Remote Sensing Image Segmentation Base On Dual-Tree Complex Wavelet Transform And Gray-level Co-occurrence Matrix

As an important development direction of image intelligent processing, image segmentation gets high attention in image processing domain. Remote sensing image segmentation, as one branch of image segmentation, obtains researchers'attention deeply. Compared with other types of images, remote sensing image has more gray levels, large information, fuzzy border, as well as'same object different spectrum'and'different object same spectrum'. Because of these characteristics, remote sensing image segmentation is too hard. However, as the earth observation satellite technology continues to mature, texture information of the remote sensing image is more and more abundant. How to use texture information in remote sensing image segmentation currently becomes one of the problems which the scholars pay close attention to. Texture feature extraction is the basis of this project.Using in remote sensing image texture analysis and texture characteristics extraction, it can advance the automation of remote sensing image interpretation. In remote sensing image segmentation, it helps to improve the final remote sensing image segmentation through the texture analysis method combined with conventional segmentation method. It also can understand the remote sensing image better and extract all kinds of useful information from the remote sensing image data.In this paper, based on the extensive literature on the remote sensing image segmentation techniques are studied, we propose a method to describe remote sensing image texture features based on Dual-Tree Complex Wavelet Transform (DT-CWT) and Gray-level Co-occurrence Matrix(GLCM). This method uses DT-CWT high-frequency sub-bands'Gamma and Lognormal parameters and features of GLCM as the feature vector of remote sensing image pixels. Then, use the K-means clustering to complete remote sensing image segmentation. The results of experiment prove that the feature based on this method can obtain more accurate remote sensing image segmentation results.