A framework for multi-sensor image segmentation using fuzzy collaborative clustering

Summary

The massive availability of remote sensing data and advances in data analytics have improved the capacity for mapping land cover and subsurface.

Specifically, in case of multi-sensor remote sensing images, the independent analysis of each image ignores the valuable complementary information available in other images. To address this problem, a framework of multisensor image analysis using fuzzy collaborative clustering is proposed. The proposed framework avoids the independent analysis of each image but combines the information available in one image with the complementary information given by the all other images for improved image understanding and segmentation. Specifically, the clustering of pixels in each image is collaborated with the clustering results of other images to refine its results. The proposed framework can simultaneously process different heterogeneous images from various sensors. The proposed framework was evaluated and validated through an experiment in which two multi-sensor images, i.e. Landsat-5 TM and ENVISAT ASAR were used over the Beijing urban area and compared with the standard fuzzy c-means clustering. Experimental results show that the proposed framework outperforms the independent image segmentation analysis in detecting the urban growth of Beijing. This framework serves as a useful tool for various earth science applications.

• image segmentation
• fuzzy collaborative clustering
• multi-sensor images
• urban growth information extraction