Denoising on hyperspectral data by energy variations

In order to denoise in spatial and spectral dimensions at the same time, a denoising method based on energy variations is provided for hyperspectral image. With the hypersurface minimal scheme, energy functional is built of spatial surface area, spectral curve length and residual l2 norm. The minimal surface area term smoothes data in spatial dimensions and the minimal curve length term works in spectral dimension whose weights are determined by two positive coefficients. The surface mean curvature and curve curvature drive suppressing of noise together. Then perturbation is abated by available neighboring reliable data. The proposed method works well in experiments on hyperspectral image from Hamamatsu camera and AVIRIS.