| The research for new methods to process and analyze high resolution remotely sensed images is one of the most important research subjects in the field of remote sensing information processing. Compared with the low-middle resolution data, structural features of the objects in the high resolution image are described in more detail, which makes the discrimination of objects by their spatial information much easier. As one of the important features of the high resolution image, texture has been widely used in remote sensing image processing. To explore new texture analysis method is of vital importance in the field of texture analysis.Local Binary Pattern (LBP) is a newly developed texture analysis method which is theoretically simple yet powerful. Although it has been widely used in the field of computer vision and shown excellent results, LBP has been rarely used in remote sensing image processing. In this study, LBP algorithm was used to extract texture information of high resolution image, and then the texture information combined with spectral information was used in the classification. Multi-dimensional extension of LBP algorithm was made to obtain multivariate texture of the multi-channel data. The effectivity of the extended algorithm was tested by adding the multivariate texture to the classification. Moreover, methods of feature extraction involved during the classification were further studied.Existed LBP algorithms were firstly employed to extract the texture information of the remote sensing image, and the obtained textures were then added in the classification to evaluate their effectivities. The existed algorithms include single-band LBP algorithm and bi-band LBP algorithm. The experiment results show that the overall accuracies of the texture-added classification results are much higher than that of the spectral-only classification results. For classes whose texture features are distinct, significant improvement was made in terms of accuracy. |
PDF Full Text Request