结合LBP变换的GF影像纹理特征提取

Texture feature extraction of GF image combining LBP transform

为了更好地提取高分辨率影像的纹理特征,该文以GF-1影像为实验对象,引入局部二进制模式(LBP)算子增强图像纹理信息,并且结合灰度共生矩阵计算图像纹理特征。实验通过使用3个不同的半径参数的LBP变换对GF-1影像的全色波段做计算,然后分别将此计算结果加入影像分类过程中。3种分类均采用相同尺度的面向对象分类方法、相同的训练样本和测试方法。最终结果显示,半径参数为3个像元的LBP变换提高分类精度幅度最大,参数为1个像元的没有提高分类精度,参数为2个像元的分类精度居中。实验表明,使用LBP变换能够很好地提取纹理信息、帮助分类,但需要找到合适的半径参数。

In order to extract the texture features of high-resolution images better,this paper took GF-1 image as the experimental object,improved classification accuracy by a texture feature extraction method fused with local binary pattern(LBP)operator and gray-level co-occurrence matrix(GLCM).Panchromatic remote sensing images were processed by three different radius of LBP operator.Then,they were classified and verified with the same classification and segmentation method,the same training and test samples.Finally,the classification accuracy of the images added three transformed band and original image were compared.Experimental results showed the highest classification accuracy was image which added band transformed in radius of 3 pixels,followed by radius of 2 pixels,which indicated the method combined LBP and GLCM could effectively improve the classification accuracy of high-resolution images with a proper radius.