论文部分内容阅读
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 traditional GMRF model is extended, a GMRF model based on the three-dimensional annular neighborhood is established, and the method for solving step-by-step model parameters is designed. This new model not only fully considers the spatial correlation between pixels in each band in multi-spectral high-resolution images, but also considers the correlation of pixels between bands. Compared with the traditional GMRF model, the GMRF model based on the three-dimensional annular neighborhood extracts texture information more abundantly. The experimental results of Forrest color texture image and QuickBird satellite remote sensing image show that the proposed texture recognition algorithm based on the three-dimensional ring neighborhood GM-RF model has a strong universal applicability. The texture image of different conditions Spatial correlation mainly exists in the band or the band) to identify can get a better classification results.