ELIN Local Optimization Algorithm

The ELIN algorithm is the latest PANCROMA technology designed at overcoming the problem of spectral distortions introduced by the nature of the panchromatic bands that represent the high-resolution information in satellite data scenes. As explained elsewhere in this manual, the Hue, Saturation, Intensity (HSI) transformation method provides a convenient way to inject high resolution spatial information into the lower resolution spectral information in order to fuse the spatial and the spectral to produce a high resolution color image. The problem with the naive approach is that the process only works if the spectral characteristics of the panchromatic band closely match those of the discarded Intensity band. Unfortunately, this may not be true. Landsat provides the most common example of the problems with this approach.

The Landsat panchromatic band is sensitive to light in the 0.52m to 0.92m wavelength range. Visible light spans the 0.45m to 0.69m wavelength range. As a result, substituting the panchromatic band for the Intensity band before reverse-transforming back to red-green-blue (RGB) color space can cause noticeable and often severe spectral distortions in the resulting pan sharpened image, especially in areas of high Near Infrared (NIR) reflectivity like grasslands and other narrow-leaf vegetation for example. The methods described in the previous sections address this problem by using the NIR band 5 to adjust the panchromatic band before substituting. These methods can produce very acceptable results. However, none of them is perfect because while NIR adjustment can correct for panchromatic NIR sensitivity, it cannot provide the missing blue band (0.45m-0.515m) information.

PANCROMA takes a different approach with its new ELIN Local Optimization algorithm. This algorithm replaces the panchromatic band with a synthesized panchromatic (SPAN) band that has virtually zero spectral distortion. When the SPAN band is substituted for the Intensity band, the reverse transformation from HSI color space to RGB color space occurs with a similar lack of spectral distortion. The result is often a pan sharpened image that matches the low resolution RGB image color tones almost exactly.

The following set of images demonstrates the advantages of ELIN. The first image is a subset from a Landsat scene L71001086_08620051201_BXX.TIF RGB color composite image from band1, band2 and band 3.

Next is a pan sharpened RGB image using the basic HSI algorithm.

This image shows some sever color distortions in the vegetated areas (other than dense forest). The final image is one made using the same four bands as used to produce the image above.

The color tones of this image match those of the reference low resolution RGB color composite almost exactly. The vegetation tones in particular match very well.

Because creating the SPAN is very computation intensive, ELIN runs can require significantly more processing time than the transformation methods discussed in previous sections. However the excellent results, especially for Landsat may be well worth the extra processing time.