The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Download
Publications Copernicus
Download
Citation
Articles | Volume XLII-2/W13
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLII-2/W13, 133–138, 2019
https://doi.org/10.5194/isprs-archives-XLII-2-W13-133-2019
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLII-2/W13, 133–138, 2019
https://doi.org/10.5194/isprs-archives-XLII-2-W13-133-2019

  04 Jun 2019

04 Jun 2019

A MODIFIED THREE-DIMENSIONAL GRAY-LEVEL CO-OCCURRENCE MATRIX FOR IMAGE CLASSIFICATION WITH DIGITAL SURFACE MODEL

L. Yan and W. Xia L. Yan and W. Xia
  • School of Geodesy and Geomatics, Wuhan University, 129 Luoyu Road, Wuhan 430079, China

Keywords: GLCM, Feature Extraction, Random Forest, Image Classification, Digital Surface Model, Image Texture analysis

Abstract. 2D texture cannot reflect the 3D object’s texture because it only considers the intensity distribution in the 2D image region but int real world the intensities of objects are distributed in 3D surface. This paper proposes a modified three-dimensional gray-level co-occurrence matrix (3D-GLCM) which is first introduced to process volumetric data but cannot be used directly to spectral images with digital surface model because of the data sparsity of the direction perpendicular to the image plane. Spectral and geometric features combined with no texture, 2D-GLCM and 3D-GLCM were put into random forest for comparing using ISPRS 2D semantic labelling challenge dataset, and the overall accuracy of the combination containing 3D GLCM improved by 2.4% and 1.3% compared to the combinations without textures or with 2D-GLCM correspondingly.