描述台风云图表面纹理结构的新模型

THE STUDY OF NEW MODEL AND ALGORITHM ON COMPUTER IMAGE TEXTURE ANALYSIS

以ＧＳ红外卫星云图为对象，采用ＳＡＲ（ＳｉｍｕｌｔａｎｅｏｕｓＡｕｔｏＲｅｇｒｅｓｓｉｖｅ）随机场纹理模型，抽取不同邻域下云图的纹理特征（ρ－粗糙度，θ－方向），确定了最佳邻域．运用分形（Ｆｒａｃｔａｌ）理论，求取了描述云图自相似性特点的多种分数维，确定了最佳分数维．在上述研究工作的基础上，建立了全新的纹理识别模型ＳＡＲＦ，研究结果表明，ＳＡＲＦ模型不仅能全方位地表征云图的纹理特征，而且具有分类精度高，速度快等优点。

Image texture analysis theory and its application in satellite imagery are discussed in this paper. Texture features of cloud imagery under different neighbourhood are extracted with the SAR random field texture model. The best neighbourhood set is choosed by using pattern recognition technique. Fractal theory is applied to the satellite imagery for its advantages in describing self similarity of high speed spinning vortex structure. Four kinds of estimation algorithms using the theory are implemented and experimental results show that information dimension is the best fractal feature to describe typhoon. A novel texture recognition model SARF model is proposed on the basis of much study stated above. The results indicate that SARF model can not only describe the total texture features of typhoon, but also help to reach a high precision and high speed in classification. The model can contribute much to the analyzing and processing of cloud imagery and hence to the development of meteorological science.