一种基于Gabor滤波器组和LBP的多特征联合SAR纹理图像分类方法

实现能够使先进飞行器根据获取的图像自动识别不同的地貌景物,是一种具有实际应用前景的技术需求.提出了联合Gabor滤波器组和局部二值模式来对SAR纹理图像进行分类的新方法SARICIT (SAR Image Classification using Inquiry Table).首先对第一套带类标的训练图像集提取两种特征,分别使用的基于非监督和监督模式相融合的混合神经网络分类器进行训练,然后使用第二套带类标的训练图像集制作二维分类信息查询表,记录两种分类器对每一幅图像的判断结果.在实际进行分类阶段,对新图像提取Gabor和LBP两种纹理特征,输入训练好的分类器.根据两种分类器给出的类型响应,结合查询表,使用一种投票的机制来确定待分类的图像的纹理属性.通过对真实SAR图像的实验结果表明,与流行的单独使用一种纹理特征进行分类相比,新方法能够对SAR图像纹理做到更准确的分类,对雷达图像更具有适用性.

SAR Texture Image Classification Based On Gabor Filter Banks and Local Binary Pattern

[Abstract] The technology of automatic recognition of geomorphologic landscape based on raw images captured by aerial craft has an extensive future. We put forward in this paper a new method called SARICIT (SAR Image Classification using Inquiry Table) which utilizes texture features based on Gabor filter banks and local binary pattern to classify the SAR images. In the first place, we used images in the training set to train the hybrid neural network (HNN). After that, we made a two-dimensional classification inquiry table by using a new set of labeled training images. In the phase of classifying a new image, we extracted two kinds of texture features, got category labels from two classifiers and gave the category result using a voting mechanism based on the classification inquiry table. The experimental results with the ground truth data bore out that our method can give more accurate result compared to the traditional approach that uses only one kind of texture feature thus justifying the applicability to SAR images.