Image Processing Reference
In-Depth Information
The last technique discussed in the monograph is based on optimizing the fusion
weights on certain desired properties of the fused image. The well-exposedness and
local contrast are considered to be the desired properties. The fused images, therefore,
possess high values of variance and average gradient. However, this output-driven
approach leads to a non-uniform contribution of bands and a shift in the radiometric
mean of the data. Therefore, the values of the fusion symmetry FS and relative bias
b
are quite high for this technique.
The band selection techniques discard a large amount of information by selecting
only 3 bands from 200
spectral bands. The contents of the composite RGB image,
therefore, are dependent upon the contents of the chosen bands which in turn depend
upon the materials present in the scene. Hence, the performance of this technique
is quite unpredictable. The PLF and the CMF techniques assign a single weight to
all the pixels in a given band. These techniques, thus, do not exploit the contents of
individual bands. The resultant fused images possess good contrast, sharpness, and
visual quality if the bands that have been assigned higher fusion weights contain data
with higher quality. Otherwise, the fused images may lack these qualities. Fusion
solutions presented in the monograph exploit the entire hyperspectral data cube
by computing fusion weights for each pixel in every band independently. These
fusion solutions process the data on a per pixel basis which makes them better
for visualization of hyperspectral data. The gain in quality of the fusion results is,
however, obtained at a cost of higher computation.
+
