42 (1): 0114001, 网络出版: 2014-12-11
基于显著性分析的自适应遥感图像融合 下载: 634次
A New Adaptive Fusion Method Based on Saliency Analysis for Remote Sensing Images
图像处理 图像融合 显著性分析 多尺度谱残差 小波变换 亮度色调饱和度变换 image processing image fusion saliency analysis multi-scale spectral residual wavelet transform intensity hue saturation transform
摘要
针对遥感图像融合中,不同地物区域对空间与光谱信息要求不同的问题,提出了一种基于显著性分析的自适应遥感图像融合算法。结合多尺度谱残差分析模型,将遥感图像分为纹理、边缘丰富的显著区域与纹理、边缘较少的非显著区域,对显著性不同的区域采用不同融合算法。针对居民区、道路等纹理、边缘信息丰富的显著区域,采用窗均值亮度色调饱和度(IHS)变换,较好地保留了空间细节;针对农田、山地等非显著区域,采用基于小波变换的融合策略保留较多光谱信息。实验结果表明,新算法能使融合结果中的显著区域保留更多空间细节,非显著区域保留更多光谱信息,为今后的遥感图像融合研究提供了一定的理论与应用价值。
Abstract
In remote sensing image fusion, these different object regions have different demands on the spatial and spectral resolution. A new adaptive fusion method based on saliency analysis for the remote sensing image is proposed. The multi-scale spectral residual (MSR) analysis model is introduced to divide the remote sensing image into the salient regions with the rich texture and edge information and the non- salient regions with the less texture and edge information. Two different fusion algorithms are applied to the two kinds of areas. The window mean intensity hue saturation (IHS) transform can be used in salient regions with the rich texture and edge information, such as urban areas and roads, to retain more texture and edge information while the wavelet transform can be used in non-salient areas, such as farmlands and mountains, to retain more spectral information. The experimental results show that the new algorithm can obtain the salient regions as urban areas and roads with more texture and edge information and non- salient areas as farmlands and mountains with more spectral information. The new method has more theory and application value in further research of remote sensing image fusion.
张立保, 章珏. 基于显著性分析的自适应遥感图像融合[J]. 中国激光, 2015, 42(1): 0114001. Zhang Libao, Zhang Jue. A New Adaptive Fusion Method Based on Saliency Analysis for Remote Sensing Images[J]. Chinese Journal of Lasers, 2015, 42(1): 0114001.