A vector SIFT operator for interest point detection in hyperspectral imagery

An algorithm for automated extraction of interest points (IP) in hyperspectral images is presented. IP are features of the image that capture information from its neighbors and are distinctive and stable under translation and rotation. IP operators for gray level images were proposed more than a decade ago and have since been studied extensively. IP are helpful in data reduction to reduce the computational burden of various algorithms by replacing an exhaustive search over the entire image domain by a probe into a concise set of highly informative points. The vector SIFT approach extends Lowe's IP operator that uses local extrema of Difference of Gaussian at multiple scales to detect interest point in gray level images by direct conversion of scalar operations such as scale-space generation, and extreme point detection into operations that take the vector nature of the image into consideration. Vector anisotropic diffusion is used for scale-space generation which enhances edges and improves IP detection. Experiments with hyperspectral images of different spatial resolutions and evaluation of IP found based on image registration is presented.