基于立体环形邻域的GMRF模型及其在遥感影像纹理识别中的应用

对传统的GMRF模型进行了扩展,建立了基于立体环形邻域的GMRF模型,并设计了分步模型参数求解方法。这种新的模型不仅全面考虑了多光谱高分辨率影像中各波段内像元之间的空间相关性,而且还顾及了波段间像元的相关性。与传统的GMRF模型相比,基于立体环形邻域的GMRF模型提取的纹理信息更为丰富。采用Forrest彩色纹理图像和QuickBird卫星遥感影像进行了实验验证,实验结果表明本文提出的基于立体环形邻域GM-RF模型的纹理识别算法具有较强的普遍适用性,对不同情况的纹理影像(即空间相关性主要存在于波段间还是波段内)进行识别均能获得较好的分类效果。

The GMRF model based on the cubic circular neighborhood and its application in remote sensing image texture recognition

Traditional GMRF models are extended, and the GMRF model based on the cubic circular neighborhood is presented, and the corresponding step-by-step parameter estimation scheme is proposed. This novel model considers not only full spatial interactions within each spectral band, but also complete interactions between different spectral bands. Compared with tradition GMRF models, the GMRF model based on the cubic circular neighborhood can extract much richer texture information. The Forrest color texture image and QuickBird satellite remote sensing image are used to verify the effectiveness of the proposed algorithm. Experimental results show that the GMRF model based on the cubic circular neighborhood is much more effective in texture recognition than traditional methods and robust whether spatial correlations dominantly exist in intra-bands or inter-bands.