Second Order Statistical Texture Analysis Of Classes Features Of Digital Remote Sensed Data

The thesis is submitted to Faculty of Basic Science and Technology, Kunming University of Science and Technology for the master degree of Mr Lv Peng in physical electronics major. The objectives of the research are to study the statistical texture features of digital natural images theoretically and obtain related results.Research and analysis of digital images is a hot research field at present. A great deal of resources and researchers is involved in the field in universities and institutes all over the world. Wide range of research topics of digital image anlysis are presented especially texture analysis in various publications, which is the hot and difficult research point in image and signal processing community. The achievements of texture research begin to play a more and more important role in pattern recognition and computer vision. Meanwhile, the research approaches and achievements can also find important applications in many fields, including industry inspection, remote sensed data analysis and mapping, medical imaging, textile defect detection, vedio image analysis, food grading, and natural texture recognition and retrivel, object recognition, etc.The objectives of the thesis start form the study on the algorithm of texture features using second order statistics, expounded the principles of GLCM and statistical features formula. On this basis, we selected 7 typical class samples of from QuickBird RGB fusion data that accuracy is 0.61 m. Calculate the GLCM of these 7 typical class samples and more than 100 sets data form the Orientation and distance of 6 statistical features.Analyzed statistical features of these samples form different perspectives and angles.The 6 statistical features are Energy, Entropy, Contrast, Homogeneity, Correlation, Dissimilarity.By the theoretical derivation,computation of the features and comparative study,we get the conclusion:(1) View of the computation results from GLCM, different types of texture samples has great difference GLCM, therefore, GLCM texture images can be considered to be the identity of a certain label. In other words, different GLCM represents a different texture, but also one correspondence. At the same time can be predicted that the different GLCM, the calculated values of the various features of the levy amount will be different, it also reflects differences in the texture of each sample. (2) GLCM have a high degree of response on both artificial and natural textures, the characteristics by GLCM have different levels of response on texture.(3) The 6 characteristics of GLCM have a certain validity on classification of the 7 typical samples form the remote sensing image of QuickBird, through the characteristic curve of the comprehensive comparison,many experiments, and the exclusion of special circumstances (some kind of characteristic curves is intersect), determine the most appropriate size of the GLCM window:between 60×60～130×130 pixels. (4) In these 6 features, Dissimilarity, Entropy, Homogeneity,and Contrast are the most different from the features in turn.They are important for the classification.(5) Correlation and Contrast may be have superior sensitivity than human eyes on the texture.(6) In the thesis,more than 100 sets data reflect the different perspectives and characteristics of statistical texture features from the samples.And the data already has some systemic. These researches of remote sensing data have an important value.for image classification, land features coverage quantitative interpretation and the thematic maps produced.