Fusion of panchromatic and multispectral images for classification using the Chinese restaurant franchise with shaped tables

Abstract

In this paper, we propose a novel framework to fuse both panchromatic (PAN) and multispectral (MS) images for classification under a Chinese restaurant franchise metaphor (CRF). In the metaphor of CRF, one kind of observations would be interpreted by using two sequent random processes: (1) a customer randomly selects a table in a restaurant to sit and (2) randomly selects a dish to eat for a newly occupied table. In our method, shaped tables (i.e., local semantic segments) are discovered from panchromatic images in the process of table selection. In the other process, a dish (i.e., an unsupervised class label) is allocated based on multispectral images for each table discovered in panchromatic images. This approach takes advantage of the rich spatial and spectral information in panchromatic and multispectral image respectively. The result indicates that the proposed algorithm outperforms these exiting state-of-art methods in all of the experiments.