Classification Algorithms Based On Texture Features For Remote Sensing Image

The application level and practical value of the remote sensing can be influenced directly by the classification accuracy of remote sensing images. As texture features consist of important information and basic character of the images, how to classify multi kinds of objects and satisfy certain accuracy has been a key problem in the study of remote sensing images.In this paper, texture classification algorithms, Gabor transform and GMRF, are researched respectively by integrating texture features of remote sensing images. A classification algorithm based on Wedgelet transform and Gabor transform for remote sensing images is proposed to reduce the influence of the window size to classification accuracy. In the algorithm, the multi-scale Wedgelet transform is applied to piecewise linear represent the remote sensing images, and then texture features are extracted from each segmented region to realize the effective classification of remote sensing images. As it's difficult to describe the global characteristics of remote sensing images by Gabor texture features and noises may be included in Gabor texture features, the Colon selection algorithm is applied to find the optimal texture features by selection, colon and mutation operating. In order to solve the defect of GMRF in estimating local texture construction and recognizing global texture features, a iterative formula is employed to obtain the optimal parameter estimation of single pixel GMRF, then by calculating the flourier power spectrum characteristic of single pixel Markov neighborhood, which can be used as the local weights of parameters of the single pixel GMRF model, we can utilize the GMRF model parameters of all the pixel to compose high-dimensional matrix. Finally, the K-L transformation is applied to the matrix and the result of the transformation is applied to classify remote sensing images.