A multiscale urban complexity index based on 3D wavelet transform for spectral–spatial feature extraction and classification: an evaluation on the 8-channel WorldView-2 imagery

Abstract

The three-dimensional wavelet transform (3D-WT) processes a multispectral remotely sensed image as a cube and hence it is able to simultaneously represent variation information in joint spectral–spatial feature space. The urban complexity index (UCI) built on the 3D-WT is defined by comparing the amount of spectral and spatial variation, since natural features have relatively smaller spatial changes than spectral changes but urban areas show more variation in the spatial domain. The calculation of the UCI is subject to the selection of window sizes; therefore, in this study, a multiscale UCI (M-UCI) is proposed by integrating the UCI features in different moving windows and decomposition levels. The performance of the M-UCI was evaluated on two WorldView-2 data sets over urban and suburban areas, respectively. Experimental results showed that the M-UCI was effective in integrating multiscale information contained in different windows and gave higher accuracies than the single-scale UCI. In experiments, the proposed M-UCI was compared with a pixel shape index (PSI), which is a texture measure extracted from the spatial domain alone. It was revealed that the PSI was more effective for the classification of urban areas than natural landscapes, whereas the M-UCI was applicable for both urban and natural areas since it represented the joint spectral–spatial domains.

Acknowledgements

This work was supported by the National Natural Science Foundation of China under Grant No. 41101336 and 41061130553, the Fundamental Research Funds for the Central Universities under Grant No. 3101016 and the LIESMARS Special Research Funding.