Texture Analysis And Classification Of Remote Sensing Image Based On Fractal Theory

With the in-depth research of remote sensing image processing and analysis, it was found using spectral information alone cannot meet the needs of remote sensing applications. But texture feature, as important information of remote sensing image, plays a crucial role in remote sensing image classification. Fractal theory is an active branch of nonlinear science. As fractal dimension is consistent with human perception of image surface texture roughness, using fractal theory in texture analysis of remote sensing images has potential theoretical and practical value.This paper not only described the current widely used texture analysis methods, but also summarized and concluded these methods. A program was constructed on Matlab platform to extract texture features of high-resolution Worldview images and moderate resolution TM/ETM image based on fractal theory. Then the acquired texture features were combined with the original spectral information for image segmentation and performing supervised classification and unsupervised classification experiment. After experimental analysis, series of research results were achieved, as follows:1. For different resolution images, the fractal dimension can reflect the roughness of texture, but the results were varied by different fractal dimension model. Experiment showed that result of double blanket coverage model was better than that of differential box-counting method.2. Compared with the traditional GLCM method, texture features based on the fractal dimension can increase the classification accuracy and was less affected by structural factors, while an appropriate scale be selected.3. Compared with simply using the spectral information, image classification accuracy was improved greatly by using texture features. And the contributions of different combination of textures to improve the classification accuracy were varied. Combining texture features based on fractal dimension with GLCM was more conducive to the improvement of classification accuracy. 4. After the image transformation, the result of image segmentation using multi-fractal dimension features of original image and transformation images was better than that of merely using a single fractal dimension characteristic.This study presented an efficient method for texture analysis of remote sensing image based on fractal theory. A technical guidance and reference was also offered for remote sensing image classification combing with texture features.