图像纹理特征提取及其分类性能分析
在过去的几十年里，纹理特征提取有了特别大的进展，纹理特征提取算法也是层出不穷。纹理特征提取的算法总体上可以分为四个大类——结构分析法、统计分析法、频域分析法和模型法。其中比较著名的算法有灰度共生矩阵、LBP、gabor变换等。纹理特征的本质是一种全局特征，它描述了图像或图像区域所对应景物的表面性质。
本文首先介绍了上述三种纹理特征提取算法，概述了算法的原理以及公式。然后采用上述三种算法提取了不同遥感图像的纹理特征，并且对提取出的特征向量做了比较。对于同一类的遥感图像（如两张图像都是农田），所提取出的特征向量应该比较接近；对于不同类的遥感图像（如一张农田，一张城市），所提取出的特征应该有一定的区分度。最后，尝试通过SVM对图像进行分类，测试分类的精度。我们很难说那种算法是最好的，但是对于不同的图像、不同的纹理，不同的算法有各自的优势和劣势。23122
关键词 纹理；特征提取；灰度共生矩阵；LBP；gabor变换；分类；SVM
毕业设计说明书（毕业论文）外文摘要
Title Image Texture Feature Extraction and Analysis of
Classification Performance
Abstract
With the popularity of Multimedia,Image as a major carrier of information, has become an indispensable part of our life.The important information of images is the texture feature. Texture is one of a key visual cue,which is hard to describe in image processing field,also is one of the most important sources of information in image analysis.Texture classification and segmentation is a hot research topic in the field of image processing. As the primary issue in texture classification and segmentation,feature extraction has been the focus for researchers.Texture analysis has an important position in pattern recognition,computer vision,image processing and so on.The significant thing of studying on feature extraction is the extraction of texture features.
In the past few decades,Texture feature extraction technique has made great progress and all kinds of texture feature extraction algorithms emerge in an endless stream.Texture feature extraction algorithm can be pided into four categories in general—structural analysis method, statistical analysis method,frequency domain analysis and modeling method.For example,Some widely used famous algorithms are gray level co-occurrence matrix,LBP,Gabor transform and so on. The essence of texture feature is a global feature,which describes the surface properties of the image or the image regions corresponding to a scene.
This dissertation first introduces three kinds of texture feature extraction algorithms,outlines each algorithm principle and formula.Then extract the texture characteristics of different remote sensing images using the three algorithms as aforementioned,and compares the extracted feature vectors.For remote sensing images of the same class (e.g. farmland),extracted feature vectors should be similar.For remote sensing images of different classes (e.g. farmland and city region),extracted features should have a large distances among classes. Finally,we use the SVM to classify images,and test the classification ability of different texture features.It is difficult to judge that which algorithm is the best, but for different image or texture,different algorithms have their own advantages and disadvantages.