Spectral feature enhancement for hyperspectral imagery

It is common practice in digital imaging to apply a spatial modulation transfer function compensation (MTFC) function as a convolution filter to accomplish image sharpening. MTFC in the spatial domain is applied to back out blurring introduced by the various MTF degraders in the image chain. Analogously, in hyperspectral imaging, there is generally a blurring in the spectral dimension due to overlapping spectral bands. This blurring effect can cause narrow-band absorption features to become less apparent when a material is imaged. In a recent study at Kodak, we showed that a hyperspectral signature can be 'sharpened' in the spectral dimension by developing a set of convolution kernels that effectively reduce the overlap among the spectral responsivity of the detectors (i.e., using an appropriate convolution kernel that effectively narrows the spectral responsivity of a detector). Our initial simulations have shown that the main limitation of this technique is its performance in noisy conditions.