Hierarchical fusion using vector quantization for visualization of hyperspectral images

Visualization of hyperspectral images that combines the data from multiple sensors is a major challenge due to huge data set. An efficient image fusion could be a primary key step for this task. To make the approach computationally efficient and to accommodate a large number of image bands, we propose a hierarchical fusion based on vector quantization and bilateral filtering. The consecutive image bands in the hyperspectral data cube exhibit a high degree of feature similarity among them due to the contiguous and narrow nature of the hyperspectral sensors. Exploiting this redundancy in the data, we fuse neighboring images at every level of hierarchy. As at the first level, the redundancy between the images is very high we use a powerful compression tool, vector quantization, to fuse each group. From second level onwards, each group is fused using bilateral filtering. While vector quantization removes redundancy, bilateral filter retains even the minor details that exist in individual image. The hierarchical fusion scheme helps in accommodating a large number of hyperspectral image bands. It also facilitates the midband visualization of a subset of the hyperspectral image cube. Quantitative performance analysis shows the effectiveness of the proposed method.