遥感图像森林植被纹理的自适应滤波

摘要

The segmentation method based on texture filtering is widely used in remote sensing image forest vegetation segmentation.The diversity of forest vegetation texture in remote sensing images makes the texture filtering method of fixed parameters can not express texture features accurately, resulting in low accuracy of segmentation.A filtering method is proposed to automatically adapt to the texture of forest vegetation.The filtering parameters are set according to the texture basic attribute of the typical forest vegetation area in remote sensing image to realize the targeted texture filtering.The attributes of the texture primitive elements in the typical forest vegetation area are obtained by the blue noise detection method and the gray level co-occurrence matrix statistic method.The texture parameters including the window size, direction, frequency and intensity of the filter and the integral window size of the local spectral histogram for expressing the texture feature are set according to the prior knowledge of the forest vegetation texture.Segmentation experiments show that this method makes full use of the characteristics of the forest vegetation texture in the image, and the feature of the texture filter is more distinguished.%基于纹理滤波的分割方法被广泛用于遥感图像森林植被分割.遥感图像中森林植被纹理的多样性使得固定参数的纹理滤波方法不能准确表达纹理的特征,导致分割精度不高.提出一种自动适应森林植被纹理的滤波方法,根据遥感图像中典型森林植被区域的纹理基础属性设置滤波参数,实现有针对性的纹理滤波处理.通过蓝噪声探测方法和灰度共生矩阵统计方法获取典型森林植被区域的纹理基元在尺度和灰度分布等方面的属性,结合森林植被纹理的先验知识设置纹理滤波参数,包括滤波器的窗口尺寸、方向、频率和强度以及用于表达纹理特征的局部谱直方图的积分窗口尺寸等.分割实验表明,该方法充分利用了图像中森林植被纹理的特点,纹理滤波表达的特征区分度更大.